Motor vehicle emission model method and system based on space-time correlation analysis and deep learning

Abstract

The invention provides a motor vehicle emission model method and system based on space-time correlation analysis and deep learning. On the basis of multi-source heterogeneous traffic big data, a geographically weighted regression model, a spatio-temporal geographically weighted regression model and the like are utilized to perform spatio-temporal analysis and measurement on numerous external and internal influence factors influencing motor vehicle emission; a new graph structure is constructed by key factors, dominant factors, sensitive factors and the like obtained after space-time analysis; on the basis of the new graph structure, a space-time graph neural network (STGCN) combining a gated loop unit (GRU) and a graph convolutional network (GCN) is built; a high-precision and localized motor vehicle emission model coupled with a vehicle dynamics principle is constructed based on deep learning, a motor vehicle emission rule is disclosed, multidisciplinary cross fusion is promoted, and support is provided for green low-carbon traffic.

Description

Technical Field

[0001] The present invention relates to the field of transportation and energy fusion technology, and in particular to a motor vehicle emission model method and system based on spatiotemporal correlation analysis and deep learning. Background Art

[0002] The statements in this section merely mention background art related to the present invention and do not necessarily constitute prior art.

[0003] High-precision motor vehicle emission models are the basis for studying motor vehicle emission issues. Classic motor vehicle emission models consider the principles of vehicle dynamics and are built based on a large amount of test data. They can be roughly divided into two categories. One category is models with average speed as the characterization parameter, such as MOBILE, EMFAC, TREMOD, COPERT, etc., which are suitable for macro and meso scales such as countries, regions, and cities; the other category is models based on motor vehicle driving characteristics (instantaneous working conditions). Such models introduce speed-acceleration and vehicle specific power (VSP). Representative models include MOVES, CMEM, IVE, VT-Micro, etc., which can be used at the micro and meso levels. At present, my country has not yet established a sound database of motor vehicle emission factors, and most of them use more mature foreign models for localization. The Ministry of Ecology and Environment took the lead in developing CVEM suitable for macro and meso scales. However, the test data of European and American models come from European and American vehicles. There are differences in the emission level and environment of my country's fleet, and the lack of localized emission data makes it difficult to calibrate and verify the model.

[0004] With the explosive growth of data and the rapid development of artificial intelligence technology, the application of deep learning in motor vehicle emission models has also become a hot topic. Domestic and foreign scholars have built emission models for different models and scenarios based on support vector regression (SVR), BP (BackPropagation, BP) neural network, multi-layer perceptron (MLP), artificial neural network (ANN), feedforward neural network (FNN), recurrent neural network (RNN), long short-term memory neural network (LSTM), etc., and the accuracy of the model has also been verified. The data-driven emission model built based on deep learning has a strong dependence on data and is mostly built based on a relatively single scenario and fixed model. How to couple the principles of vehicle dynamics and enhance the mechanism and portability of data-driven emission models is the direction of development.

[0005] Both the classic motor vehicle emission model and the data-driven vehicle emission model based on deep learning have the problem of insufficient correlation analysis and measurement of influencing factors. Although the classic motor vehicle emission model is based on measured data and has strong mechanistic rationality, it is difficult to fully consider the influencing factors due to the large number of influencing factors and the complexity of the model. The data-driven vehicle emission model relies on input data, which are basically influencing factors. Due to the incomplete coverage of the constructed scenarios and vehicle models, there are fewer influencing factors, resulting in insufficient correlation analysis between the influencing factors. Summary of the invention

[0006] The present invention provides a motor vehicle emission model method and system based on spatiotemporal correlation analysis and deep learning. Based on multi-source heterogeneous traffic big data, the geographical weighted regression model, spatiotemporal geographical weighted regression model, etc. are used to perform spatiotemporal analysis and measurement of many external and internal influencing factors affecting motor vehicle emissions; and a new graph structure is constructed based on the key factors, dominant factors, sensitive factors, etc. obtained after spatiotemporal analysis; based on the new graph structure, a spatiotemporal graph neural network (STGCN) network combining a gated recurrent unit (GRU) and a graph convolutional network (GCN) is constructed; based on deep learning, a high-precision, localized motor vehicle emission model coupled with the principle of vehicle dynamics is constructed to reveal the laws of motor vehicle emissions, promote the cross-integration of multiple disciplines, and provide support for green and low-carbon transportation.

