CN117649007A - Lightweight urban space-time data prediction method and device - Google Patents

Abstract

This application discloses a lightweight urban spatiotemporal data prediction method, device, electronic equipment and storage medium, which belongs to the field of urban data prediction technology. The method includes: constructing a causal expansion convolution module, and the causal expansion convolution module is used for mining The temporal correlation of urban spatio-temporal data; construct a graph expansion convolution module, which is used to mine the spatial correlation of urban spatio-temporal data; build a spatio-temporal graph expansion model based on the causal expansion convolution module and the graph expansion convolution module; Historical urban spatiotemporal data is brought into the spatiotemporal graph expansion model to obtain the prediction results of urban spatiotemporal data. This method improves the prediction accuracy of urban spatio-temporal data by mining the time correlation and spatial correlation of urban spatio-temporal data. At the same time, it reduces the parameter scale and improves the calculation speed by introducing expansion operations, which can simultaneously improve the prediction of this prediction method. Precision and ease of use.

Description

A lightweight urban spatiotemporal data prediction method and device

Technical field

This application belongs to the technical field of lightweight urban spatiotemporal data prediction, and specifically relates to a lightweight urban spatiotemporal data prediction method, device, electronic equipment and storage medium.

Background technique

Spatiotemporal prediction is a popular research topic in the field of urban computing and is of great significance to urban planning and intelligent transportation. Currently, there are many attempts to use various deep learning models to predict the spatiotemporal state of unknown systems. However, most of the existing prediction models tend to improve the prediction accuracy while ignoring the ease of use in practical applications, such as low time efficiency and large parameter scale.

Although the current spatiotemporal prediction model has achieved excellent prediction accuracy in spatiotemporal prediction tasks, there are still shortcomings. Specifically, most existing spatiotemporal prediction models focus on improving the prediction accuracy of the model, but ignore the ease of use of the model. The improvement of model prediction accuracy is often accompanied by an increase in model complexity, which not only increases the difficulty of model implementation, but also makes the calculation efficiency of the model low and the parameter scale large. In actual scenarios, the ease of use of models has always been the focus of many scholars. Improving model prediction accuracy without significantly increasing model calculation time/model parameter scale is a top priority in urban computing. However, most existing spatiotemporal prediction models still struggle to strike a balance between model prediction accuracy and ease of use.

Contents of the invention

The purpose of this application is to provide a lightweight urban spatiotemporal data prediction method, a lightweight urban spatiotemporal data prediction device, electronic equipment and storage media to solve the problem of existing lightweight urban spatiotemporal data prediction that is difficult to take into account the prediction accuracy. and model usability issues.

According to the first aspect of the embodiments of this application, a lightweight urban spatiotemporal data prediction method is provided. The method may include:

Construct a causal dilated convolution module, which is used to mine the time correlation of urban spatiotemporal data;

Construct a graph dilation convolution module, which is used to mine the spatial correlation of the city's spatiotemporal data;

Construct a space-time graph expansion model based on the causal expansion convolution module and the graph expansion convolution module;

The historical urban spatiotemporal data is brought into the spatiotemporal graph expansion model to obtain the prediction results of the urban spatiotemporal data.

In some optional embodiments of the present application, building a causal dilated convolution module includes:

Construct causal dilation convolution operator;

The causal dilation convolution module is obtained by adding a residual link and a regularization module to the causal dilation convolution operator.

In some optional embodiments of this application, the forward propagation formula of the causal dilated convolution operator is as follows:

in, For the time series of urban spatiotemporal data/> middle Perform causal expansion operation with expansion factor d;/> is the urban spatio-temporal data observed by urban sensor node v _i in the time window τ;/> is the convolution kernel; K is the size of the convolution kernel.

In some optional embodiments of this application, the forward propagation formula of the causal dilated convolution module is as follows:

in, To perform the forward propagation operation of the causal expansion convolution module on the time series of urban spatiotemporal data;/> is the intermediate variable of city sensor node v _i in the causal dilation convolution module; H _i is the output variable of city sensor node v _i in the causal dilation convolution module; e _h is the number of convolution kernels; For the time series of urban spatiotemporal data/> Perform causal convolution operation; Norm is the parameter regularization function; Relu is the activation function.

