CN117332818A - Fault diagnosis method based on self-adaptive graph neural network multi-source data fusion - Google Patents
Fault diagnosis method based on self-adaptive graph neural network multi-source data fusion Download PDFInfo
- Publication number
- CN117332818A CN117332818A CN202210795584.0A CN202210795584A CN117332818A CN 117332818 A CN117332818 A CN 117332818A CN 202210795584 A CN202210795584 A CN 202210795584A CN 117332818 A CN117332818 A CN 117332818A
- Authority
- CN
- China
- Prior art keywords
- graph
- neural network
- adaptive
- fusion
- graph neural
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a fault diagnosis method based on self-adaptive graph neural network multi-source data fusion, which comprises the following steps: constructing an adaptive graph neural network model, wherein the adaptive graph neural network model learns a plurality of graph structures by utilizing an adjacent matrix of the graph, estimates fault types by fusing the plurality of graph structures, and comprises the following steps: a 1 x 1 convolution layer for fusing different source data; k layers of graph neural networks, wherein each layer of graph neural network performs feature fusion of different layers in a jumping connection mode; and collecting data of different sensing sources, preprocessing each data, and inputting the preprocessed data into the self-adaptive graph neural network model. The invention can improve the capability of fusing and learning task for complex and changeable multi-source data.
Description
Technical Field
The invention relates to the field of fault diagnosis, in particular to a fault diagnosis method based on self-adaptive graph neural network multi-source data fusion.
Background
The high-efficiency operation and maintenance of the industrial system are not separated from the professional fault diagnosis technology. The state change of a large industrial system is related to various information such as system environment, internal complex structure, physical working principle and the like, the corresponding monitoring sensors are various, and meanwhile, the collected data has the characteristics of large scale, high dimension, complex coupling relation, high noise, time varying, non-stability and the like. Thus, there is a complex association between the different variables of the data. The traditional fault diagnosis method based on data fusion considers time and space dimension characteristics by using a fixed model, so that the interpretation capability of the model on complex relationships among multidimensional variables is insufficient, and further, the equipment state/fault diagnosis requirements in an industrial scene are difficult to meet.
The existing fault diagnosis technology based on multi-source data fusion mainly comprises the following steps:
(1) Based on data-level fusion fault diagnosis, the original data collected by a sensor is generally used for directly processing, various sensing signals such as vibration, sound, current, angular speed and the like are fused into a single-path signal, and then equipment faults are mined or inferred from the original data through means such as machine learning and the like. The method has the advantages of small loss of sensing information, high communication requirement and high fault diagnosis and reasoning difficulty.
(2) The fault diagnosis technology based on feature level fusion firstly extracts features of data acquired by a plurality of sensors, and then performs feature fusion. The method generally adopts the theory of evidence theory, decision theory, fuzzy theory and the like to combine with priori knowledge to construct a fault diagnosis system. Therefore, the dimension, communication load requirement and calculation capability requirement of the fusion data can be reduced, but the accuracy of fault diagnosis is affected due to the loss of the sensing information.
(3) The decision stage fusion is carried out, firstly, primary decision results are obtained by processing the characteristics of the multi-source sensing information through pattern recognition, bayesian network reasoning and the like, and then a plurality of primary decision results are fused by using methods such as D-S evidence theory and the like to obtain final decision results. Because the decision-level fusion depends on the internal operation mechanism of the equipment and the prior expert knowledge base, the fault diagnosis of complex relations among different variables of multi-source data is difficult to meet.