[0007] Terminology explanation: Gated recurrent unit (GRU): GRU is a type of recurrent neural network (RNN). It can solve the problems of long-term memory and gradient in back propagation in RNN. It is a simplification of LSTM (Long-Short Term Memory). The output gate and forget gate of LSTM are coupled as update gates, and the reset gate corresponds to the input gate of LSTM. This retains the ability of LSTM network to maintain memory state and selectively remember or forget information, while reducing the number of parameters, greatly shortening the training time, making it easy to train and suitable for processing time series data.

[0008] Graph Convolutional Network (GCN): GCN draws on the ideas of convolutional networks, recurrent networks, and deep autoencoders. It is based on the graph structure in traditional graph theory and is used to process unstructured data. It is a framework that uses deep learning to directly learn graph structured data.

[0009] In a first aspect, the present invention provides a motor vehicle emission model method based on spatiotemporal correlation analysis and deep learning; A motor vehicle emission model method based on spatiotemporal correlation analysis and deep learning, comprising: Obtain data on factors affecting motor vehicle emissions in the study area, perform map matching and preprocessing on the data, and build a multi-source heterogeneous traffic spatiotemporal database; Use geographically weighted regression models and other methods to analyze and measure the spatiotemporal correlation of various factors affecting motor vehicle emissions; A new graph structure is constructed based on the key factors, dominant factors, sensitive factors, etc. obtained after spatiotemporal analysis, and a spatiotemporal graph neural network (STGCN) network combining GRU and GCN is built based on the newly constructed graph structure; A high-precision, localized motor vehicle emissions model coupled with vehicle dynamics principles is constructed based on deep learning.

[0010] In the second aspect, a motor vehicle emission model method system based on spatiotemporal correlation analysis and deep learning is provided; A motor vehicle emission model system based on spatiotemporal correlation analysis and deep learning, comprising: Data acquisition and processing module: obtain the data on the influencing factors of motor vehicle emissions in the study area, perform map matching and preprocessing on the data, and build a multi-source heterogeneous traffic spatiotemporal database; Data analysis module: Use geographically weighted regression models to analyze and measure the spatiotemporal correlation of various factors affecting motor vehicle emissions; Graph construction module: construct a new graph structure based on the key factors, dominant factors, sensitive factors, etc. obtained after spatiotemporal analysis, and build a spatiotemporal graph neural network (STGCN) network combining GRU and GCN based on the newly constructed graph structure; Model building module: Based on deep learning, a high-precision, localized motor vehicle emission model is built that couples vehicle dynamics principles.

[0011] In a third aspect, the present invention further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the first aspect when executing the program.

[0012] In a fourth aspect, the present invention further provides a storage medium that non-temporarily stores computer-readable instructions, wherein when the non-temporary computer-readable instructions are executed by a computer, the instructions of the method described in the first aspect are executed.

[0013] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a motor vehicle emission model method and system based on spatiotemporal correlation analysis and deep learning. Based on multi-source heterogeneous traffic big data, the present invention uses geographical weighted regression models, spatiotemporal geographical weighted regression models, etc. to perform spatiotemporal analysis and measurement of external and internal factors affecting motor vehicle emissions.

[0014] The present invention constructs a new graph structure and network structure. The new graph structure is constructed based on the key factors, dominant factors, sensitive factors, etc. obtained after spatiotemporal analysis, and based on the new graph structure, a spatiotemporal graph neural network (STGCN) network combining a gated recurrent unit (GRU) and a graph convolutional network (GCN) is constructed.