In some optional embodiments of this application, building a graph dilation convolution module includes:

Construct an expanded adjacency matrix of urban sensor nodes;

Construct a graph dilated convolution operator based on the dilated adjacency matrix;

A graph dilation convolution module is constructed based on the graph dilation convolution operator.

In some optional embodiments of this application, the calculation formula of the expanded adjacency matrix is as follows:

Among them, A ^dilation=d is the expanded adjacency matrix; is the expansion adjacency relationship between urban sensor nodes v _i and v _j ; d is the expansion factor; A ^d is the dth power of the first-order topological adjacency matrix.

In some optional embodiments of this application, the calculation formula of the graph dilation convolution operator is as follows:

in, is the graph expansion convolution operation with the expansion factor d for the city sensor node v _i ; γ _ji is the influence weight of the sensor node v _j on the sensor node _vi ; H _i is the city sensor node v _i in the causal expansion convolution module The output variable;/> is the expansion adjacency matrix with expansion factor d; W _q and W _v are learnable parameters; Relu is the activation function; exp is the exponential function; [·||·] is the matrix connection function.

In some optional embodiments of this application, the forward propagation formula of the graph dilation convolution module is as follows:

in, is a space-time tensor, which is the output of the causal dilated convolution module.

In some optional embodiments of the present application, building a spatiotemporal graph expansion model based on the causal expansion convolution module and the graph expansion convolution module includes:

Construct a spatiotemporal graph dilation convolution block, which includes the causal dilation convolution module and the graph dilation convolution module in series;

Concatenate N of the spatio-temporal graph dilation convolution blocks to obtain the spatio-temporal graph dilation model;

Among them, N is a natural number greater than 1.

In some optional embodiments of this application, the calculation formula of the space-time graph expansion model is as follows:

in, For historical city spatiotemporal data,/> is the output of the Nth space-time graph dilation convolution block;/> is the prediction result of urban spatiotemporal data; q is the prediction step length of the model;/> is the space-time tensor/> Do the convolution operation.

According to the second aspect of the embodiment of the present application, a lightweight urban spatiotemporal data prediction device is provided. The device may include:

The first data processing module is used to construct a causal dilated convolution module, which is used to mine the time correlation of urban spatiotemporal data;

The second data processing module is used to construct a graph expansion convolution module. The graph expansion convolution module is used to mine the spatial correlation of the city's spatiotemporal data;

A third data processing module, configured to construct a spatiotemporal graph expansion model based on the causal expansion convolution module and the graph expansion convolution module;

A prediction module is used to bring historical urban spatiotemporal data into the spatiotemporal graph expansion model to obtain prediction results of urban spatiotemporal data.

According to a third aspect of the embodiment of the present application, an electronic device is provided, and the electronic device may include:

processor;

Memory used to store instructions executable by the processor;

Wherein, the processor is configured to execute instructions to implement the lightweight urban spatiotemporal data prediction method as described in any embodiment of the first aspect.

According to a fourth aspect of the embodiments of the present application, a storage medium is provided. When instructions in the storage medium are executed by a processor of an information processing device or a server, the information processing device or the server implements any of the first aspects. The lightweight urban spatiotemporal data prediction method described in the embodiment.

The above technical solution of the present application has the following beneficial technical effects:

The lightweight urban spatiotemporal data prediction method provided by the embodiments of this application improves the prediction accuracy of urban spatiotemporal data by mining the time correlation and spatial correlation of urban spatiotemporal data. At the same time, the parameter scale is reduced by introducing expansion operations. , which improves the calculation speed and can simultaneously improve the prediction accuracy and ease of use of the prediction method.

Description of drawings

Figure 1 is a schematic flowchart of the construction process of the space-time graph expansion model in an exemplary embodiment of the present application;

Figure 2 is a schematic flow chart of a lightweight urban spatiotemporal data prediction method provided by an exemplary embodiment of the present application;

Figure 3 is a schematic structural diagram of the space-time graph expansion model in an exemplary embodiment of the present application;

Figure 4 is a schematic diagram comparing causal convolution and causal dilation convolution in an exemplary embodiment of the present application;

Figure 5 is a schematic diagram of the expanded adjacency relationship of the target graph node in an exemplary embodiment of the present application;

Figure 6 is a schematic diagram of the construction process of the expanded adjacency matrix in an exemplary embodiment of the present application;

Figure 7 is a schematic structural diagram of a lightweight urban spatiotemporal data prediction device in an exemplary embodiment of the present application;

Figure 8 is a schematic structural diagram of an electronic device in an exemplary embodiment of the present application;

Figure 9 is a schematic diagram of the hardware structure of an electronic device in an exemplary embodiment of the present application.

Detailed ways

In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the specific embodiments and the accompanying drawings. It should be understood that these descriptions are exemplary only and are not intended to limit the scope of the application. Furthermore, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily confusing the concepts of the present application.

A schematic diagram of a layer structure according to an embodiment of the present application is shown in the accompanying drawings. The drawings are not drawn to scale, with certain details exaggerated for clarity and may have been omitted. The shapes of the various regions and layers shown in the figures, as well as the relative sizes and positional relationships between them are only exemplary. In practice, there may be deviations due to manufacturing tolerances or technical limitations, and those skilled in the art will base their judgment on actual situations. Additional regions/layers with different shapes, sizes, and relative positions can be designed as needed.

Obviously, the described embodiments are part of the embodiments of the present application, but not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of this application.

In the description of the present application, it should be noted that the terms "first", "second" and "third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

In addition, the technical features involved in different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

Research has found that although existing spatiotemporal prediction models have achieved better prediction accuracy in spatiotemporal prediction tasks, they still have shortcomings, such as the high complexity of the model. The high complexity of the model not only increases the difficulty of model implementation, but also makes the model's calculation efficiency low and its parameter scale large. In actual scenarios, the ease of use of models has always been the focus of many scholars. However, most existing spatiotemporal prediction models are still difficult to balance between model prediction accuracy and ease of use.

In view of this, this application proposes a novel lightweight spatiotemporal prediction model that takes into account model prediction accuracy and ease of use, namely the Spatiotemporal Graph Dilated Convolutional Network (STGDN), for prediction tasks of urban spatiotemporal data. The STGDN model aims to improve the prediction accuracy of lightweight urban spatiotemporal data prediction tasks without significantly increasing the model calculation time/model parameter scale. Based on the spatiotemporal graph expansion model, this application proposes a lightweight urban spatiotemporal data prediction method, a lightweight urban spatiotemporal data prediction device, electronic equipment and storage media.

Definitions in this application include:

Graph G (Graph): As shown in (a) and (b) in Figure 1, the research area of urban spatiotemporal data can be abstracted into a graph structure G=<V,E,A>, where Represents n city sensors in graph G, that is, n graph nodes; E represents the correlation between city sensors. For simplicity, the connection relationship between city sensors can be matrix/> represents, where A _ij represents the topological connection relationship between node _vi and node v _j .

Spatiotemporal State Matrix: The urban spatiotemporal data monitored by all urban sensors in Figure G can be expressed as a spatiotemporal state matrix. Among them, urban spatiotemporal data/> Represents the urban spatiotemporal data value (traffic flow and air quality per unit time) monitored by a single city sensor v _i within the time window t, Represents the spatial sequence monitored by all city sensors within the time window t.

As shown in (c) in Figure 1, this application aims to establish a function model This model can rely on the graph structure G to mine the spatiotemporal patterns in urban sensor data from the spatiotemporal state matrix X, thereby accurately predicting future urban spatiotemporal data. Given a space-time state matrix X, its modeling process is shown in formula (1).

In the formula: Indicates that this application proposes the STGDN model;/> Represents the historical spatiotemporal data that the STGDN model needs to input, where p represents the time-dependent step;/> Represents future (prediction) spatio-temporal data, q represents the prediction step, q=1 represents single-step prediction, q>1 represents multi-step prediction; W represents the learnable parameters in the STGDN model.

The lightweight urban spatiotemporal data prediction method, lightweight urban spatiotemporal data prediction device, electronic equipment and storage medium provided by the embodiments of the present application will be described in detail below with reference to the accompanying drawings through specific embodiments and application scenarios.

