CN120781175B

CN120781175B - Charging pile data transmission method

Info

Publication number: CN120781175B
Application number: CN202511279631.6A
Authority: CN
Inventors: 黄瑞娟; 王金涛; 李瑶; 朱永康
Original assignee: Jiangyin Furen High Tech Co Ltd
Current assignee: Jiangyin Furen High Tech Co Ltd
Priority date: 2025-09-09
Filing date: 2025-09-09
Publication date: 2025-11-14
Anticipated expiration: 2045-09-09
Also published as: CN120781175A

Abstract

The invention discloses a transmission method of charging pile data, which relates to the technical field of charging pile data transmission and comprises the following steps of comparing an uploading attached tag sequence with real-time charging behaviors through an integrated state recognition algorithm formed by a graph neural network and a long-term and short-term memory network, recognizing a message context failure state caused by a user terminating charging operation in advance in a frame splicing uploading process, outputting a context validity vector, constructing a frame segment transmission influence map according to the context validity vector, calculating behavior attached strength and validity weight of each remaining non-uploading data segment through the influence map, and recognizing a data segment set needing to be interrupted from uploading by utilizing a preset weight threshold. The invention solves the difficult problem of interruption identification in the frame splicing uploading of the charging pile, realizes accurate interruption and splicing avoidance of invalid data segments, and ensures the semantic consistency and the system stability of messages.

Description

Charging pile data transmission method

Technical Field

The invention relates to the technical field of data transmission of charging piles, in particular to a data transmission method of charging piles.

Background

The transmission of the charging pile data refers to a data interaction process between the charging pile and an external association system (such as a background monitoring platform, a management server, a user terminal, an energy management system and the like) in the operation process of the charging pile equipment. The transmission involves the contents of charging pile operating state, control commands, metering data, fault information, communication protocol negotiation, device configuration parameters, remote upgrade information, etc., and communicates by wired (e.g., ethernet) or wireless (e.g., 4G/5G, wi-Fi, NB-IoT) means. In order to ensure stability, security and real-time performance of data transmission, communication processing technologies such as protocol design, data encapsulation and analysis, transmission encryption, buffering and retransmission, network fault tolerance and the like are generally involved. The process plays a core supporting role in ensuring reliable operation, remote control, data analysis, energy scheduling and the like of the charging pile.

The existing charging pile data transmission technology generally establishes network connection with an external system through a communication module embedded in the charging pile so as to realize remote interaction and real-time management of contents such as operation data, control instructions, charging information and the like. The specific implementation method comprises the steps of firstly, establishing connection between a charging pile and a background system through a built-in communication module (such as 4G/5G, ethernet, wi-Fi or NB-IoT), secondly, data transmission is generally based on a standard or customized communication protocol (such as OCPP, modbus, DL/T645 and the like) to prescribe a message format, an interaction flow and an exception handling mechanism, thirdly, data is subjected to structuring and packaging before transmission, and is assisted with means of data encryption, signature verification, compression and the like to improve safety and transmission efficiency, and at a data link layer, the system also introduces mechanisms such as breakpoint continuous transmission, cache retransmission, network health detection and the like to solve the problem of data packet loss or communication interruption in an unstable network environment. In addition, the whole transmission process also comprises a plurality of links such as collection and report of state data, issuing and response of remote control commands, pushing of remote upgrading data packages of equipment firmware, and cooperative data interfaces required by a butt joint power system or an energy scheduling system, which form a complete realization chain of the current charging pile data transmission technology, and provide a technical foundation for the intellectualization, networking and operational maintenance of the charging pile.

The prior art has the following defects:

In the process that the charging pile adopts a segmented frame splicing mechanism to carry out data transmission, when the communication quality is poor, the pile end usually splits a complete data message into a plurality of data segments to be sequentially uploaded, and each data segment needs to wait for background confirmation after being sent so as to realize that the data splicing is finally completed in the background. If the user terminates the charging operation in advance in the frame sharing uploading process, the pile end will immediately generate and send a termination message for notifying the background of the termination of the charging action, but the residual data segments still in the queue to be uploaded are not interrupted in time due to the lack of a mechanism for identifying the frame sharing revocation status in the existing transmission logic, and still continue to be sent to the background according to the original logic. When receiving the data segments, the background still executes conventional splicing processing because the validity of the data segments cannot be judged, so that the invalid residual data and the previous data segments are spliced to generate a false message which is complete in format and invalid in semanteme. The existing charging pile data transmission technology cannot interrupt continuous uploading and splicing processing of the residual data segments according to the message context failure state caused by the specific operation of stopping charging in advance by a user in the frame splicing uploading process, so that the error recognition of the equipment state by a background is caused, abnormal data archiving and charging analysis can be triggered, and errors in scheduling strategies and misjudgment of running states can be caused, and even the overall stability and compliance of the system are affected.

The above information disclosed in the background section is only for enhancement of understanding of the background of the disclosure and therefore it may include information that does not form the prior art that is already known to a person of ordinary skill in the art.

Disclosure of Invention

The invention aims to provide a transmission method of charging pile data, which aims to solve the problems in the background technology.

In order to achieve the above purpose, the invention provides a method for transmitting charging pile data, which specifically comprises the following steps:

s1, constructing a behavior context identification chain for each group of frame splicing data to be uploaded, generating a behavior perception matrix by the behavior context identification chain through a charging behavior event sequence, and generating an uploading attachment tag sequence according to the behavior perception matrix to establish a binding relation between a data message and charging behavior;

S2, comparing the uploading attached tag sequence with the real-time charging behavior through an integrated state recognition algorithm formed by the graph neural network and the long-term and short-term memory network, recognizing a message context failure state caused by the fact that a user terminates a charging operation in advance in the frame splicing uploading process, and outputting a context validity vector;

s3, constructing a frame segment transmission influence spectrum according to the context effectiveness vector, calculating behavior attaching strength and effectiveness weight of each remaining data segment which is not uploaded through the influence spectrum, and identifying a data segment set needing to be interrupted for uploading by utilizing a preset weight threshold;

S4, adding a logic isolation mark to the identified data segment set, transmitting the logic isolation mark as input to a frame processing path screening structure constructed based on an attention mechanism, and determining the data segment set which does not participate in uploading and splicing processing according to the priority of the data segment, a behavior attachment chain and a time synchronization field;

S5, carrying out real-time clustering on the rest data segments to be uploaded through a behavior clustering engine constructed by the multi-layer perception structure, generating a frame splicing consistency confidence factor, comparing the frame splicing consistency confidence factor with a strategy threshold interval, and executing dynamic regulation and control of a data uploading path and a splicing processing path based on the comparison result.

Preferably, S1 specifically comprises the following steps:

collecting charging behavior events in a charging process, constructing a charging behavior event sequence, wherein each charging behavior event comprises a time field, an operation type field and an execution source field, and taking the charging behavior event sequence as an input basis of a behavior context identification chain for defining a behavior context range attached by frame spelling data;

The method comprises the steps of taking a charging behavior event sequence as input, constructing a three-dimensional behavior perception matrix according to the sequence of events and the relevance between event types, wherein three dimensions of the behavior perception matrix respectively represent the behavior event sequence, the behavior event type and an event behavior driving level and are used for carrying out structural expression on charging behavior characteristics;

Analyzing the behavior perception matrix, extracting behavior event sections overlapped with each group of frame splicing data to be uploaded in time, and distributing an uploading attachment label for the frame splicing data, wherein the uploading attachment label comprises a corresponding behavior event sequence index, a behavior state type and a driving level identification and is used for representing the attachment relation between the frame splicing data and a specific charging behavior;

Binding the uploading attachment tag to a data message structure of corresponding frame splicing data, establishing one-to-one binding mapping between the data message and the charging behavior, embedding a behavior context identification chain into a frame splicing data uploading scheduling path, and realizing continuous synchronization relation maintenance of uploading data and behavior events.