[0015] Based on deep learning, the present invention constructs a high-precision, localized motor vehicle emission model that couples the principles of vehicle dynamics. It gives full play to the powerful nonlinear fitting ability of deep learning and the advantages of traditional emission models that consider the principles of vehicle dynamics, providing support for green and low-carbon transportation. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0017] Figure 1 is a flow chart of the method of embodiment 1; Figure 2 A schematic diagram of a newly constructed graph structure of Example 1 Figure 3 Schematic diagram of the newly constructed graph node matrix of Example 1 Figure 4 This is a GRU network diagram of Example 1 Figure 5 Schematic diagram of the newly constructed STGCN network in Example 1 Implementation

[0018] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0019] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0020] In this embodiment, all data is obtained based on the legal application of data in compliance with laws and regulations.

[0021] Embodiment 1 This embodiment provides a motor vehicle emission model method based on spatiotemporal correlation analysis and deep learning, such as Figure 1 As shown in Figure 1, a motor vehicle emission model method based on spatiotemporal correlation analysis and deep learning is proposed. include: S101: Obtain the data on the influencing factors of motor vehicle emissions in the study area, perform map matching and preprocessing on the data, and build a multi-source heterogeneous traffic spatiotemporal database; S102: Use geographically weighted regression models to analyze and measure the spatiotemporal correlation of various factors affecting motor vehicle emissions; S103: Construct a new graph structure based on the key factors, dominant factors, sensitive factors, etc. obtained after spatiotemporal analysis, and build a spatiotemporal graph neural network (STGCN) network combining GRU and GCN based on the newly constructed graph structure; S104: Build a high-precision, localized motor vehicle emission model based on deep learning coupled with vehicle dynamics principles.

[0022] Furthermore, the S101: obtaining the data of influencing factors of motor vehicle emissions in the study area, performing map matching and preprocessing on the data, and constructing a multi-source heterogeneous traffic spatiotemporal database, specifically includes: S101-1: Obtain data on factors affecting motor vehicle emissions in the study area, including natural economic data such as temperature, humidity, slope, weather, GDP, population density, land use type, roads, POI points, motor vehicle ownership, vehicle model, mileage, vehicle age distribution, fuel type, etc. at different times; traffic condition data such as GPS trajectory, mobile phone signaling, acceleration, average speed, number of stops, delays, etc.; driver data such as driver composition and driving habits; vehicle dynamics data such as degradation coefficient, engine displacement, aerodynamic drag coefficient, etc.; second-by-second motor vehicle emission data (CO2, CO, HC, NO2, PM 2.5 )wait; S101-2: Match natural economic data, traffic condition data, driver data, vehicle dynamics data, and emission data to a unified high-precision map; S101-3: Based on accurate map matching, clean the multi-source heterogeneous data to remove errors and redundant data; and convert and standardize the cleaned data; S101-4: Import the standardized data into the database to form a corresponding data sequence, as shown in Table 1, and construct a multi-source heterogeneous traffic spatiotemporal database.

[0023] Table 1 Vehicle emission data series ID time Model Vehicle weight Vehicle age Fuel Engine displacement slope temperature humidity GDP speed Acceleration Delays <![CDATA[CO2]]> <![CDATA[NO2]]> <![CDATA[PM 2.5 ]]> <![CDATA[PM 10 ]]> … Furthermore, the S101-2: matching the natural economy data, traffic condition data, driver data, vehicle dynamics data, and emission data to a unified high-precision map specifically includes: S101-21: Project socio-economic and physical geographic data with coordinate systems, such as land use, POI points, GDP, road network, slope, etc., into the same coordinate system; S101-22: Obtain the longitude and latitude information of the location of the traffic facility data such as road zoning information, signal control information, text, and picture information according to the address matching function, and match them on the map; S101-23: Match the trajectory, mobile phone signaling and other data to the map using the map matching model of the "distance-direction-connectivity" criterion; the matching criteria are as follows: (a) distance criterion: the distance from the target location to the matched section is the shortest; (b) direction criterion: the current moving direction of the target is closest to the direction of the matched section; (c) connectivity criterion: if the matching result at the previous moment is on a certain section, the matching result at the current moment is still on this section or on a section adjacent to this section; after completing the map matching, if two adjacent positioning points are on the same section, the driving path is along this section, otherwise, the Dijkstra algorithm is used to obtain the shortest path between two points as the driving path based on the topological information of the road network; S101-24: Match non-spatial data such as fuel type, oil quality, vehicle ownership, vehicle model, vehicle age, vehicle weight, and mileage with trajectory data; S101-25: Based on map matching, various types of acquired data are associated.