In order to cope with complex relations among different variables of data, a deep learning-based method is widely applied to fault diagnosis with high diagnosis accuracy, and a processing method of multi-branch feature fusion in a time dimension is generally adopted by a convolutional neural network (Convolutional Neural Network, CNN) and a long-short-time memory network (Long Short Term Memory Network, LSTM), but the interpretation power of the association relation features among multi-source data is insufficient. For this reason, patent CN112783940a discloses a fault diagnosis method and medium for multi-source time sequence data based on graph neural network, which can combine the relevant features and time sequence features of the multi-source data, and improve the accuracy and noise immunity of fault diagnosis. However, because the graph neural network of the scheme is of a fixed topological structure, the graph neural network is still difficult to be suitable for complex and changeable multi-source data fusion and learning tasks.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: aiming at the technical problems in the prior art, the invention provides a fault diagnosis method based on adaptive graph neural network multi-source data fusion, which can improve the capability of fusion and learning tasks of complex and changeable multi-source data.
In order to solve the technical problems, the technical scheme provided by the invention is as follows:
the fault diagnosis method based on the adaptive graph neural network multi-source data fusion is characterized by comprising the following steps of:
constructing an adaptive graph neural network model, wherein the adaptive graph neural network model learns a plurality of graph structures by utilizing an adjacent matrix of the graph, and estimates the fault type by fusing the plurality of graph structures;
and collecting data of different sensing sources, preprocessing each data, and inputting the preprocessed data into the self-adaptive graph neural network model.
Further, the adaptive graph neural network model includes:
the 1 multiplied by 1 convolution layer is used for fusing different source data and then inputting the fused source data into the K-layer graph neural network;
in the K-layer graph neural network, each graph neural network performs feature fusion of different levels in a jump connection mode, and each graph neural network comprises:
the time convolution layer is used for extracting the time dimension characteristics of the data;
and the adaptive graph convolution layer is used for generating an adaptive adjacency matrix to characterize different graph structures and fusing the different graph structures.
Further, the gating mechanism expression in the time convolution layer is:
wherein X is input time series data having S time steps,wherein N is the variable type, namely the sensing source number, D is the node hidden vector dimension, h (k) Output characteristics of time convolution layer of neural network for k-th layer graph d Indicating an extended causal convolution operation, by which is an element-wise multiplication operation, tanh (·) indicates a tanh nonlinear activation function, are all network model parameters.
Further, the step of generating the adaptive adjacency matrix specifically includes:
convolving the last layer of time into a networkOutput characteristic h of (2) K As a node vector dictionary D 1 Setting a parameter matrix as a node vector dictionary D 2 Using two node vector dictionaries D 1 ，D 2 Respectively expressed as a source node vector set and a target node vector set, D 1 ，D 2 ∈R N×(D×S) According to the node vector dictionary D 1 ，D 2 Calculating node dependency relation weight z;
after normalizing the node dependency weight z, intercepting a positive value z in the node dependency weight z cut ；
For z cut Normalizing to obtain an adaptive adjacency matrix A adp 。
Further, intercept the positive value z in node dependency weight z cut The method specifically comprises the following steps: results z normalized to node dependency weights z norm And performing nonlinear activation on the ReLU and the tanh, and further cutting off the positive value in the activated weight.
Further, the specific steps of fusing different graph structures include:
according to the structure G of the previous fusion diagram i-1 And the initial graph structure of the current inputObtain the reset gate r i Updating door z i State of (2);
reset the current gate r i With the previous fusion diagram structure G i-1 And (3) carrying out Hadamard product and updating the current candidate fusion graph G' i ；
Using update gate z i Combine the above fusion graph structure G i-1 And candidate fusion map G' i Obtaining a current fusion graph G i According to the current fusion graph G i Updating the output of the adaptive graph convolution layer.
Further, reset gate r i Updating door z i The expression of (2) is:
where σ (·) is the Sigmoid activation function,as a parameter of the weight-bearing element,is a bias parameter.
Further, the current candidate fusion map G' i The expression is:
wherein the method comprises the steps ofFor the network weight parameter, +.>Is a bias parameter for the network.
Further, the current fusion graph G i The expression is:
G i ＝z i ⊙G i-1 +(1-z i )⊙G′ i
wherein +.
Further, the output expression of the adaptive graph convolution layer is:
wherein, K is the number of propagation steps,respectively represent forward transfer matrix M f Backward directionTransfer matrix M f Backward transfer matrix M b And adaptive graph G i Power series of>The network parameters of forward propagation convolution, backward propagation convolution, and adaptive propagation convolution, respectively.