Preferably, S2 specifically comprises the following steps:

s201, inputting an uploading attaching tag sequence into a graph neural network, constructing an uploading tag behavior graph, generating attaching connection edges among nodes according to graph structure relations between behavior event sequence indexes and behavior state types of the uploading attaching tag, extracting structure association features of the uploading attaching tag sequence in a behavior space, and forming graph neural embedded representation;

s202, constructing a behavior evolution sequence of a real-time charging behavior in a time sequence, inputting a long-period memory network, and extracting evolution feature vectors of the charging behavior in each stage through time gating state conduction for expressing continuity and termination trend of the charging behavior;

S203, based on an integrated state recognition algorithm, the graph nerve embedded representation is compared with the long-period memory feature vector, the context failure state of the message caused by the fact that a user terminates the charging operation in advance in the frame spelling uploading process is recognized, and the context validity vector composed of context validity marks corresponding to the uploading attachment labels is output.

Preferably, S201 is specifically:

Mapping each uploading attachment tag in the uploading attachment tag sequence into a node in the graph structure, wherein each node comprises a behavior event sequence index field and a behavior state type field and is used for defining a graph node set of an uploading tag behavior graph;

Based on the precedence relation of the behavior event sequence index and the semantic proximity of the behavior state type, constructing attachment connection edges between the nodes of the graph, wherein each attachment connection edge represents the structural dependency relationship between the nodes through an adjacency matrix;

inputting the constructed uploading tag behavior diagram into a graph neural network, extracting the structure embedded characteristics of graph nodes through graph convolution operation, and performing graph space structure coding on the uploading attached tag sequence to form graph neural embedded representation for subsequent behavior comparison.

Preferably, S203 is specifically:

An integrated state recognition algorithm is established, the integrated state recognition algorithm comprises a structure input channel and a behavior input channel, the structure input channel extracts a space attaching mode between uploading attaching labels based on graph nerve embedding representation, the behavior input channel captures an evolution track of a charging behavior based on long-short-term memory feature vectors, the graph nerve embedding representation is input into the structure input channel, the long-term memory feature vectors are input into the behavior input channel, and the structure input channel is used for forming a structure-behavior comparison standard;

Respectively establishing index consistency alignment relations in a structure input channel and a behavior input channel, performing node level comparison operation on an index dimension of an uploading attachment tag in an attention alignment mode, extracting a context consistency score sequence between a graph nerve embedded representation and a long-period memory characteristic vector, and forming a context difference map for failure recognition;

On the basis of the context difference map, a behavior interruption judging threshold range is set, uploading attachment labels with consistency scores lower than a threshold lower limit are marked as invalid states, uploading attachment labels with scores higher than a threshold upper limit are marked as valid states, other labels are judged according to a neighbor inference method, and finally a context validity vector composed of context validity marks corresponding to all uploading attachment labels is output.

Preferably, S3 specifically comprises the following steps:

S301, establishing a one-to-one mapping relation between the validity marks of all uploading attachment labels in the context validity vector and uploading attachment labels of the remaining non-uploading data segments, constructing a frame segment transmission influence map according to the mapping result, wherein each node in the frame segment transmission influence map represents one remaining non-uploading data segment, and establishing edge connection between the nodes based on the behavior event sequence index sequence in the uploading attachment labels, the time field continuity and the similarity calculation among behavior state types to form a graph structure reflecting the data uploading structure logic;

S302, calculating behavior attaching strength and effectiveness weight of each remaining non-uploaded data segment based on nodes and edge weight structures thereof in a frame segment transmission influence map, wherein the behavior attaching strength calculates path compactness according to sequence differences of behavior event sequence indexes in uploading attaching labels among the nodes, and the effectiveness weight is weighted and averaged according to effectiveness labels associated with uploading attaching labels in context effectiveness vectors and combining with node adjacent relations to obtain a context transmission evaluation factor of each node;

And S303, comparing the context transmission evaluation factor with a preset weight threshold, if the context transmission evaluation factor is lower than the lower limit of the termination uploading judgment threshold, marking the corresponding data segment as an object of interruption uploading, if the context transmission evaluation factor is higher than the upper limit of the maintenance uploading judgment threshold, continuing uploading the data segment, classifying and identifying other data segments by adopting a deviation degree judgment strategy based on local weighted average according to the centrality weight distribution of the local subgraph of the graph structure, and finally forming a data segment set needing interruption uploading.

Preferably, S302 is specifically:

Taking the index value of the behavior event sequence in the uploading attached label of each node in the frame segment transmission influence map as an index coordinate, calculating an index difference value between the current node and a directly adjacent node, setting an edge weight according to the reciprocal of the index difference value, and measuring the behavior compactness between uploading labels;

defining the behavior attaching strength of each node as a weighted average value of the edge weights of all adjacent nodes according to the edge weights among the nodes in the graph structure, and reflecting the attaching strength aggregation degree of the current data segment and other data segments in the uploading sequence;

Extracting a corresponding validity mark value of each node uploading attachment label in a context validity vector, carrying out weighted summation on the validity mark value and a validity mark value of an adjacent node according to the edge weight to obtain the validity weight of each node, and carrying out normalized weighted fusion on the behavior attachment strength and the validity weight to obtain a context transmission evaluation factor corresponding to the node.

Preferably, S4 is specifically:

For the identified data segment set, inserting a logic isolation mark according to the information identification field, wherein the logic isolation mark is composed of Boolean isolation bits and context isolation factors and is used for indicating the independent state and the behavior failure level of the data segment in the subsequent processing flow;

Taking the data segments embedded with the logic isolation marks as input and inputting a frame processing path screening structure, constructing the frame processing path screening structure based on an attention mechanism, taking each data segment as an independent query unit, and generating attention distribution vectors in the whole data segment sequence to be processed;

And according to the priority value of each data segment, the node depth in the behavior attachment chain and the alignment offset of the time synchronization field, carrying out multi-factor weighting scoring by combining the logic isolation marks, determining whether the data segments enter the splicing path or not through the attention weighting score, and marking the data segments with the score lower than a preset threshold value as not participating in uploading and splicing processing objects.

Preferably, S5 is specifically:

Constructing a multi-layer perception structure comprising an input embedding layer, a feature extraction layer and a semantic aggregation layer, inputting context feature vectors of the rest data segments to be uploaded into the structure, and extracting behavior mode representation of each data segment;

Based on the extracted behavior pattern representation, clustering the rest data segments to be uploaded in an embedded space by using a behavior clustering engine, calculating the consistency distribution density inside the clusters to which each data segment belongs, and generating a corresponding framing consistency confidence factor;

comparing the frame splicing consistency confidence factor with a preset strategy threshold interval, triggering the priority scheduling of the uploading path if the factor value is higher than the threshold upper limit, marking the uploading path as a splicing avoidance object if the factor value is lower than the threshold lower limit, and dynamically configuring the uploading path and the splicing processing path by the rest data segments according to the confidence factor and the time sequence characteristic.

In the technical scheme, the invention has the technical effects and advantages that:

1. According to the invention, by constructing the charging behavior context identification chain and uploading the attached tag sequence, accurate binding between the framing data and the charging behavior is realized, and when a user terminates the charging operation in advance, an integrated state identification algorithm constructed by combining the graph neural network and the long-short-term memory network can be combined, and the behavior interruption trend in the data uploading process is dynamically perceived, so that the failure state of the message context is identified. Based on the failure state identification result, the system further builds a frame segment transmission influence map, quantifies the behavior adherence intensity and the context validity weight of each remaining unoccupied data segment, accurately identifies and interrupts the uploading path of the data segment with failed semantics according to the context transmission evaluation factors and the preset weight threshold, prevents invalid data from entering the background splicing logic, and remarkably improves the semantic consistency and the framing accuracy of data transmission.

2. According to the invention, a logic isolation mechanism and a frame processing path screening structure are introduced, on the basis of identifying a data segment to be interrupted, the behavior failure level is marked by using Boolean isolation bits and context isolation factors, and the objects which do not have splicing effectiveness are dynamically removed by combining a priority value, a behavior dependent chain structure and a time synchronization field and executing multi-factor weighted screening based on an attention mechanism. Further, behavior pattern clustering is carried out on the data segments to be uploaded through the multi-layer sensing structure and the clustering engine, joint regulation and control of an uploading path and a splicing processing path are implemented based on a splicing frame consistency confidence factor and a strategy threshold interval, intelligent and self-adaptive adjustment of a data uploading strategy is achieved, false message generation and rear end recognition errors are effectively avoided, and system stability, accuracy and scheduling compliance are enhanced.

Drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments described in the present application, and other drawings may be obtained according to these drawings for those skilled in the art.

Fig. 1 is a flow chart of a method for transmitting charging pile data according to the present invention.

Detailed Description

Example embodiments will now be described more fully with reference to the accompanying drawings. However, the example embodiments may be embodied in many different forms and should not be construed as limited to the examples set forth herein, but rather, the example embodiments are provided so that this disclosure will be more thorough and complete, and will fully convey the concept of the example embodiments to those skilled in the art.

The invention provides a transmission method of charging pile data as shown in fig. 1, which specifically comprises the following steps:

in this embodiment, S1 specifically includes the following steps:

When collecting charging behavior events in the charging process and constructing a charging behavior event sequence, an event monitoring task can be set on the charging pile side to perform high-frequency sampling on a data communication instruction stream, a user operation record, a charging gun physical state change, a vehicle feedback signal and the like related in an operation flow, and an ordered event stream is generated by combining a time stamp. Each charging behavior event needs to comprise three key fields, namely a time field is used for recording a specific time node of the event and used as a time sequence reference of a behavior sequence, an operation type field is used for identifying a behavior category corresponding to the event, such as starting charging, suspending charging, user card swiping, vehicle disconnection and the like, so that subsequent event semantics are conveniently classified, and an execution source field is used for indicating that the event is triggered by a user action, autonomous logic triggering of a charging pile or vehicle signal triggering, so that a behavior source is defined when a behavior dependency chain is constructed. The charging behavior event sequence is formed by combining the three types of fields into a structured data unit and arranging the structured data unit in time sequence. The behavior event sequence is an input basis of a behavior context identification chain and is used for definitely determining the behavior phase context of the frame splicing data in the process of generating and uploading, so that the accurate binding and tracing of the follow-up uploading data and behavior states can be realized. In the implementation process, the original behavior event stream can be stored through the intermediate buffer queue, then the event analysis engine analyzes and constructs the charging behavior event sequence with definite semantics and ordered time sequence according to the fields.

When a three-dimensional behavior perception matrix is constructed by taking a charging behavior event sequence as input, each behavior event in the event sequence is firstly required to be uniformly encoded to form a structural data format with processibility. The three dimensions of the three-dimensional behavior perception matrix are behavior event sequence, behavior event type and event behavior driving level respectively. The behavior event sequence dimension is sequenced through time fields of the events to ensure that the matrix keeps continuity in the time dimension, the behavior event type dimension is subjected to category mapping through operation type fields, for example, operations such as 'start charge', 'stop charge', 'card swiping authentication', 'vehicle leaving pile', and the like are mapped into discrete category indexes for expressing different behavior modes, the event behavior driving level dimension is deduced together through an execution source field and a behavior triggering intensity parameter, and a numerical value reflecting the behavior triggering intensity is endowed to each event to express the dominance degree or response priority of the event to the whole charging behavior path. The build process may insert each charging behavior event sequentially into the tensor index location by initializing the three-dimensional tensor structure, and fill in the classification identifier in the corresponding event type dimension, and fill in the driving weight calculated by the behavior source evaluation engine in the driving level dimension. The finally generated three-dimensional behavior perception matrix can accurately express the time sequence relation, type composition and influence weight of the behavior event in the charging process, realizes structural coding and visual modeling of the charging behavior sequence, and provides high-resolution behavior semantic support for subsequent behavior state identification and uploading dependency construction.

When analyzing the behavior perception matrix to extract the behavior event section overlapped with each group of frame data to be uploaded in time, firstly, calibrating a generation time window of the frame data, and determining a start-stop time stamp corresponding to each group of frame data. And then, identifying all behavior events overlapped with the time range of the frame spelling data by comparing the time stamp interval with the time index marked by the sequence dimension of the behavior events in the behavior perception matrix. These overlapping events form the behavior event section to which the frame data belongs, and are the basis for the generation of subsequent tags. For each spelling data segment, the most representative behavior event characteristics in the behavior event segment are extracted to construct an uploading attachment tag. The tag comprises three fields, namely a behavior event sequence index, a behavior state type, a driving level identification and a driving level identification, wherein the behavior event sequence index is used for identifying the positioning of the data segment in a behavior event sequence, the behavior state type is determined according to an operation type field of the extracted behavior event and used for expressing behavior semantics of the data segment, the driving level identification is used for extracting a numerical value based on a driving weight dimension in a behavior perception matrix, and the dominant strength of behavior triggering is reflected after normalization processing. After the tag is generated, the tag is bound to a transmission header field corresponding to the frame splicing data segment, so that semantic attachment mapping of the frame splicing data and the charging behavior to which the frame splicing data belongs is realized. The process can realize high-efficiency event matching by constructing an index mapper, automatically complete label structure filling by a rule engine, ensure the integrity and accuracy of uploading attached labels, and provide semantic support for subsequent context identification and uploading decisions.

When the uploading attachment tag is bound to the data message structure of the corresponding frame splicing data, the message structure of the frame splicing data is expanded first, and a special field is reserved for storing the uploading attachment tag. The tag field is embedded in a control header area of the data message in a structured form and comprises three contents, namely a behavior event sequence index, a behavior state type and a driving level identification. The behavior event sequence index is used for establishing an accurate index mapping between the data segment and the behavior nodes in the event sequence in the uploading process, the behavior state type identifies the charging process stage of the data segment, and the driving level identification is used for quantifying the behavior dependency degree of the uploading of the data segment. After the label is embedded, the uploading strategy logic related to the charging behavior can be dynamically invoked according to the label content by maintaining a behavior context identification chain in an uploading scheduling path, so that one-to-one binding between the frame splicing data uploading behavior and the charging behavior event is realized. In order to realize continuous synchronous relation maintenance, a behavior scheduling matching engine can be constructed, and priority ordering and transmission control are realized in an uploading queue according to label content, so that the frame splicing data is always kept to be synchronously propelled with real-time charging behaviors in the uploading process. The structured binding mode not only realizes the strong association of data and behaviors, but also provides basic semantic support and traceability for the follow-up identification of behavior context failure and the judgment of the execution uploading interruption.

in this embodiment, S2 specifically includes the following steps:

The real-time charging behavior builds a behavior evolution sequence in time sequence, firstly, event timestamp ordering is needed to be carried out on collected original charging behavior data, all behavior events are ensured to be arranged according to the actual occurring sequence, each behavior event consists of an event type, a behavior state, a time field and an event trigger source field, and the behavior event is used for describing state transition in the current charging process. After the sorting is completed, the behavior sequence is subjected to sectional coding through a sliding window, continuous behavior fragments are mapped into a time fragment vector sequence, the length and the characteristic dimension are unified, and preparation is made for inputting a long-short-term memory network. The behavior evolution sequence is taken as input and is transmitted into a long-term and short-term memory network, and in the network, an input vector sequentially conducts state conduction through a time gating structure constructed by an input gate, a forgetting gate and an output gate, the hidden state of each time node captures the characteristic expression of a corresponding behavior segment, and long-term dependence and short-term fluctuation are fused and encoded to generate a characteristic vector sequence crossing time evolution. The sequence reflects the continuous change and the potential termination trend of the charging behavior in different stages, and further provides a semantic expression basis for the interruption of the subsequent recognition behavior. The long-term and short-term memory network avoids gradient dissipation through a gating mechanism, and can accurately memorize the influence of early behavior events on the current state, so that the evolution feature vector has context relevance and behavior dynamic sensitivity.

The method is used for accurately identifying the message context failure state caused by the fact that a user terminates the charging operation in advance in the frame splicing uploading process, so that the key problem that the data validity cannot be dynamically judged based on the charging behavior context in the prior art is solved. The method comprises the steps of inputting an uploading attaching tag sequence into a graph neural network, constructing an uploading tag behavior graph, extracting structural association features, constructing a real-time charging behavior into a behavior evolution sequence, inputting a long-term memory network, describing the evolution trend and a termination signal of the charging behavior in a time dimension, enabling a system to have time perception capability, and carrying out structure-behavior double-channel comparison on the graph neural embedded representation and the long-term memory feature vector through an integrated state recognition algorithm on the basis of the time perception capability, wherein the method is used for capturing the context failure state caused by the advanced termination operation of a user by combining the attaching structure between uploading data and user operation and fusing the continuity features of the charging behavior. The whole flow realizes the data semantic validity identification under the dynamic behavior background through the fusion of the structural modeling and the time sequence modeling, and has high intelligence and pertinence.