[0024] Furthermore, the S102: using a geographically weighted regression model or the like to perform spatiotemporal correlation analysis and measurement on various factors affecting motor vehicle emissions, specifically includes: S102-1: Analyze and measure the spatial heterogeneity of factors affecting vehicle emissions using geographically weighted regression (GWR); (1) In the formula, is the regional sample point The spatial geographic coordinates of For sample points On regression parameters, and spatial position Related, is the error term of the sample point; S102-2: Analyze and measure the spatiotemporal non-stationarity of the factors affecting vehicle emissions using the spatiotemporal geographically weighted regression model GTWR; (2) In the formula, For the The space-time coordinates of the sample points, It is The regression coefficient of the independent variable is a function of the spatial and temporal position. is the error term of the sample point; S102-3: Use SGWR to analyze and measure the impact scale of factors affecting vehicle emissions; (3) In the formula, is the regression coefficient of the global variable, is the number of independent variables, is the regression parameter of global regression, is the number of local variables, is the error of the sample point; S102-4: Use MGWR to analyze and measure the impact of different factors affecting vehicle emissions; (4) In the formula, For the The bandwidth of the regression coefficient of each variable, express Place The regression coefficients of the variables; S102-5: Based on the results of spatiotemporal analysis, obtain the key factors, sensitive factors, dominant factors, threshold factors, correction factors, and measurement factors for correlation factor analysis that affect vehicle emissions by vehicle type and emission substances (CO2, CO, HC, NO2, PM2.5).

[0025] Further, S103: construct a new graph structure based on the key factors, dominant factors, sensitive factors, correction factors, etc. obtained after spatiotemporal analysis, and build a spatiotemporal graph neural network (STGCN) network combining GRU and GCN based on the newly constructed graph structure, specifically including: S103-1: Constructing a new graph structure , where the node The key factors, sensitive factors, dominant factors, threshold factors, correction factors and other influencing factors (such as speed, acceleration, vehicle model, etc.) that affect vehicle emissions are nodes. ,side is the correlation between nodes (influencing factors), edge weight is the measurement data of the correlation factor, that is, the measurement factor, as follows Figure 2 (Sample data); Edge weights is the relationship between the influencing factors, that is, the correlation data measured by the geographically weighted regression model, that is, the measurement factor data, and its adjacency matrix is ​​as follows Figure 3 (sample data); S103-2: constructing a motor vehicle emission function based on the newly constructed graph structure; CO2, HC, CO, NO emitted by vehicles at all times x 、PM 2.5The emission can be expressed as an N-dimensional vector, where N is the node data. The emission X at a certain moment can be expressed as: Based on the graph, we learn information samples to obtain a function that reflects future information. : (5) S103-3: Based on the newly constructed graph structure, construct the STGCN network based on GRU+GCN; regard the vehicle emission data as a time series. CO2, HC, CO, NO at all times x 、PM 2.5 Emissions are taken as a function of the influencing factors at time t, as shown in Formula 5. A GRU network suitable for processing time series data is selected. The internal structure of GRU is as follows: Figure 4 ; The GCN variant composed of the constructed graph structure replaces the fully connected layer inside the GRU, so that the GRU can process emission-related data; the full connection inside the GRU is replaced, the GCN and GRU are not simply stacked, and the matrix multiplication in the GRU is replaced by convolution, and the STGNN network is proposed, such as Figure 5 .