Compared with the prior art, the invention has the advantages that:
aiming at the problem of multi-source sensing data fusion of complex association relations in the existing fault diagnosis, the invention fully excavates association relations of different types and different degrees among variables through self-adaptive learning multi-graph structure to construct a graph neural network model. And then, inputting the learned graph structure into a graph neural network sharing parameters, and combining input data to obtain corresponding feature vectors, thereby improving the complex and changeable multi-source data fusion and learning task capacity.
Drawings
FIG. 1 is a flow chart of a method according to an embodiment of the invention.
Fig. 2 is a schematic diagram of a gating mechanism in a time convolution layer according to an embodiment of the present invention.
FIG. 3 is a schematic diagram of a gating mechanism of a gating unit in a convolutional layer of an adaptive graph according to an embodiment of the present invention.
Detailed Description
The invention is further described below in connection with the drawings and the specific preferred embodiments, but the scope of protection of the invention is not limited thereby.
The embodiment provides a fault diagnosis method based on adaptive graph neural network multi-source data fusion, as shown in fig. 1, the flow mainly comprises: and (3) carrying out proper pretreatment (resampling and data normalization) on signals of various acquired sensing data, such as vibration, sound, heat, current, voltage, capacitance, inductance and the like, wherein the result is used as the input of a next-stage adaptive graph neural network model. The adaptive graph neural network model learns multiple graph structures using the neighbor matrix of the graph and estimates device status/fault type by fusing the multiple graph structures. Each step is described in detail below:
(1) Multisource data preprocessing
For industrial multisource sensing data, data preprocessing is required to obtain an optimal fault diagnosis model between deep learning method processes. The main flow comprises the following steps: resampling and data normalization. The resampling is used for ensuring that the time intervals of sampling points of variables with different dimensions are consistent, and generally adopts two types of downsampling and upsampling. The data normalization is to convert data into dimensionless data, and mainly adopts three methods of normalization, centering and standardization.
(2) Self-adaptive graph neural network model fault diagnosis
Aiming at the intricate and complex association relations among different sensing data, the embodiment provides a self-adaptive heterogeneous graph structure learning method, a gating fusion unit applicable to graph structures is built by taking a gating circulation unit (Gate Recurrent Unit, GRU) as a core, and a plurality of learned graph structures are fused finally, so that a richer variable relation is obtained.
As shown in fig. 1, the adaptive graph neural network model of the embodiment firstly fuses different source data through a 1×1 convolution layer, then inputs the fused source data into a K-layer graph neural network, performs feature fusion of different levels through a jump connection mode, and finally outputs device state/fault type classification through softmax. For each layer of graph neural network, as shown in fig. 1, it includes a time convolution layer and an adaptive graph convolution layer, and the design of the time convolution layer and the adaptive graph convolution layer in this embodiment is as follows:
(a) Time convolution layer:
the time convolution network is used for extracting features in the time dimension, namely the time dependency relationship. Because the gating mechanism plays an important role in the information flow of the time convolution network, as shown in fig. 2, the method adopts the combination of Sigmoid and tanh nonlinear activation functions to realize the gating mechanism. Given a period of input time series data X ε R with S time steps N×D×S Where N is the variable type, i.e. the number of sensing data sources, D is the node hidden vector dimension, and the input time convolutions are shown in fig. 1. The gate control mechanism is specifically expressed as:
wherein h is (k) Output characteristics of extended causal convolution with gating mechanism for a neural network of a k-th layer graph d Indicating an extended causal convolution operation, by which is an element-wise multiplication operation, tanh (·) indicates a tanh nonlinear activation function. Are all network model parameters. Where h is (k) An adaptive graph convolutional layer is normalized into a k-layer graph neural network.