As shown in Figure 2, in the first aspect of the embodiment of this application, a lightweight urban spatiotemporal data prediction method is provided. The method may include:

Step S101: Construct a causal dilated convolution module, which is used to mine the time correlation of urban spatiotemporal data;

Step S102: Construct a graph dilation convolution module, which is used to mine the spatial correlation of urban spatiotemporal data;

Step S103: Construct a space-time graph expansion model based on the causal expansion convolution module and the graph expansion convolution module;

Step S104: Bring historical urban spatiotemporal data into the spatiotemporal graph expansion model to obtain prediction results of urban spatiotemporal data.

In this embodiment, the urban spatio-temporal data can be any of the following: traffic data, PM2.5 data and temperature data. The lightweight urban spatiotemporal data prediction method provided by this embodiment improves the prediction accuracy of urban spatiotemporal data by mining the time correlation and spatial correlation of urban spatiotemporal data. At the same time, the parameter scale is reduced by introducing expansion operations. The calculation speed is improved, which can simultaneously improve the prediction accuracy and ease of use of the prediction method.

Specifically, as shown in Figure 3, the STGDN model mainly consists of multiple serial spatio-temporal graph dilated convolution blocks (STGDNCell). Specifically, each STGDNCell contains a causal dilated convolution module (Causal Dilated ConvolutionModule, CDC) and a graph dilated convolution module (Graph Dilated Convolution Module, GDC), where the causal dilated convolution module is used to mine time in urban spatiotemporal data. Correlation,The graph dilated convolution module is used to mine,spatial correlations in urban sensor data. The STGDN model is mainly composed of multiple STGDNCells. Specifically, each STGDNCell contains a causal dilated convolution module and a graph dilated convolution module in series.

Existing time-correlation relationship mining models can be divided into iterative models and non-iterative models. The two types of models have advantages and disadvantages. For example, iterative models such as RNN have lower model parameters but slower computing efficiency, and non-iterative models such as convolutional neural networks have The model calculation efficiency of the iterative model is faster but the number of model parameters is larger. In order to simultaneously reduce the time complexity of the model (to improve the model's computing efficiency) and the space complexity (to reduce the number of model parameters), this application uses a novel lightweight causal dilated convolutional network to mine the time correlation in the data, namely STGDN. The causal dilated convolution module in .

The core of the causal dilation convolution module lies in the causal dilation convolution operator. Compared with the causal dilation convolution operator, the causal dilation convolution operator can greatly reduce the depth of the model, thereby reducing the number of parameters of the model. Take a time series containing 9 timestamps As an example, Figure 4 further illustrates the difference between causal convolution and causal dilated convolution. (a) in Figure 4 is used to characterize causal convolution, and (b) in Figure 4 is used to characterize causal dilation convolution. If using convolution kernel For time series/> To perform causal convolution operations, a 5-layer causal convolutional neural network is required. If using convolution kernel/> For time series/> To perform a causal dilated convolution operation with an expansion factor of d, only three layers of causal dilated convolutional neural networks are needed. Therefore, the causal dilated convolution operator reduces the number of parameters of the model by reducing the depth of the neural network. In time series/> For example, formula (2) further demonstrates the forward propagation process of the causal expansion convolution operator.

In the formula: Represents the time series/> in/> Perform causal expansion operation with expansion factor d;/> Represents the spatio-temporal data observed by graph node v _i in the time window τ;/> Represents the convolution kernel; K represents the size of the convolution kernel.

On the basis of defining the causal dilated convolution operator, this application further defines the forward propagation process of the causal dilated convolution module. In the causal dilated convolution module, residual linkage and parameter regularization are used to improve the prediction accuracy of the model. in time series For example, the forward propagation process of the causal dilated convolution module is shown in formula (3).

In the formula: Represents the forward propagation operation of the causal dilation convolution module on the time series;/> represents the intermediate variable of graph node v _i in the causal dilation convolution module; H _i represents the output variable of graph node v _i in the causal dilation convolution module; e _h represents the number of convolution kernels;/> The meaning is consistent with formula (2);/> Represents the time series/> Perform causal convolution operation, mainly used for residual link; Norm represents parameter regularization function; Relu represents activation function.

After all graph nodes pass through the causal expansion convolution module, a space-time tensor can be obtained In order to mine the spatial correlation relationships in spatiotemporal tensors, this application proposes a novel lightweight graph dilation convolution network, namely the graph dilation convolution module in STGDN. The core of the graph dilation convolution module lies in the graph dilation convolution operator. Similar to the causal dilation convolution operator, the graph dilation convolution operator can also significantly reduce the number of parameters of the model by reducing the depth of the model. As the expansion factor d increases, the graph dilation convolution operator can use fewer neural network layers to model long-range spatial dependencies.

The difficulty of the graph dilated convolution operator is to efficiently discover the dilated adjacency relationships of graph nodes. Figure 5 shows the expanded adjacency relationship of the target graph nodes. Specifically, when the expansion factor d is 1, the expansion adjacency relationship of the target graph node is equal to the first-order topological neighbor node. When the expansion factor d is 2, the expansion adjacency relationship of the target graph node is equal to the second-order topological neighbor node (the graph node that the target graph node can reach through the first-order topological neighbor node).

In order to efficiently obtain the expanded adjacency relationship of graph nodes, this application can obtain the expanded adjacency matrix A ^dilation=d with the expansion factor d through matrix multiplication of the first-order topological adjacency matrix A. Figure 6 shows the construction process of the expanded adjacency matrix. According to the first-order topological adjacency matrix A, this application can find that the first-order topological neighbor nodes of graph node v ₅ are v ₁ , v ₂ , and v ₃ . By further performing matrix multiplication on matrix A with an expansion factor of d=2, this application can find that the value equal to 1 in matrix A ² is the second-order topological adjacency of graph node v ₅ , that is, the nodes are v ₄ and v ₆ , _v7 . In the same way, the value equal to 1 in the matrix v ³ is the third-order topological adjacency of the graph node v ₅ , that is, the nodes are v ₈ and v _9. Based on this matrix property, the calculation method of the expanded adjacency matrix with the expansion factor d It can be defined by formula (4).

In the formula: A ^dilation=d represents the expanded adjacency matrix; Represents the expanded adjacency relationship between graph nodes v _i and v _j ; d represents the expansion factor. When d=1, the expanded adjacency matrix is the first-order topological adjacency matrix of the graph; A ^d represents the dth power of the first-order topological adjacency matrix.

After obtaining the dilation adjacency matrix, this application further defines the graph dilation convolution operator. The graph dilation convolution operator aggregates the information conveyed by spatial neighbors in a weighted manner. Specifically, the calculation method of the graph dilation convolution operator is shown in formula (5).

In the formula: Represents the graph expansion convolution operation with expansion factor d on graph node v _i ; γ _ji represents the influence weight of graph node v _j on graph node v _i ; The meaning of H _i is the same as formula (3);/> Represents the expansion adjacency matrix with expansion factor d; W _q and W _v represent learnable parameters; Relu represents the activation function; exp represents the exponential function; [·||·] represents the matrix connection function.

Based on the graph dilation convolution operator, this application further defines the forward propagation process of the graph dilation convolution module. Since the graph dilation convolution module only contains one graph dilation convolution operator, the forward propagation process of the graph dilation convolution operator is the forward propagation process of the graph dilation convolution module, as shown in formula (6).

In the formula: Represents a space-time tensor that is the output of the causal dilated convolution module.

Enter spatiotemporal data After the causal expansion convolution module and the graph expansion convolution module, the output of a single STGDNCell will be obtained, that is/> The output of the last STGDNCell can be used to obtain the final prediction result of the model through the convolution operation. hypothesis is the output result of the last STGDNCell, and the final prediction result is shown in formula (7).

In the formula: represents the final prediction result of the model; q represents the prediction step size of the model;/> Represents a space-time tensor/> Do the convolution operation.

The STGDN model passes the spatiotemporal data of the first p time windows Predict the spatiotemporal data of q time windows in the future/> In the process of model optimization, this application uses the mean square error to optimize the predicted value/> and true value/> between losses. Specifically, the loss function of the STGDN model is shown in formula (8).

In the formula: Represents the space-time state truth value (ground truth) of the t+jth time window; Represents the spatio-temporal state prediction value of the t+jth time window.

The STGDN model proposed in the lightweight urban spatiotemporal data prediction method provided in this embodiment is a lightweight spatiotemporal prediction model rather than a time series prediction model. The proposed STGDN model not only inherits the advantages of high prediction accuracy, fast operation speed, and small parameter scale of causal dilation convolution, but can also directly serve the spatiotemporal prediction task of urban sensor data. This application defines a novel graph dilated convolution (Graph Dilated Convolutional Operator) operator. The graph dilation convolution operator can effectively and quickly capture the spatial dependencies in urban sensor data without significantly increasing the model calculation time/model parameter scale. This application uses three real spatio-temporal data sets (traffic data set, PM2.5 data set, temperature data set) to evaluate the performance of the STGDN model, such as prediction accuracy, operating efficiency and model parameter quantity.

Since the prediction performance of knowledge-driven spatiotemporal prediction models is often lower than that of data-driven models, this application mainly compares the STGDN model with popular data-driven methods. There are 9 baseline methods used in this application, which can be roughly divided into two categories. The first category is machine learning models, including ST-KNN model and BTMF model. The second category is deep learning models, including T-GCN model, BiSTGN model, STGODE model, STA-ODE model, GDGCN model, ASTGCN model and DSTAGNN model. The prediction accuracy of the STGDN model and the baseline method is shown in the table below.

The results show that on the three data sets, the prediction accuracy of the second type of model is slightly higher than that of the first type of model, that is, the prediction performance of the deep learning model is slightly higher than the prediction performance of shallow machine learning. Specifically, on the traffic data set and PM2.5 data set, the prediction accuracy of the STGDN model is higher than the prediction accuracy of the ST-KNN, BTMF, T-GCN, BiSTGN, STGODE and ASTGCN models, and is close to STA-ODE and DSTAGNN. , the prediction accuracy of the GDGCN model. In terms of temperature data, the prediction accuracy of the STGDN model is higher than that of the ST-KNN, BTMF, T-GCN, BiSTGN, STGODE, ASTGCN and DSTAGNN models, and is close to the prediction accuracy of the STA-ODE and GDGCN models.

As shown in Figure 7, in the second aspect of the embodiment of the present application, a lightweight urban spatiotemporal data prediction device is provided. The sending device may include:

The first data processing module 11 is used to construct a causal dilated convolution module. The causal dilated convolution module is used to mine the time correlation of urban spatiotemporal data;

The second data processing module 12 is used to construct a graph expansion convolution module. The graph expansion convolution module is used to mine the spatial correlation of urban spatiotemporal data;

The third data processing module 13 is used to construct a space-time graph expansion model based on the causal expansion convolution module and the graph expansion convolution module;

The prediction module 14 is used to bring historical urban spatiotemporal data into the spatiotemporal graph expansion model to obtain prediction results of urban spatiotemporal data.

The lightweight urban spatiotemporal data prediction device in the embodiment of the present application may be a device, or may be a component, integrated circuit, or chip in a terminal. The device may be a mobile electronic device or a non-mobile electronic device. For example, the mobile electronic device may be a mobile phone, a tablet computer, a notebook computer, a handheld computer, a vehicle-mounted electronic device, a wearable device, an ultra-mobile personal computer (UMPC), a netbook or a personal digital assistant (personal digital assistant). assistant, PDA), etc., and the non-mobile electronic device can be a server, Network Attached Storage (NAS), personal computer (PC), television (television, TV), teller machine or self-service machine, etc., this application The examples are not specifically limited.

The lightweight urban spatiotemporal data prediction device provided by the embodiments of the present application can implement the lightweight urban spatiotemporal data prediction method provided by the above embodiments. To avoid duplication, the details will not be described here.

Optionally, as shown in Figure 8, this embodiment of the present application also provides an electronic device 1100, including a processor 1101, a memory 1102, and programs or instructions stored on the memory 1102 and executable on the processor 1101. When this program or instruction is executed by the processor 1101, each process of the above lightweight urban spatiotemporal data prediction method or data processing method embodiment is implemented, and the same technical effect can be achieved. To avoid duplication, it will not be described again here.

It should be noted that the electronic devices in the embodiments of the present application include the above-mentioned mobile electronic devices and non-mobile electronic devices.

FIG. 9 is a schematic diagram of the hardware structure of an electronic device implementing an embodiment of the present application.

The electronic device 1200 includes but is not limited to: radio frequency unit 1201, network module 1202, audio output unit 1203, input unit 1204, sensor 1205, display unit 1206, user input unit 1207, interface unit 1208, memory 1209, processor 1210, etc. part.

Those skilled in the art can understand that the electronic device 1200 may also include a power supply (such as a battery) that supplies power to various components. The power supply may be logically connected to the processor 1210 through a power management system, thereby managing charging, discharging, and function through the power management system. Consumption management and other functions. The structure of the electronic device shown in Figure 9 does not constitute a limitation on the electronic device. The electronic device may include more or less components than shown in the figure, or combine certain components, or arrange different components, which will not be described again here. .

It should be understood that in the embodiment of the present application, the input unit 1204 may include a graphics processor (Graphics Processing Unit, GPU) 12041 and a microphone 12042. The graphics processor 12041 is responsible for the image capture device (such as Process the image data of still pictures or videos obtained by the camera). The display unit 1206 may include a display panel 12061, which may be configured in the form of a liquid crystal display, an organic light emitting diode, or the like. The user input unit 1207 includes a touch panel 12071 and other input devices 12072. Touch panel 12071, also known as touch screen. The touch panel 12071 may include two parts: a touch detection device and a touch controller. Other input devices 12072 may include but are not limited to physical keyboards, function keys (such as volume control keys, switch keys, etc.), trackballs, mice, and joysticks, which will not be described again here. Memory 1209 may be used to store software programs as well as various data, including but not limited to application programs and operating systems. The processor 1210 can integrate an application processor and a modem processor. The application processor mainly processes the operating system, user interface, application programs, etc., and the modem processor mainly processes wireless communications. It can be understood that the above modem processor may not be integrated into the processor 1210.

Embodiments of the present application also provide a readable storage medium. Programs or instructions are stored on the readable storage medium. When the program or instructions are executed by a processor, the above lightweight urban spatiotemporal data prediction method is implemented and can achieve The same technical effects will not be repeated here to avoid repetition.

Wherein, the processor is the processor in the electronic device described in the above embodiment. The readable storage medium includes computer readable storage media, such as computer read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disk, etc.

An embodiment of the present application further provides a chip. The chip includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to realize the above-mentioned lightweight urban space-time. Each process of the embodiment of the data prediction method can achieve the same technical effect. To avoid duplication, it will not be described again here.

It should be understood that the chips mentioned in the embodiments of this application may also be called system-on-chip, system-on-a-chip, system-on-a-chip or system-on-a-chip.

It should be noted that, in this document, the terms "comprising", "comprises" or any other variation thereof are intended to cover a non-exclusive inclusion, such that a process, method, article or device that includes a series of elements not only includes those elements, It also includes other elements not expressly listed or inherent in the process, method, article or apparatus. Without further limitation, an element defined by the statement "comprises a..." does not exclude the presence of additional identical elements in a process, method, article, or device that includes that element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, but may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved. Functions may be performed, for example, the methods described may be performed in an order different from that described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus the necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is better. implementation. Based on this understanding, the technical solution of the present application can be embodied in the form of a computer software product that is essentially or contributes to the existing technology. The computer software product is stored in a storage medium (such as ROM/RAM, disk , optical disk), including several instructions to cause a terminal (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in various embodiments of this application.

The embodiments of the present application have been described above in conjunction with the accompanying drawings. However, the present application is not limited to the above-mentioned specific implementations. The above-mentioned specific implementations are only illustrative and not restrictive. Those of ordinary skill in the art will Inspired by this application, many forms can be made without departing from the purpose of this application and the scope protected by the claims, all of which fall within the protection of this application.

Claims (13)

1. A lightweight urban spatiotemporal data prediction method, which is characterized by: Construct a causal dilated convolution module, which is used to mine the time correlation of urban spatiotemporal data; Construct a graph dilation convolution module, which is used to mine the spatial correlation of the city's spatiotemporal data; Construct a space-time graph expansion model based on the causal expansion convolution module and the graph expansion convolution module; The historical urban spatiotemporal data is brought into the spatiotemporal graph expansion model to obtain the prediction results of the urban spatiotemporal data. 2. A lightweight urban spatiotemporal data prediction method according to claim 1, characterized in that constructing a causal dilated convolution module includes: Construct causal dilation convolution operator; The causal dilation convolution module is obtained by adding a residual link and a regularization module to the causal dilation convolution operator. 3. A lightweight urban spatiotemporal data prediction method according to claim 2, characterized in that the forward propagation formula of the causal expansion convolution operator is as follows: in, For the time series of urban spatiotemporal data/> in/> Perform causal expansion operation with expansion factor d;/> is the urban spatio-temporal data observed by urban sensor node v _i in the time window τ;/> is the convolution kernel; K is the size of the convolution kernel. 4. A lightweight urban spatiotemporal data prediction method according to claim 3, characterized in that the forward propagation formula of the causal dilated convolution module is as follows: in, To perform the forward propagation operation of the causal dilated convolution module on the time series of urban spatiotemporal data; H′ _i , /> is the intermediate variable of city sensor node v _i in the causal dilation convolution module; H _i is the output variable of city sensor node v _i in the causal dilation convolution module; e _h is the number of convolution kernels; For the time series of urban spatiotemporal data/> Perform causal convolution operation; Norm is the parameter regularization function; Relu is the activation function. 5. A lightweight urban spatiotemporal data prediction method according to claim 1, characterized in that constructing a graph dilation convolution module includes: Construct an expanded adjacency matrix of urban sensor nodes; Construct a graph dilated convolution operator based on the dilated adjacency matrix; A graph dilation convolution module is constructed based on the graph dilation convolution operator. 6. A lightweight urban spatiotemporal data prediction method according to claim 5, characterized in that the calculation formula of the expanded adjacency matrix is as follows: Among them, A ^dilation=d is the expanded adjacency matrix; is the expansion adjacency relationship between urban sensor nodes v _i and v _j ; d is the expansion factor; A ^d is the dth power of the first-order topological adjacency matrix. 7. A lightweight urban spatiotemporal data prediction method according to claim 6, characterized in that the calculation formula of the graph dilation convolution operator is as follows: in, is the graph expansion convolution operation with the expansion factor d for the city sensor node v _i ; γ _ji is the influence weight of the sensor node v _j on the sensor node _vi ; H _i is the city sensor node v _i in the causal expansion convolution module The output variable;/> is the expansion adjacency matrix with expansion factor d; W _q and W _v are learnable parameters; Relu is the activation function; exp is the exponential function; [·||·] is the matrix connection function. 8. A lightweight urban spatiotemporal data prediction method according to claim 7, characterized in that the forward propagation formula of the graph dilation convolution module is as follows: in, is a space-time tensor, which is the output of the causal dilated convolution module. 9. A lightweight urban spatiotemporal data prediction method according to any one of claims 1 to 8, characterized in that a spatiotemporal graph expansion is constructed based on the causal expansion convolution module and the graph expansion convolution module. Models, including: Construct a spatiotemporal graph dilation convolution block, which includes the causal dilation convolution module and the graph dilation convolution module in series; Concatenate N of the spatio-temporal graph dilation convolution blocks to obtain the spatio-temporal graph dilation model; Among them, N is a natural number greater than 1. 10. A lightweight urban spatiotemporal data prediction method according to claim 9, characterized in that the calculation formula of the spatiotemporal graph expansion model is as follows: in, For historical city spatiotemporal data,/> is the output of the Nth space-time graph dilation convolution block;/> is the prediction result of urban spatiotemporal data; q is the prediction step length of the model;/> is the space-time tensor/> Do the convolution operation. 11. A lightweight urban spatiotemporal data prediction device, which is characterized by including: The first data processing module is used to construct a causal dilated convolution module, which is used to mine the time correlation of urban spatiotemporal data; The second data processing module is used to construct a graph expansion convolution module. The graph expansion convolution module is used to mine the spatial correlation of the city's spatiotemporal data; A third data processing module, configured to construct a spatiotemporal graph expansion model based on the causal expansion convolution module and the graph expansion convolution module; A prediction module is used to bring historical urban spatiotemporal data into the spatiotemporal graph expansion model to obtain prediction results of urban spatiotemporal data. 12. An electronic device, characterized in that it includes: a processor, a memory, and a program or instruction stored on the memory and executable on the processor. When the program or instruction is executed by the processor Implement the lightweight urban spatiotemporal data prediction method as described in any one of claims 1-10. 13. A readable storage medium, characterized in that the readable storage medium stores programs or instructions, and when the programs or instructions are executed by a processor, the lightweight process as claimed in any one of claims 1 to 10 is achieved. Level urban spatiotemporal data prediction method.