In this embodiment, S201 specifically includes:

In the process of constructing the uploading tag behavior graph, the mapping operation can be completed by establishing a graph node corresponding to each uploading tag in the uploading tag sequence. The mapping process is based on the behavior event sequence index field and behavior state type field contained in the label, each graph node uses the unique behavior event sequence index as its position identification in the graph structure, and the behavior state type field is used for defining the role of the node in behavior semantics, such as the operation states of "start charging", "stop request", "power adjustment", etc. In this way, the uploading attachment tag sequence is converted into a structured graph node set, each node not only carries binding information between the data segment and the charging behavior, but also forms a semantic link between behavior events in the graph structure, and provides a semantic and structural foundation for the follow-up construction of the attachment connection edge in the uploading tag behavior graph. The definition mode of the graph node set ensures the simultaneous expression of time sequence and semantics among labels, so that the graph neural network can fully learn the structural characteristics and the logic attachment relation of the uploaded data under the charging behavior background.

When the attachment connection edges between the graph nodes are constructed, the connection sequence of the graph nodes in the time dimension can be determined according to the time sequence relation of the behavior event sequence indexes in the uploading attachment labels, and meanwhile, the logic connection between the nodes in the semantic dimension is established by combining the semantic proximity between behavior state types. The time sequence relation is judged through the increasing trend of index values of the behavior event sequence, if indexes of the two nodes are continuous or close, the strong time dependence exists, the semantic proximity is quantified through a predefined behavior state type similarity matrix, and the higher the functional similarity degree between the types is, the larger the semantic dependence value is. When the time dependence and the semantic proximity meet the set threshold condition, an attachment connection edge can be established between two nodes, and the existence of the edge is recorded in a form of an adjacent matrix. In the adjacency matrix, each row corresponds to each column to form a graph node, non-zero values in the matrix indicate that the nodes have an dependency relationship, and the numerical value represents the dependency strength. The attachment connection edge constructed in the method can accurately reflect the structural dependence and semantic continuity between nodes in the uploaded tag behavior graph, and provides a clear connection basis for the structure learning process of the graph neural network.

After the uploading tag behavior diagram is input into the graph neural network, the graph nodes can be subjected to feature aggregation and updating by adopting graph convolution operation. Each graph node represents an uploading attachment tag, the initial characteristics of the graph node consist of a behavior event sequence index field and a behavior state type field, the graph neural network collects characteristic information from adjacent nodes in each round of convolution through a structural relation defined by an adjacent matrix, and the graph neural network combines the characteristics of the graph node to carry out weighted fusion so as to extract high-order characteristic expression reflecting the semantic status of the node in the whole graph structure. By stacking the multi-layer graph rolling network, the attachment patterns between remote nodes can be recursively captured and aggregated in a global structure. After the graph convolution is completed, the embedded features generated by all graph nodes can be arranged according to the index sequence of the uploading attachment tag sequence, so that a group of structure coding vector sequences with space attachment semantics are constructed, namely, graph space structure coding of the uploading attachment tag sequence is completed. The finally formed graph nerve embedded representation not only maintains the information characteristics of the labels, but also fuses the structural relationship and the context dependence among the behavior labels, and provides a high-distinction representation basis for the subsequent behavior comparison.

In this embodiment, S203 specifically is:

When the integrated state recognition algorithm is constructed, a two-channel neural network architecture can be adopted, wherein a structure input channel adopts a graph neural network as a core structure and is used for processing graph neural embedded representation of an uploading attachment label, extracting a spatial attachment mode of the graph neural embedded representation in a frame spelling data uploading process, and a behavior input channel adopts a long-period memory network, receives a charging behavior evolution vector constructed according to a time sequence and extracts time dynamic characteristics of the charging behavior evolution vector in a charging behavior process. In order to realize the collaborative comparison between two channels, the graph nerve embedded representation and the long-short-period memory feature vector are respectively subjected to normalization, dimension alignment and embedded stretching treatment in the respective channels, are uniformly mapped to a shared representation space, and are subjected to vector cascading and cross attention fusion in the uniform space to form a structure-behavior comparison reference capable of measuring the context consistency. The reference can capture the graph structure relation between the uploading attached labels, can be related to the state evolution trend of the charging behavior of the user, and is used for supporting the subsequent accurate identification of the context failure state of the message. The integrated state recognition algorithm enables the context effectiveness analysis to have high expression capacity and discrimination through joint modeling of structure dependence and behavior continuity.

In order to realize the context consistency analysis between the graph nerve embedded representation and the long-term and short-term memory feature vector, first, index consistency alignment relations are respectively established in the structure input channel and the behavior input channel. Each tag in the uploading attached tag sequence has a corresponding timestamp and event index in the behavior event sequence, and the indexes can be used as a unified alignment standard. By respectively carrying out index normalization, position coding mapping and time synchronization coding processing on input data in two channels, the image nerve embedded representation and the long-period memory feature vector have consistent reference coordinates in the index dimension of the uploading attachment tag. After the consistency alignment relation is constructed, the feature representation in the two input channels can be ensured to realize one-to-one comparison of node levels based on the shared index, so that a unified index foundation is laid for the subsequent comparison operation.

On the basis of index consistency alignment, a focus alignment mode is introduced, and node level comparison operation is performed on the index dimension of the uploading attachment tag. By constructing a cross-channel attention fusion structure, the graph nerve embedded representation is used as a query vector, the long-term and short-term memory feature vector is used as a key value vector, and an adaptive attention weight mechanism is adopted to calculate the attention weight value under each pair of corresponding indexes, so that the context consistency score between the nodes is obtained. These scoring sequences may be organized as one-dimensional vector sequences, with each position representing the degree of matching between the structural and behavioral representations of a particular uploading attachment tag. And further constructing a context difference map through a sliding window smoothing, local extremum detection and behavior change sensitive weighting mechanism. The map can be used for accurately identifying the area with inconsistent structural behaviors, thereby being used as a direct basis for identifying the failure of the context of the message in the frame assembly uploading. The context difference map has the sensitive response capability to the dynamic charging behavior mutation while keeping the behavior index consistency.

In order to convert inconsistent areas between structures and behaviors reflected in a context difference map into operable uploading data screening basis, a behavior interruption judgment threshold range needs to be set, and a context consistency score sequence is divided into intervals. The threshold range is composed of an upper bound and a lower bound, and can be calculated by using a clustering center of historical behavior interrupt sample data and a boundary statistical value. A score below the lower bound typically indicates a serious mismatch in structural behavior, reflecting that the framing data to which the uploaded tag is attached is no longer relevant to the current behavior context, and thus the type of tag is marked directly as a dead state. A score above the upper bound indicates that the behavior continuity is clear and the attachment strength is stable, and the tag can be determined to be in a valid state. And (3) since the label with the score in the threshold interval is in the fuzzy critical state due to the consistency of the context, in order to avoid misjudgment, a neighbor inference method is introduced to carry out fine judgment, namely, the current label state is inferred by constructing a label index adjacency graph and referring to the duty ratio trend of adjacent validity marks in the context difference graph. The marking operation forms a context validity marking set covering all uploading attachment labels to form a context validity vector for filtering judgment of subsequent uploading data segments.

In the actual execution process, the context consistency score sequence can be standardized firstly, mapped into a uniform interval scale, and then the local mean and variance are calculated based on sliding window convolution to assist in judging the score fluctuation threshold. The behavior interruption judgment threshold can be set according to the joint generation of a plurality of dimensions, such as the comprehensive modeling of event index span, behavior state change rate and historical failure rate model, and the optimal upper and lower limit boundaries are determined. When the context validity vector is output, a mapping structure corresponding to the uploading attached tag index one by one is needed to be constructed, wherein the tag value corresponding to each tag is determined through regular state mapping, the valid state is assigned to be 1, and the invalid state is assigned to be 0. Aiming at labels in a threshold interval, the neighbor inference method is specifically implemented by taking a target label as a center in a label index diagram, expanding a certain depth to left and right adjacent nodes, counting the valid state distribution of the adjacent labels, and determining the most probable state of the current label in a majority vote or weight scoring mode. The process gives consideration to the local consistency characteristic of the label context, and effectively avoids state identification errors caused by abnormal disturbance.

in this embodiment, S3 specifically includes the following steps:

The construction of the frame segment transmission influence map depends on the binding relationship between the validity flag of each uploading attachment tag and the remaining non-uploading data segment in the context validity vector. Each upload tag contains semantic features of the charging behavior, including a behavior event sequence index, behavior state type, and time field. By comparing these labels with the labels bound to the remaining data segments, a one-to-one mapping can be achieved. On the basis, each node in the graph structure is used for representing a remaining data segment which is not uploaded, and connecting edges between the nodes are calculated based on three core association features, namely, a sequence relation of behavior event sequence indexes and causal sequence logic reflecting behaviors, continuity of time fields and semantic similarity between behavior state types, wherein the continuity of the time fields is used for measuring time sequence tightness of the data segments in a charging process, and the semantic similarity between the behavior state types is used for judging whether functional dependence or connection exists between the two behavior states. The side weight between each pair of data segments is calculated through the three-dimensional index, and then a complete frame segment transmission influence map is constructed, and the map structure can accurately reflect the context association relation and the behavior attachment structure between the data segments in the data transmission regulation and control process, so that the subsequent context transmission assessment and uploading decision judgment are facilitated.

In order to construct a frame segment transmission influence map, firstly, mapping and confirming the validity mark of each uploading attachment label in the context validity vector and labels bound by the remaining non-uploading data segments in an index hash mapping mode to ensure that nodes are matched with the labels one by one. Then, each data segment is taken as one node in the graph, and a graph node set is established. According to each pair of nodes, edge connection in the graph is built based on three characteristics, namely, firstly, absolute difference of index values of behavior event sequences in two nodes uploading attached labels is calculated to be used as a time sequence distance, secondly, time field start-stop values of two data segments are extracted to judge whether intersection or close proximity exists on a time window, if so, time continuity scoring is distributed, and finally, semantic connection scoring is given to label pairs with similarity higher than a threshold value through embedding dictionary query behavior state types. After three scores are integrated, a linear weighting strategy is used for synthesizing the three scores into an edge weight, and weighted edge connection is established in the graph structure. The finally generated graph structure is a frame segment transmission influence map, not only the context behavior information of the uploading data segment is reserved, but also the causal and time sequence dependency relationship among the data segments is established, and a high-dimensional structure foundation for the subsequent context effective propagation, node evaluation and data uploading interruption judgment is supported.

The context transmission evaluation factor is a composite quantization index fused with behavior attachment strength and validity weight, and is used for reflecting the comprehensive state of each remaining unoccupied data segment in terms of behavior continuity and context semantic validity. And comparing the factor with a preset weight threshold value, so that dynamic classification decision of the uploading strategy can be realized. If the evaluation factor is higher than the upper limit of the upload termination judgment threshold, the context consistency and the behavior dependency relationship are stable, the method has high reliability uploading value, and the method should continue to participate in uploading. And for the middle region data segment between the two thresholds, the importance degree and the stability degree of the data segment in the adjacent behavior chain are evaluated by analyzing the local sub-graph structure in the transmission influence map of the frame segment where the data segment is positioned and combining the centrality weight distribution of the nodes in the graph. The centrality weight can be calculated by the edge weight intensity of the node and the adjacent node, and is used for representing the influence degree of the node on information conduction in the graph structure. And based on a deviation degree judging strategy of the local weighted average, further classifying the data segment by calculating the mean deviation degree of the nodes and the adjacent group nodes on the context transmission evaluation factors. The strategy can effectively avoid fuzzy judgment of boundary areas, improve robustness of frame splicing interruption identification, ensure that only data segments with effective behavior support are uploaded, optimize data quality and avoid semantic pollution.

In this embodiment, S302 is specifically:

In the frame segment transmission influence map, each node corresponds to a remaining non-uploaded data segment, and each node is bound with an uploading attachment tag, and the tag comprises a behavior event sequence index value used for indicating the time sequence position of the data segment in the whole charging behavior event sequence. In order to describe the behavior sequence relationship and context compactness between the data segments, the index value of the behavior event sequence in the node uploading attachment label is required to be used as a basic coordinate, the difference of the index values between any two directly adjacent nodes is calculated, and the smaller the difference is, the closer the two data segments are in time in the charging behavior flow, and the more continuous the behavior semantics is. To reflect this degree of closeness, the inverse of the index difference is used as the weighting value of the edge, i.e. the larger the edge weight, the stronger the behavior dependence between the two data segments. The method for calculating the edge weight based on the index reciprocal can logically map the behavior time into the graph structure, so that the weight of the edge in the graph has time sequence semantics and context expression capability, and a structural basis is provided for the follow-up behavior attachment strength and effectiveness weight calculation.

Taking an uploading task containing 5 remaining data segments not uploaded as an example, assume that the index values of the action event sequences in the corresponding uploading attachment labels are respectively [10, 12, 13, 17 and 19], wherein each data segment is a node in the graph structure. First, a connection relationship between node pairs is established, and the connection is defined as directly adjacent index pairs, for example, nodes 10 and 12,12 and 13,13 and 17,17 and 19 form directly connected edges. For the nodes 10 and 12, the index difference is |12-10|=2, the edge weight is 1/2=0.5, the difference is 4, and the edge weight is 1/4=0.25. And so on, the weights of all the connecting edges are calculated according to the corresponding inverse index difference values. The setting of the side weight not only reflects the uploading sequence among the data segments, but also quantitatively characterizes the bonding degree of the data segments on the action time sequence through the numerical weight. The finally obtained frame segment transmission influence map forms a graph model with behavior semantic structure weights, and provides an accurate behavior dependent structure foundation for subsequent aggregation calculation based on graph structures. The method has good interpretability and expandability, and can be directly applied to complex scene modeling under multi-source behavior data.

In the frame segment transmission influence map, in order to measure the tightness of a certain residual unoccupied data segment and other data segments in the behavior adherence chain, a value representing the behavior adherence strength of the node is calculated according to the side weights of all sides directly connected with the node in the map structure. Specifically, the edge weights between the node and all adjacent nodes are taken out, and a weighted average operation is performed to obtain the tightness mean value of the data segments of the node attached to other behaviors in the graph. The higher the value of the behavior dependency strength, the stronger the association of that data segment with other data segments in the upload behavior sequence, the more apparent the context continuity, generally meaning that its transmission behavior is more likely to be semantically part of a continuous upload operation. The calculation mode takes the side weight as the measurement basis of the behavior relevance, so that the behavior attachment strength has the structural semantic expression, and the subsequent judgment of the transmission necessity of the data segment can be assisted.

Taking a frame segment transmission influence map as an example, assume that an uploading data segment represented by a certain node a is respectively connected with three adjacent nodes B, C, D, and the side weights are respectively 0.5, 0.33 and 0.25. These edge weights are calculated from the reciprocal of the index difference of the sequence of preamble behavior events, and contain the behavior precedence logic. To calculate the behavior adherence strength of the node a, the edge weights of the three edges are added and averaged, adherence strength= (0.5+0.33+0.25)/(3) +0.36. The value characterizes the average behavior adhesiveness between the node A and the directly related data segment, if the value is obviously higher than the average attaching strength of other nodes in the graph, the node A can be inferred to play a stronger behavior accepting role in the data uploading process and should be preferentially reserved, otherwise, if the value is too low, the node A possibly belongs to an isolated or behavior interrupt node and needs to enter an interrupt uploading judging flow. The structure-driven measurement mode can realize automatic modeling of the context relation strength in an unsupervised scene, and the intelligent decision making capability of frame data scheduling is remarkably improved.