(b) Adaptive graph convolution layer:
the graph rolling network adjusts the graph neural network structure through adaptive adjacency matrix generation, and fuses a plurality of heterogeneous graph structures through a gating fusion unit. The self-adaptive adjacent matrix and the gating fusion unit are specifically designed as follows:
(b-1) adaptive adjacency matrix generation: the adjacency matrix is a learnable node dependency matrix. Using two node vector dictionaries D 1 ，D 2 Respectively expressed as a source node feature set and a target node vector set, wherein D 1 Output characteristic h for a time convolution network K (K is the last layer time convolution network), D 2 For parameter matrix, randomly initializing it, D 1 ，D 2 ∈R N ×(D×S) . The node dependency weight z is:
normalizing z and normalizing the normalized result z norm Non-linear activation of ReLU and tanh, and further cutting off positive value in weight
Where ε is the truncation constant. Finally, the normalized index (Softmax) function is used for the pair z cut Normalizing to obtain an adjacent matrix A adp ：
Therefore, the adjacency matrix learns and constructs the dependency relationship among the sensing nodes according to the characteristics extracted by the time convolution network.
(b-2) gating fusion unit, for inputting different time points of time sequence data, obtaining q initial adjacent matrixes by a generation method of the adaptive adjacent matrixes, and representing different graph structuresThe initial graph structure is subjected to depth fusion by using a gating fusion unit, as shown in fig. 3.
The gate control unit mainly comprises two gate control mechanisms, namely an update gate and a reset gate. Let h denote the number of gating units, and according to the fusion graph structure G transmitted by the previous graph neural network i-1 And the initial graph structure of the current inputObtain the reset gate r i Updating door z i State of (2):
wherein sigma (·) is Sigmoid activation function, W ar ，W az ∈R N×N ，W hr ，W hz ∈R N×N B is a weight parameter of the model r ，b z ∈R 1×N The last fusion graph structure of the first layer graph neural network is obtained by random initialization as a bias parameter of the model.
Reset gate r is then used i And calculating a candidate fusion map. Reset the current gate r i With the previous fusion diagram structure G i-1 And (5) carrying out Hadamard product to update the candidate fusion map of the current map. Candidate fusion map G' i ∈R N×N Expressed as:
wherein W is ah ，W hh ∈R N×N B is a network weight parameter h ∈R 1×N Is a bias parameter for the network.
Finally use update door z i Combine the above fusion graph structure G i-1 And candidate fusion map G' i Obtaining a current fusion graph G i ∈R N×N ：
G i ＝z i ⊙G i-1 +(1-z i )⊙G′ i (9)
The update gate mechanism is used for learning the dependency relationship between the graph structures with farther position intervals, and the reset gate mechanism is used for updating the dependency relationship between the graph structures with adjacent positions, so that the more complex association relationship between the multiple variables is learned. The final adaptive graph convolution layer is:
wherein Z is i Is G i The corresponding adaptive graph convolving layer outputs, K is the number of propagation steps,respectively represent forward transfer matrix M f Backward transfer matrix M f Backward transfer matrix M b And adaptive fusion map G i Power series of> The network parameters of the forward propagation convolution, the backward propagation convolution and the self-adaptive propagation convolution are respectively shown, and X is the standard layer output result of the upper layer graph neural network.
Results Z of adaptive graph convolutional layer i The results of the fault classification were obtained through conventional standardized, fully-connected, softmax layers.
In summary, compared with the prior art, the method has the main advantage that the association relations of different types and different degrees between the variables can be fully mined. The multi-graph neural network structure is adaptively learned, and the association characteristics among more variables are extracted through fusion of the multi-graph structure, so that the complex and changeable multi-source data fusion and learning task capacity is improved.
The foregoing is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. While the invention has been described with reference to preferred embodiments, it is not intended to be limiting. Therefore, any simple modification, equivalent variation and modification of the above embodiments according to the technical substance of the present invention shall fall within the scope of the technical solution of the present invention.