[0026] Further, S104: constructing a high-precision, localized motor vehicle emission model based on deep learning coupled with vehicle dynamics principles, specifically including: S104-1: Preprocess the key influencing factors, dominant factors, measurement factors, and PEMS measurement data such as speed, acceleration, temperature and humidity according to the created graph structure format to form a data set; normalize the emission data as label data; S104-2: Divide the data set into training set, validation set and test set in a ratio of 6:2:2, and further divide the divided training set to perform 4-cross validation; S104-3: Set relevant parameters, including the maximum number of iterations, learning rate, and learning rate decay coefficient. Use the learning rate hot start + cosine annealing learning rate strategy. Use Adam as the optimizer. Add the L2 regularization coefficient as a penalty term to prevent network overfitting. Use MSE Loss as the loss function to supervise network learning. S104-4: Iterative training. After each epoch of training, the validation set and test set will be used to calculate the accuracy. The network structure and parameters with the highest accuracy in the training process will be selected as the emission model. S104-5: Repeat the above steps to construct a motor vehicle emission model based on vehicle type and pollutant.

[0027] Embodiment 2 This embodiment provides a motor vehicle emission model method system based on spatiotemporal correlation analysis and deep learning; A motor vehicle emission model system based on spatiotemporal correlation analysis and deep learning, comprising: The data acquisition and processing module is configured as follows: obtaining the data on the influencing factors of motor vehicle emissions in the study area, performing map matching and preprocessing on the data, and constructing a multi-source heterogeneous traffic spatiotemporal database; The data analysis module is configured to: use a geographically weighted regression model and other methods to perform spatiotemporal correlation analysis and measurement of various factors affecting motor vehicle emissions; The graph construction module is configured as follows: a new graph structure is constructed based on the key factors, dominant factors, sensitive factors, etc. obtained after spatiotemporal analysis, and a spatiotemporal graph neural network (STGCN) network combining GRU and GCN is built based on the newly constructed graph structure; A model building module is configured to build a high-precision, localized motor vehicle emission model based on deep learning coupled vehicle dynamics principles.

[0028] It should be noted here that the above-mentioned data acquisition and processing module, quantity analysis module, graph construction module and model construction module correspond to steps S101 to S104 in the first embodiment, and the examples and application scenarios implemented by the above-mentioned modules and the corresponding steps are the same, but are not limited to the contents disclosed in the first embodiment. It should be noted that the above-mentioned modules, as part of the system, can execute the proposed system in a computer system such as a set of computer executable instructions, and can be implemented in other ways; for example, the system embodiment described above is only schematic, and for example, the division of the above-mentioned modules is only a logical function division. In actual implementation, there may be other division methods, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0029] Embodiment 3 This embodiment also provides an electronic device, including: one or more processors, one or more memories, and one or more computer programs; wherein the processor is connected to the memory, and the one or more computer programs are stored in the memory. When the electronic device is running, the processor executes the one or more computer programs stored in the memory so that the electronic device executes the method described in the above embodiment one.

[0030] The method in the first embodiment can be performed by combining hardware and software modules in the processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it will not be described in detail here.

[0031] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0032] Embodiment 4 This embodiment also provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, a motor vehicle emission model method based on spatiotemporal correlation analysis and deep learning as described in Embodiment 1 is completed.

[0033] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A motor vehicle emission model method based on spatiotemporal correlation analysis and deep learning, Its characteristics include: obtaining the data of influencing factors of motor vehicle emissions in the study area, performing map matching and preprocessing on the data, and constructing a multi-source heterogeneous traffic spatiotemporal database; using geographically weighted regression models and other methods to perform spatiotemporal correlation analysis and measurement on various factors affecting motor vehicle emissions; constructing a new graph structure based on the key factors, dominant factors, sensitive factors, etc. obtained after spatiotemporal analysis, and building a spatiotemporal graph neural network STGCN network combining GRU and GCN based on the newly constructed graph structure; A high-precision, localized motor vehicle emissions model coupled with vehicle dynamics principles is constructed based on deep learning.

2. The motor vehicle emission model method based on spatiotemporal correlation analysis and deep learning as claimed in claim 1, characterized in that: Obtain data on factors affecting motor vehicle emissions in the study area, map match and preprocess the data, and build a multi-source heterogeneous traffic spatiotemporal database, including: obtaining data on factors affecting motor vehicle emissions in the study area, including natural economic data such as temperature, humidity, slope, weather, GDP, population density, land use type, roads, POI points, motor vehicle ownership, vehicle model, mileage, vehicle age distribution, fuel type, etc. at different times; traffic condition data such as GPS trajectory, mobile phone signaling, acceleration, average speed, number of stops, delays, etc.; driver data such as driver composition and driving habits; vehicle dynamics data such as degradation coefficient, engine displacement, and aerodynamic drag coefficient; second-by-second motor vehicle emission data (CO2, CO, HC, NO2, PM 2.5 ) etc.; match natural economic data, traffic condition data, driver data, vehicle dynamics data, and emission data to a unified high-precision map; on the basis of accurate map matching, clean the multi-source heterogeneous data to remove errors and redundant data; convert the format of the cleaned data and standardize it; import the standardized data into the database to form a corresponding data sequence, and build a multi-source heterogeneous traffic spatiotemporal database.

3. The method of obtaining the influencing factor data of motor vehicle emissions in the study area as claimed in claim 2, performing map matching and preprocessing on the data, and constructing a multi-source heterogeneous traffic spatiotemporal database, is characterized in that: The natural economy data, traffic condition data, driver data, vehicle dynamics data, and emission data are all matched to a unified high-precision map. Specifically, the socio-economic and natural geographic data with coordinate systems, such as land use, POI points, GDP, road network, and slope, are uniformly projected to the same coordinate system; the longitude and latitude information of the location of traffic facility data such as road zoning information, signal control information, text, and image information is obtained based on the address matching function, and matched to the map; the trajectory, mobile phone signaling and other data are matched to the map using the map matching model based on the "distance-direction-connectivity" criterion; the matching criteria are as follows: (a) Distance criterion: the distance from the target location to the matched road The distance between the two segments is the shortest; (b) Direction criterion: the current moving direction of the target is closest to the direction of the matched segment; (c) Connectivity criterion: if the matching result at the previous moment is on a certain segment, the matching result at the current moment is still on this segment or on a segment adjacent to this segment; after completing the map matching, if two adjacent positioning points are on the same segment, the driving path is along this segment, otherwise, the Dijkstra algorithm is used to obtain the shortest path between the two points as the driving path based on the topological information of the road network; match non-spatial data such as fuel type, oil quality, vehicle ownership, vehicle model, vehicle age, vehicle weight, mileage, etc. with the trajectory data; on the basis of map matching, associate the various types of data obtained.

4. The motor vehicle emission model method based on spatiotemporal correlation analysis and deep learning as claimed in claim 1, characterized in that: The geographically weighted regression model is used to conduct spatiotemporal correlation analysis and measurement of various factors affecting motor vehicle emissions, including: using geographically weighted regression (GWR) to analyze and measure the spatial heterogeneity of factors affecting vehicle emissions; using the spatiotemporal geographically weighted regression model GTWR to analyze and measure the spatiotemporal non-stationarity of factors affecting vehicle emissions; using SGWR to analyze and measure the impact scale of factors affecting vehicle emissions; using MGWR to analyze and measure the impact range of different factors affecting vehicle emissions; obtaining the key factors, sensitive factors, dominant factors, threshold factors, correction factors, and measurement factors for correlation factor analysis that affect vehicle emissions by vehicle type and emission substances (CO2, CO, HC, NO2, PM2.5).