The nodes in each frame segment transmission impact map represent one remaining data segment that is not uploaded, and the uploading attachment labels carried by the nodes have unique corresponding validity flag values in the context validity vector (usually a numerical binary representation, for example, "1" represents valid and "0" represents invalid). In order to comprehensively evaluate the persistence of a certain data segment in the context semantics, the validity marks of the node and the validity marks of all adjacent nodes are fused and calculated, wherein the contribution of the adjacent nodes is determined by the side weight between the adjacent nodes and the central node, namely, the stronger side weight indicates that two nodes are closer in the behavior semantics, and the validity inference effect on the central node is larger. This fusion is accomplished by means of weighted summation, forming the validity weight of each node. And then, carrying out normalization processing on the behavior attaching strength of the node and the effectiveness weight, and then carrying out weighted superposition so as to construct a context transmission evaluation factor. The evaluation factor covers two dimensions of structure attachment and semantic effectiveness at the same time, and provides quantitative support for subsequent interrupt judgment.

Taking node X as an example, its corresponding validity flag in the context validity vector is 0.8, and the validity flag values of three nodes Y, Z, W adjacent to it are 0.9, 0.7, and 0.4, respectively, and the edge weights between them and node X are 0.5, 0.3, and 0.2, respectively. First, a weighted sum is performed according to the validity flag of the adjacent node and the edge weight, and an adjacent contribution term is calculated:

effectiveness weight = =. The following 0.9X0.5+0.7X0.3 +0.4×0.2) =0.45+ 0.21+0.08=0.74.

Combining the node X's own validity flag with the adjacency weighting result may be integrated, for example, in an average manner, to obtain a final validity weight value (e.g., (0.8+0.74)/2=0.77). Assuming that the behavior attachment strength of the node X is 0.6, normalizing the effective weight and the behavior attachment strength (for example, linearly normalizing to the interval of [0,1 ]), and then weighting and superposing according to a set weight proportion (for example, 0.5:0.5), so as to finally obtain the context transmission evaluation factor of the node X as (0.77+0.6)/2=0.685. This factor can be used to determine if the data segment corresponding to node X has value to continue uploading, with lower values indicating weaker context consistency and behavior continuity, more likely to be candidates for interrupting uploading. The calculation mode has good structural adaptability and interpretation capability, and is beneficial to accurately regulating and controlling the data frame splicing uploading strategy.

In this embodiment, S4 specifically is:

In the transmission control process of the charging pile frame data, the insertion of the logic isolation mark for the identified data segment set can be realized by embedding a special meta field in the structured message format of the data segment. The logical isolation mark is composed of two parts, namely a Boolean isolation bit and a context isolation factor. The boolean isolate bit characterizes in a binary manner whether the current data segment has been identified as a failing object, i.e. should be isolated from the subsequent upload and splice flows. The context isolation factor quantitatively represents the isolation level of the data segment under the failure behavior background by combining the information such as the failure marking strength in the context validity vector, the node centrality of the data segment in the frame segment transmission influence map, the coupling degree with the failure behavior event and the like. The insertion mode of the isolation mark can be realized by field expansion or defining independent 'ContextFlag' and 'DETACHLEVEL' fields at the head of the data segment, and the isolation field is identified and analyzed in a communication protocol, so that a subsequent processing module can accurately identify the isolated data segment.

In a specific implementation, assuming that the context validity flag corresponding to the uploading attachment tag of a certain non-uploaded data segment is 0 (invalid), and a plurality of nodes in adjacent nodes are also in a invalid state, the data segment is identified as a high isolation priority at the structural level. When constructing a message, the boolean isolation bit ContextFlag is set to 1, and the context isolation factor DETACHLEVEL is calculated according to the sum of the edge weights of the node and other failed nodes in the frame segment transmission influence map. For example, if the edge weights of the node a and the three failure nodes are 0.6, 0.7 and 0.9 respectively, then DETACHLEVEL is obtained to be 0.74 through normalization summation, which indicates that the behavior isolation degree of the data segment in the structure diagram is higher. The isolation information is added into the data segment as a part of the message structure, and is used as a key characteristic field when the subsequent incoming frame processes the path screening structure, and is used for assisting behavior filtering and screening judgment of the splicing path. The design ensures that the isolation expression of the behavior semantic state has clear and quantifiable description while the message is in a complete format.

in the process of frame splicing data processing path screening, a data segment embedded with a logic isolation mark is used as an input frame processing path screening structure, and the key is to conduct context-aware selective modeling on the data segment to be processed by using an attention mechanism. The frame processing path screening structure enables the system to establish a dynamic dependency weight relationship between all data segments to be processed by constructing a multi-head attention network with each data segment as an independent query unit. Each data segment includes its structural features (e.g., data segment priority, time synchronization field, behavior dependency chain) and context Wen Yuyi (e.g., boolean isolation bits and context isolation factors). These features are uniformly encoded as vector inputs, respectively as Query, key (Key) and Value (Value) input sources in the attention mechanism. When each data segment is processed, the model calculates the attention weight according to the similarity between the data segment vector and all other data segment vectors, and then forms an attention distribution vector which represents the processing relevance of the data segment in the whole context. The logical isolation markers will have a significant modulating effect on the formation of the attention profile, reducing the attention of the failed data segment to other segments when generating the attention vector, thereby causing the model to automatically weaken its selection weights in the upload path.

In practical implementations, a frame processing path screening structure may be constructed based on a transducer encoder framework, all data segments to be uploaded are encoded into d-dimensional embedded vectors, and the ContextFlag and DETACHLEVEL fields in each vector are mapped to one isolated modulation factor matrix. Assuming that the input sequence contains 5 data segments, a Query-Key scoring operation is performed on each data segment and the Value sums are weighted to generate a representation of its attention in the sequence. If ContextFlag of a data segment is 1 (logical isolation), then the generated Query will be multiplied by a suppression coefficient λ smaller than 1, typically 0.2, to suppress its interference with other data segments, and at the same time, its contribution when referenced as Key by other data segments is also reduced. The mechanism ensures that the logic isolation section is in an edge or weak connection state in the attention map, and is finally marginalized or directly screened out in the behavior uploading strategy, so that the logic isolation section is prevented from entering a subsequent splicing processing flow. The whole path screening structure keeps structural continuity between the embedded layer and the attention layer, ensures that the screening decision process is dynamically fused based on context semantics, structural coupling and priority, and improves the accuracy and robustness of a processing path.

In the path screening of frame splicing data processing, a multi-factor scoring system is built for each data segment to be processed by fusing the priority value, the node depth in the behavior dependent chain and the alignment offset of the time synchronization field, so as to realize fine granularity quantitative judgment of the uploading and splicing value of the data segment. The priority value reflects the importance level of the data segment in the whole data scheduling, and is preset generally according to the real-time performance, the data type or the uploading strategy, the node depth in the behavior dependence chain represents the behavior dependence level of the data segment in the user charging behavior, the greater the depth is, the stronger the influence of uploading on the complete behavior chain reconstruction is indicated, the alignment offset of the time synchronization field is used for describing the synchronization error of the data segment between the time dimension and the main behavior time sequence axis, the smaller the offset is, the higher the time consistency of the data segment is, and the stronger the context suitability is provided. These three types of structural indicators are encoded together with logical isolation markers into scoring vectors, one for each factor, and the composite score is calculated by weighted summation. The attention mechanism is used for dynamically adjusting the weight of each dimension so as to adapt to the behavior association strength change under different scenes, and the adaptive modeling of the context semantics and the structural influence of the data segment is realized. And finally, comparing the grading result with a preset threshold value, and if the grading result is lower than the threshold value, judging the data segment as a behavior failure and low-value object, and not entering an uploading and splicing path any more, so that behavior pollution and resource waste caused by invalid data splicing are effectively avoided. The mechanism can improve the context consistency, the resource utilization efficiency and the behavior data integrity of the framing processing strategy, and is particularly suitable for the scene of frequent interruption of the charging behavior of the user.

In this embodiment, S5 specifically is:

A multi-layer perception structure comprising an input embedding layer, a feature extraction layer and a semantic aggregation layer is constructed, and the purpose of deep expression modeling is achieved on context feature vectors of the rest data segments to be uploaded. The input embedding layer performs dense vectorization coding on discrete fields in the context feature vector such as behavior state types, uploading attached tag indexes and the like through an embedding matrix, and inputs the model in a unified vector space form. The feature extraction layer adopts a Convolutional Neural Network (CNN) or a Multi-head Attention mechanism (Multi-head Attention) for capturing local behavior variation and context coupling features between data segments in a time dimension. The semantic aggregation layer extracts semantic feature distribution of the global behavior sequence through pooling operation or a gating circulating unit (GRU) to form high-dimensional behavior pattern representation, and the model is ensured to be capable of carrying out joint expression on behavior states, context association and time sequence features of the data segments.

In particular embodiments, the context feature vectors of the remaining data segments to be uploaded may be organized in a triplet structure, each item including a timestamp, a behavior tag, and an attachment state value. Firstly, inputting the triples into an embedding matrix through a table look-up method to obtain dense representation with unified dimensions, for example, embedding each field into 64-dimensional vectors, splicing the vectors to form a 192-dimensional input vector, then inputting the vectors into a multi-head attention block with residual connection to capture the dependency relationship and behavior change trend among contexts, and finally, carrying out sequence modeling through a layer of bidirectional GRU network to output the aggregate semantic vector of each data segment as behavior mode representation. The method improves the semantic hierarchy of the feature expression on the premise of not losing time sensitivity, and provides an accurate behavior basis for the consistency judgment of the subsequent spelling frames.

The behavior clustering engine performs clustering operation on the rest data segments to be uploaded in a high-dimensional embedded space based on the extracted behavior pattern representation, and aims to identify semantic similarity among the data segments and measure aggregation compactness of the data segments. In the clustering process, each behavior pattern representation is regarded as a data point, and an unsupervised clustering algorithm such as K-means, DBSCAN or spectral clustering is adopted to divide the embedded vector space, so that data segments with similar semantic features are aggregated into the same cluster. After the clustering is completed, the distance from each data segment in each cluster to the mass center of the cluster is calculated, and the degree of agreement between the data segment and the current clustering behavior semantic is reflected. And constructing a consistency index based on the density distribution inside the clusters, converting the compactness of each data segment in the clusters into a probability confidence value, and taking the probability confidence value as a frame splicing consistency confidence factor of the data segment to measure the behavior consistency and the context matching degree of the data segment in a splicing structure.

In specific implementation, the behavior pattern representation of each data segment is mapped to a low-dimensional space through a Principal Component Analysis (PCA) or t-SNE mode and other dimension-reducing modes, so that clustering efficiency and interpretability are ensured. And presetting a plurality of clustering centers by using a K-means algorithm, distributing all data segments to corresponding clusters according to a minimum Euclidean distance principle, and recording the data density distribution in each cluster. For each data segment, calculating the normalized distance from the data segment to the cluster centroid as a consistency score, and further mapping the score into a frame splicing consistency confidence factor of an interval [0,1] by using a Softmax function, wherein the higher the numerical value is, the more structural continuity and splicing suitability of the data segment are shown in the current context. The method effectively establishes a quantization bridge between the behavior expression and the data splicing strategy.

The frame consistency confidence factor reflects the structural consistency and the behavior adaptation degree of each data segment to be uploaded in the clustering context. The system compares the confidence factor with a preset strategy threshold value interval in a numerical value mode to form a multi-level uploading regulation and control basis. When the confidence factor of a certain data segment is higher than the upper threshold, the confidence factor is high in semantic consistency with other data segments in the cluster, the data segment is suitable for preferential splicing and uploading, an uploading path preferential scheduling mechanism is triggered at the moment, and the priority of the segment in a scheduling queue is improved. Otherwise, if the confidence factor is lower than the threshold lower limit, the system marks the confidence factor as a splicing avoidance object, and the confidence factor is regarded as being deviated from the current semantic structure or isolated in behavior and is excluded from the uploading splicing flow. For the data segment with the confidence factor between the upper limit and the lower limit, the system further carries out comprehensive evaluation by combining the time sequence, the synchronous alignment offset degree and the like of the data segment in the charging behavior sequence, and dynamically configures the data uploading path and the splicing processing path of the data segment according to the context structure so as to realize finer frame structure control and behavior driving uploading strategy.

In a specific implementation process, the platform first sets a confidence factor threshold interval through the policy engine, for example, defines that the confidence factor is higher than 0.85 as the uploading priority and lower than 0.45 as the avoidance boundary. For each data segment, the data segment is initially classified according to its confidence factor, if the value is 0.9, the data segment is immediately added into the upload priority queue and inserted in front of the frame scheduling path, and if the value is 0.3, the data segment is assigned a logic avoidance flag and temporarily stored in the low priority buffer. For the intermediate state data segment between 0.45 and 0.85, the system invokes the timestamp field, the behavior state type and the uploading attached tag sequence index of the behavior event, evaluates the behavior continuity through the offset in the behavior time line, synthesizes the confidence factor to carry out weighted scoring, determines whether to incorporate a splicing priority group or delay uploading according to the weighted scoring result, and ensures that the high consistency and the dynamic suitability between the whole data uploading structure and the behavior context are maintained.

The above embodiments may be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented in software, the above-described embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the processes or functions in accordance with embodiments of the present application are produced in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from one website site, computer, server, or data center to another website site, computer, server, or data center by wired or wireless means (e.g., infrared, wireless, microwave, etc.). Computer readable storage media can be any available media that can be accessed by a computer or data storage devices, such as servers, data centers, etc. that contain one or more collections of available media. The usable medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium may be a solid state disk.

It should be understood that, in various embodiments of the present application, the sequence numbers of the foregoing processes do not mean the order of execution, and the order of execution of the processes should be determined by the functions and internal logic thereof, and should not constitute any limitation on the implementation process of the embodiments of the present application.

Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.

In the several embodiments provided in the present application, it should be understood that the disclosed system and method may be implemented in other manners. For example, the embodiments described above are merely illustrative, e.g., the partitioning of elements is merely a logical functional partitioning, and there may be additional partitioning in actual implementation, e.g., multiple elements or components may be combined or integrated into another system, or some features may be omitted, or not implemented. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other forms.

The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed over a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in the embodiments of the present application may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit.

The foregoing is merely illustrative embodiments of the present application, but the scope of the present application is not limited thereto, and any person skilled in the art can easily think about variations or substitutions within the technical scope of the present application, and the application should be covered. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A method for transmitting data from a charging pile, characterized by comprising the following steps:

S1. Construct a behavior context identifier chain for each set of frame data to be uploaded. The behavior context identifier chain generates a behavior perception matrix through the charging behavior event sequence, and generates an upload attachment tag sequence based on the behavior perception matrix, which is used to establish the binding relationship between data packets and charging behavior.

S2. Using an integrated state recognition algorithm composed of graph neural network and long short-term memory network, the uploaded attached tag sequence is compared with the real-time charging behavior to identify the message context failure state caused by the user terminating the charging operation in advance during the frame-by-frame upload process, and output the context validity vector.

S3. Construct a frame segment transmission impact map based on the context validity vector. Based on the nodes and their edge weight structure in the frame segment transmission impact map, calculate the behavioral dependency strength and validity weight of each remaining unuploaded data segment. The behavioral dependency strength is calculated based on the order difference of the behavioral event sequence index in the upload dependency label between nodes to determine the path tightness. The validity weight is calculated by weighting the validity label of the associated upload dependency label in the context validity vector and combining it with the node adjacency relationship to obtain the context transmission evaluation factor of each node. Compare the context transmission evaluation factor with the preset weight threshold to identify the set of data segments that need to be interrupted for uploading.

S4. Add logical isolation markers to the identified data segment set, and pass the logical isolation markers as input to the frame processing path filtering structure built based on the attention mechanism. Based on the data segment priority, behavior dependency chain and time synchronization field, determine the set of data segments that do not participate in the upload and splicing process.

S5. The behavior clustering engine built through the multi-layer perception structure performs real-time clustering of the remaining data segments to be uploaded, generates a frame consistency confidence factor, compares the frame consistency confidence factor with the policy threshold range, and performs dynamic adjustment of the data upload path and splicing processing path based on the comparison result.

2. The method for transmitting charging pile data according to claim 1, characterized in that S1 specifically includes the following steps:

Collect charging behavior events during the charging process, construct a charging behavior event sequence, each charging behavior event includes a time field, an operation type field, and an execution source field, and use the charging behavior event sequence as the input basis for the behavior context identifier chain to define the scope of the behavior context to which the framed data is attached;

Using the sequence of charging behavior events as input, a three-dimensional behavior perception matrix is constructed based on the order of events and the correlation between event types. The three dimensions of the behavior perception matrix represent the order of behavior events, the type of behavior events, and the level of event behavior driving, respectively, which are used to express the characteristics of charging behavior in a structured way.

The behavior perception matrix is analyzed to extract the behavior event segments that overlap with each group of frame data to be uploaded in time, and an upload dependency label is assigned to the frame data. The upload dependency label includes the corresponding behavior event sequence index, behavior state type and drive level identifier, which are used to represent the dependency relationship between the frame data and a specific charging behavior.

The upload attachment tag is bound to the data message structure of the corresponding frame data to establish a one-to-one binding mapping between the data message and the charging behavior, and the behavior context identifier chain is embedded into the frame data upload scheduling path to realize the continuous synchronization relationship maintenance between the uploaded data and the behavior event.

3. The method for transmitting charging pile data according to claim 1, characterized in that S2 specifically includes the following steps:

S201. Input the uploaded dependent label sequence into the graph neural network to construct the uploaded label behavior graph. Generate dependent connection edges between nodes based on the graph structure relationship between the behavior event sequence index and behavior state type of the uploaded dependent label. Extract the structural association features of the uploaded dependent label sequence in the behavior space to form a graph neural embedding representation.

S202. Construct a behavior evolution sequence of real-time charging behavior in chronological order, input it into a long short-term memory network, and extract the evolution feature vector of charging behavior at each stage through time-gated state transmission to express the continuity and termination trend of charging behavior.

S203. Based on the integrated state recognition algorithm, the graph neural embedding representation is compared with the long short-term memory feature vector to identify the message context failure state caused by the user terminating the charging operation in advance during the frame-by-frame upload process, and outputs a context validity vector composed of context validity tags corresponding to each uploaded attachment tag.

4. The method for transmitting charging pile data according to claim 3, wherein step S201 specifically comprises:

Each upload dependency tag in the upload dependency tag sequence is mapped to a node in the graph structure. Each node contains a behavior event sequence index field and a behavior state type field, which are used to define the set of graph nodes of the upload tag behavior graph.

Based on the sequential relationship of behavioral event sequence index and the semantic proximity of behavioral state type, dependency connection edges between graph nodes are constructed. Each dependency connection edge represents the structural dependency relationship between nodes through an adjacency matrix.

The constructed upload tag behavior graph is input into a graph neural network. The structural embedding features of the graph nodes are extracted through graph convolution operations, and the uploaded dependent tag sequence is encoded in graph space structure to form a graph neural embedding representation for subsequent behavior comparison.

5. The method for transmitting charging pile data according to claim 4, wherein S203 specifically comprises:

An integrated state recognition algorithm is constructed, which includes a structural input channel and a behavioral input channel. The structural input channel extracts the spatial attachment pattern between uploaded attachment tags based on graph neural embedding representation, and the behavioral input channel captures the evolution trajectory of charging behavior based on long short-term memory feature vectors. The graph neural embedding representation is input into the structural input channel, and the long short-term memory feature vector is input into the behavioral input channel to form a structure-behavior comparison benchmark.

Index consistency alignment relationships are established in the structural input channel and the behavioral input channel respectively. Node-level comparison operations are performed on the uploaded dependent label index dimension through attention alignment. The context consistency score sequence between the graph neural embedding representation and the long short-term memory feature vector is extracted to form a context difference map for failure identification.

Based on the contextual difference graph, a threshold range for behavior interruption judgment is set. Upload attachment tags with consistency scores below the lower limit of the threshold are marked as invalid, and upload attachment tags with scores above the upper limit of the threshold are marked as valid. The remaining tags are judged according to the nearest neighbor inference method. Finally, the contextual validity vector is output, which is composed of the contextual validity tags corresponding to all upload attachment tags.

6. The method for transmitting charging pile data according to claim 1, characterized in that S3 specifically includes the following steps:

S301. Establish a one-to-one mapping relationship between the validity markers of each upload dependency label in the context validity vector and the upload dependency labels of the remaining unuploaded data segments. Construct a frame segment transmission influence graph based on the mapping result. Each node in the frame segment transmission influence graph represents a remaining unuploaded data segment. Nodes are connected by edge calculation based on the similarity between the behavior event sequence index order, time field continuity and behavior state type in the upload dependency labels, forming a graph structure that reflects the data upload structure logic.

S302. Based on the nodes and their edge weight structure in the frame segment transmission influence graph, calculate the behavioral dependency strength and validity weight of each remaining unuploaded data segment. The behavioral dependency strength is calculated based on the order difference of the behavioral event sequence index in the upload dependency label between nodes to determine the path tightness. The validity weight is calculated based on the validity label associated with the upload dependency label in the context validity vector and combined with the node adjacency relationship to perform a weighted average, thus obtaining the context transmission evaluation factor of each node.

S303. The context transmission evaluation factor is compared with the preset weight threshold. If it is lower than the lower limit of the termination upload judgment threshold, the corresponding data segment is marked as an interrupted upload object. If it is higher than the upper limit of the maintain upload judgment threshold, the data segment continues to be uploaded. The remaining data segments are classified and identified according to the centrality weight distribution of the local subgraph of the graph structure, using a deviation judgment strategy based on local weighted average, and finally forming a set of data segments that need to be interrupted for upload.

7. The method for transmitting charging pile data according to claim 6, wherein step S302 specifically comprises:

Using the index value of the behavior event sequence in the upload-dependent tag of each node in the frame transmission influence graph as the index coordinate, the index difference between the current node and its direct neighboring nodes is calculated, and the edge weight is set according to the reciprocal of the index difference to measure the closeness of the behavior sequence between upload tags.

Based on the edge weights between nodes in the graph structure, the behavioral dependency strength of each node is defined as the weighted average of the edge weights of all its neighboring nodes, reflecting the degree of aggregation of the dependency strength between the current data segment and other data segments in terms of upload order.

Extract the corresponding validity tag value of each node's uploaded attachment tag in the context validity vector, and sum the validity tag value with the validity tag values of the adjacent nodes according to the edge weight to obtain the validity weight of each node. Then, normalize and weight the behavior attachment strength and validity weight to obtain the context transmission evaluation factor corresponding to the node.

8. The method for transmitting charging pile data according to claim 1, wherein S4 specifically comprises:

For the identified set of data segments, logical isolation markers are inserted based on the information identification field. The logical isolation markers consist of Boolean isolation bits and context isolation factors, which are used to indicate the independent state and behavior failure level of the data segments in subsequent processing.

The data segments with embedded logical isolation tags are input into the frame processing path filtering structure. The frame processing path filtering structure is built based on the attention mechanism, treating each data segment as an independent query unit and generating an attention distribution vector in the entire sequence of data segments to be processed.

Based on the priority value of each data segment, the node depth in the behavior dependency chain, and the alignment offset of the time synchronization field, a multi-factor weighted score is performed in conjunction with the logical isolation mark. The attention-weighted score determines whether the data segment enters the splicing path. Data segments with scores below a preset threshold are marked as objects that do not participate in uploading and splicing processing.

9. The method for transmitting charging pile data according to claim 1, wherein S5 specifically comprises:

A multi-layer perceptual structure containing an input embedding layer, a feature extraction layer, and a semantic aggregation layer is constructed. The context feature vectors of the remaining data segments to be uploaded are input into this structure to extract the behavioral pattern representations of each data segment.

Based on the extracted behavioral pattern representation, the remaining data segments to be uploaded are clustered in the embedding space using a behavioral clustering engine. The consistency distribution density within the cluster to which each data segment belongs is calculated, and the corresponding frame consistency confidence factor is generated.

The frame consistency confidence factor is compared with the preset policy threshold range. If the factor value is higher than the upper limit of the threshold, the upload path is prioritized and scheduled. If it is lower than the lower limit of the threshold, it is marked as a splicing avoidance object. The upload path and splicing processing path of the remaining data segments are dynamically configured according to the confidence factor and time series characteristics.