Claims (10)
1. The fault diagnosis method based on the adaptive graph neural network multi-source data fusion is characterized by comprising the following steps of:
constructing an adaptive graph neural network model, wherein the adaptive graph neural network model learns a plurality of graph structures by utilizing an adjacent matrix of the graph, and estimates the fault type by fusing the plurality of graph structures;
and collecting data of different sensing sources, preprocessing each data, and inputting the preprocessed data into the self-adaptive graph neural network model.
2. The fault diagnosis method based on adaptive graph neural network multi-source data fusion according to claim 1, wherein the adaptive graph neural network model comprises:
the 1 multiplied by 1 convolution layer is used for fusing different source data and then inputting the fused source data into the K-layer graph neural network;
in the K-layer graph neural network, each graph neural network performs feature fusion of different levels in a jump connection mode, and each graph neural network comprises:
the time convolution layer is used for extracting the time dimension characteristics of the data;
and the adaptive graph convolution layer is used for generating an adaptive adjacency matrix to characterize different graph structures and fusing the different graph structures.
3. The fault diagnosis method based on adaptive graph neural network multi-source data fusion according to claim 2, wherein the gating mechanism expression in the time convolution layer is:
wherein X is input time series data having S time steps,wherein N is the variable type, namely the sensing source number, D is the node hidden vector dimension, h (k) Output characteristics of time convolution layer of neural network for k-th layer graph d Indicating an extended causal convolution operation, by which is an element-wise multiplication operation, tanh (·) indicates a tanh nonlinear activation function, +.> Are all network model parameters.
4. The fault diagnosis method based on adaptive graph neural network multi-source data fusion according to claim 2, wherein the step of generating the adaptive adjacency matrix specifically comprises:
output characteristic h of last layer time convolution network K As a node vector dictionary D 1 Setting a parameter matrix as a node vector dictionary D 2 Using two node vector dictionaries D 1 ，D 2 Respectively expressed as a source node vector set and a target node vector set, D 1 ，D 2 ∈R N×(D×S) According to the node vector dictionary D 1 ，D 2 Calculating node dependency relation weight z;
after normalizing the node dependency weight z, intercepting a positive value z in the node dependency weight z cut ；
For z cut Normalizing to obtain an adaptive adjacency matrix A adp 。
5. The fault diagnosis method based on adaptive graph neural network multi-source data fusion according to claim 4, wherein a positive value z in node dependency relation weight z is intercepted cut The method specifically comprises the following steps: results z normalized to node dependency weights z norm And performing nonlinear activation on the ReLU and the tanh, and further cutting off the positive value in the activated weight.
6. The fault diagnosis method based on adaptive graph neural network multi-source data fusion according to claim 2, wherein the specific step of fusing different graph structures comprises the following steps:
fusion graph structure G according to last transferred i-1 And the initial graph structure of the current inputObtain the reset gate r i Updating door z i State of (2);
reset the current gate r i With the previous fusion diagram structure G i-1 And (3) carrying out Hadamard product and updating the current candidate fusion graph G' i ；
Using update gate z i Combine the above fusion graph structure G i-1 And candidate fusion map G' i Obtaining a current fusion graph G i According to the current fusion graph G i Updating the output of the adaptive graph convolution layer.
7. The fault diagnosis method based on adaptive graph neural network multi-source data fusion according to claim 6, wherein the reset gate r is i Updating door z i The expression of (2) is:
where σ (·) is the Sigmoid activation function,b is a weight parameter r ，Is a bias parameter.
8. The fault diagnosis method based on adaptive graph neural network multi-source data fusion according to claim 6, wherein the current candidate fusion graph G' i The expression is:
wherein the method comprises the steps ofFor the network weight parameter, +.>Is a bias parameter for the network.
9. The fault diagnosis method based on adaptive graph neural network multi-source data fusion according to claim 6, wherein the current fusion graph G i The expression is:
G i ＝z i ⊙G i-1 +(1-z i )⊙G′ i
wherein +..