5. The motor vehicle emission model method based on spatiotemporal correlation analysis and deep learning as claimed in claim 1, characterized in that: A new graph structure is constructed based on the key factors, dominant factors, sensitive factors, correction factors, etc. obtained after spatiotemporal analysis, and a spatiotemporal graph neural network (STGC) network combining GRU and GCN is built based on the newly constructed graph structure, specifically including: constructing a new graph structure G = (V, E, W), where node v is the key factor, sensitive factor, dominant factor, threshold factor, correction factor and other influencing factors (such as speed, acceleration, vehicle model, etc.) affecting vehicle emissions as node V, edge E is the correlation between nodes, edge weight W is the measurement data of the correlation factor, that is, the measurement factor, and the edge weight W (i, j) is the relationship between the influencing factors, that is, the correlation relationship data measured by the geographical weighted regression model, that is, the measurement factor data; on the basis of the newly constructed graph structure, a motor vehicle emission function is constructed; the CO2, HC, CO, NO emitted by the vehicle at time t x 、PM 2.5 The emission can be expressed as an N-dimensional vector, where N is the node data. The emission X at time t+h can be expressed as: Based on the graph, we learn information samples to obtain a function f that reflects future information: X t+h =f(G r ,G w ;(x t-W-1 ,…,x t-1 ,x t )) Based on the newly constructed graph structure, a STGCN network based on GRU+GCN is constructed; the vehicle emission data is regarded as a time series, and the CO2, HC, CO, and NO of the vehicle at time t+h are x 、PM 2.5 Emissions are taken as a function of the influencing factors at time t, and a GRU network suitable for processing time series data is selected; the GCN variant composed of the constructed graph structure replaces the fully connected layer inside the GRU, so that the GRU can process emission-related data; the full connection inside the GRU is replaced, rather than simply stacking GCN and GRU, and convolution is used instead of matrix multiplication in the GRU, and the STGNN network is proposed.

6. The motor vehicle emission model method based on spatiotemporal correlation analysis and deep learning as claimed in claim 1, characterized in that: Based on deep learning, a high-precision, localized motor vehicle emission model is constructed based on the principle of coupled vehicle dynamics, including: pre-processing the key influencing factors such as speed, acceleration, temperature and humidity, dominant factors, measurement factors, and PEMS measurement data according to the created graph structure format to form a data set; normalizing the emission data as label data; dividing the data set into training set, validation set and test set in a ratio of 6:2:2, and further dividing the divided training set for 4 cross-validation; setting relevant parameters, including the maximum number of iterations, learning rate, and learning rate decay coefficient, using the learning rate hot start + cosine annealing learning rate strategy; using Adam as the optimizer, adding the L2 regularization coefficient as a penalty term to prevent network overfitting; using MSE Loss as the loss function to supervise network learning; iterative training, using the validation set and test set for accuracy calculation after each epoch of training, and selecting the network structure and parameters with the highest accuracy in the training process as the emission model; repeating the above steps to construct a motor vehicle emission model for different vehicle types and pollutants.

7. A motor vehicle emission model system based on spatiotemporal correlation analysis and deep learning, characterized in that: include: Data acquisition and processing module: obtain the data on the influencing factors of motor vehicle emissions in the study area, map match and preprocess the data, and build a multi-source heterogeneous traffic spatiotemporal database; data analysis module: use geographically weighted regression models and other methods to conduct spatiotemporal correlation analysis and measurement of various factors affecting motor vehicle emissions; graph construction module: construct a new graph structure based on the key factors, dominant factors, sensitive factors, etc. obtained after spatiotemporal analysis, and build a spatiotemporal graph neural network (STGCN) network combining GRU and GCN based on the newly constructed graph structure; model construction module: build a high-precision, localized motor vehicle emission model based on deep learning coupled vehicle dynamics principles.

8. An electronic device, characterized in that: include: A memory for non-temporarily storing computer-readable instructions; a computer program on the memory and executable by the processor; wherein the computer-readable instructions and the computer program, when executed by the processor, execute the method described in any one of claims 1 to 7.

9. A storage medium, characterized in that: Non-temporarily storing data and a computer program, which, when executed by a processor, executes the instructions of the method according to any one of claims 1 to 7.

Images