10. The fault diagnosis method based on adaptive graph neural network multi-source data fusion according to claim 6, wherein the output expression of the adaptive graph convolution layer is:
wherein, K is the number of propagation steps,respectively represent forward transfer matrix M f Backward transfer matrix M f Backward transfer matrix M b And adaptive graph G i Power series of>The network parameters of forward propagation convolution, backward propagation convolution, and adaptive propagation convolution, respectively.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210795584.0A CN117332818A (en) | 2022-07-07 | 2022-07-07 | Fault diagnosis method based on self-adaptive graph neural network multi-source data fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210795584.0A CN117332818A (en) | 2022-07-07 | 2022-07-07 | Fault diagnosis method based on self-adaptive graph neural network multi-source data fusion |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117332818A true CN117332818A (en) | 2024-01-02 |
Family
ID=89281748
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210795584.0A Pending CN117332818A (en) | 2022-07-07 | 2022-07-07 | Fault diagnosis method based on self-adaptive graph neural network multi-source data fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117332818A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118331147A (en) * | 2024-05-15 | 2024-07-12 | 希格玛电气(珠海)有限公司 | Intelligent energy-saving control system of electrical equipment |
-
2022
- 2022-07-07 CN CN202210795584.0A patent/CN117332818A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118331147A (en) * | 2024-05-15 | 2024-07-12 | 希格玛电气(珠海)有限公司 | Intelligent energy-saving control system of electrical equipment |
| CN118331147B (en) * | 2024-05-15 | 2024-10-01 | 希格玛电气(珠海)有限公司 | Intelligent energy-saving control system of electrical equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Malekian et al. | Concepts, procedures, and applications of artificial neural network models in streamflow forecasting | |
| Haarnoja et al. | Latent space policies for hierarchical reinforcement learning | |
| CN112801404B (en) | Traffic prediction method based on self-adaptive space self-attention force diagram convolution | |
| CN119067028B (en) | Integrated circuit process parameter optimization method and system based on machine learning | |
| Das et al. | Artificial neural network trained by particle swarm optimization for non-linear channel equalization | |
| Leng et al. | Design for self-organizing fuzzy neural networks based on genetic algorithms | |
| CN113176776B (en) | Unmanned ship weather self-adaptive obstacle avoidance method based on deep reinforcement learning | |
| CN112765896A (en) | LSTM-based water treatment time sequence data anomaly detection method | |
| CN115618296A (en) | Anomaly detection method for time-series data of dam monitoring based on graph attention network | |
| CN109241291A (en) | Knowledge mapping optimal path inquiry system and method based on deeply study | |
| CN108596327B (en) | A Deep Learning-Based Artificial Intelligence Picking Method for Seismic Velocity Spectrum | |
| Barzola-Monteses et al. | Energy consumption of a building by using long short-term memory network: a forecasting study | |
| CN117556310B (en) | Spacecraft residual life prediction method | |
| CN116384450B (en) | Deep Convolutional Fuzzy Neural Networks for Medical Data and Their Training Methods | |
| Hsu et al. | Remaining useful life prediction based on state assessment using edge computing on deep learning | |
| CN115564155A (en) | A distributed wind turbine power prediction method and related equipment | |
| CN117035013A (en) | Method for predicting dynamic network link by adopting impulse neural network | |
| JP7541692B1 (en) | A method for evaluating networks of neuro-fuzzy systems with learning capabilities | |
| Delgado et al. | A multiobjective genetic algorithm for obtaining the optimal size of a recurrent neural network for grammatical inference | |
| Chouikhi et al. | Single-and multi-objective particle swarm optimization of reservoir structure in echo state network | |
| CN108537366A (en) | Reservoir scheduling method based on optimal convolution bidimensionalization | |
| CN112651519A (en) | Secondary equipment fault positioning method and system based on deep learning theory | |
| Xing et al. | Predicting evolving chaotic time series with fuzzy neural networks | |
| CN117727178A (en) | A large-scale traffic prediction method and system integrating long-distance spatio-temporal correlation | |
| CN112015894B (en) | Text single class classification method and system based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |