A method and device for predicting and visualizing carbon emissions from motor vehicles on urban streets

Abstract

The present invention discloses a method for predicting and visualizing carbon emissions from motor vehicles on urban streets, comprising using a CNN semantic segmentation model to extract features from urban streetscape images and encode them into feature vectors; converting the urban road network structure into a graph structure; constructing a graph neural network model, matching the GPS trajectory data of motor vehicles on urban streets and the feature vectors of streetscape images into the graph structure as the input of the graph neural network, using the matched graph structure as the graph structure of the graph neural network, and using the calculated true carbon emission values ​​as label data for training the graph neural network; and evaluating the graph neural network model and optimizing the model parameters. A platform for predicting and visualizing carbon emissions from motor vehicles on urban streets is constructed, and the model output results are returned. The technical solution of the present invention provides a reliable method for predicting carbon emissions at the scale of urban street sections, thereby realizing refined carbon emission prediction for sections.

Description

Technical Field

[0001] The present invention relates to the fields of information technology and urban traffic carbon emissions, and in particular to a method and device for predicting and visualizing carbon emissions from motor vehicles on urban streets. Background Art

[0002] With the continuous increase in urban traffic volume, vehicle carbon emissions have become one of the main sources of urban carbon emissions. As an important means of public transportation, taxis have become one of the hot issues in the field of urban street transportation research.

[0003] Many researchers, both domestically and internationally, have calculated individual vehicle emissions based on motor vehicle carbon emission factors. Common models include the COPERT and CMEM models. However, these methods only measure the emission levels of individual vehicles and are limited by numerous factors, such as vehicle speed and fuel consumption. This makes it difficult to directly measure street-level carbon emissions. Other researchers have combined deep learning models with overall city carbon emissions to predict large-scale urban carbon emissions. However, this approach lacks the granularity required to accurately measure street-level emissions, making it difficult to provide decision-making support to relevant departments.

[0004] In recent years, with the rapid development of artificial intelligence and deep learning technologies, researchers in related fields have begun applying them to areas such as urban traffic management. Deep learning has been widely applied in the study of urban traffic emissions, achieving impressive results. Researchers have proposed corresponding deep learning models for different modes of transportation, such as a deep learning-based bus pollutant emission prediction model and a deep learning-based rail transit operational indicator prediction model. In the area of ​​urban street traffic, researchers have begun applying deep learning to predict pollution emissions from urban street traffic. Compared to traditional methods, deep learning methods not only accurately predict traffic emissions but also provide more refined predictions for emissions across different streets and time periods. In the transportation sector, some research has begun using deep learning to predict traffic flow and street congestion. Furthermore, some research has begun applying deep learning to areas such as traffic environment and air quality. However, there is still a significant research gap in the field of fine-grained carbon emission prediction at the street level. Generally, carbon emission predictions can only be achieved for individual vehicles or entire cities, without dynamic perception of the street environment. Researchers both domestically and internationally have also made some progress in addressing urban traffic emissions:

[0005] From a microscopic perspective, some physical models are already available for calculating the relevant carbon emissions of individual motor vehicles. Currently, motor vehicle microscopic models are relatively mature, and motor vehicle microscopic emission models primarily determine motor vehicle emissions by simulating motor vehicle emission characteristics and substituting the vehicle's own usage characteristics for calculations. "Analysis of Spatiotemporal Patterns of Carbon Dioxide Emissions from Taxi Trajectories in Wuhan" uses the CMEM model to invert the CO2 emissions of trajectory points in taxi travel trajectories, and "Analysis of Motor Vehicle Pollutant Emission Characteristics in Wuhan Based on the MOBILE6.2 Model" uses the MOBILE model for localized research on the emission characteristics of motor vehicle pollutants in Wuhan, calculating emission factors for different vehicle models. From a macroscopic perspective, some people use deep learning methods to predict the overall carbon emissions of a country or city.

[0006] Although traffic carbon emissions can be predicted in different ways, from a micro perspective, even if micro factors are combined with deep learning models to predict vehicle carbon emissions, it is not possible to directly predict carbon emissions at the road segment perception granularity. From a macro perspective, most predictions are made for the overall carbon emissions of the region, and there are few prediction methods that are as detailed as the road segment perception granularity.

[0007] In recent years, graph neural networks (GNNs) have shown great potential in geospatial applications due to their support for topological structures. Many researchers have applied GNNs to urban transportation, most notably for traffic flow prediction. However, few methods have been developed to predict street-level carbon emissions by integrating road network structure. Summary of the Invention

[0008] To solve the above problems, the present invention provides a method and apparatus for predicting and visualizing carbon emissions from motor vehicles on urban streets, wherein the method comprises the following steps:

[0009] S1. Calculate the true carbon emission value of motor vehicles on urban streets based on GPS trajectory data of motor vehicles on urban streets, obtain urban street images and urban road network structure, and pre-process the true carbon emission value, urban street images, and urban road network structure;

[0010] S2, converting the pre-processed urban road network structure into a graph structure, with urban streets as nodes of the graph structure;

[0011] S3. Extract features from the pre-processed urban street image using the CNN semantic segmentation model as node features of the graph structure, and encode the extracted node features into node feature vectors.

[0012] S4. Construct a graph neural network model, match the street motor vehicle GPS trajectory data and street view image feature vectors into a graph structure, use the street motor vehicle GPS trajectory data and street view image feature vectors matched into the graph structure as the input of the graph neural network, use the matched graph structure as the graph structure of the graph neural network, and use the true value of carbon emissions calculated in S1 as label data for training the graph neural network; provide feedback on the output value of the model and the error of the label data, continuously optimize the model, and evaluate the graph neural network model;

[0013] S5. Establish network services for encapsulating, calling and displaying the graph neural network model, and construct a platform for predicting and visualizing carbon emissions from motor vehicles on urban streets. By inputting GPS trajectory data of motor vehicles on urban streets, urban street view images and urban road network structures, the output results are displayed.

[0014] The present invention also proposes a device for predicting and visualizing carbon emissions of motor vehicles on urban streets, comprising:

[0015] processor;

[0016] a memory having stored thereon a computer program executable on the processor;

[0017] Wherein, when the computer program is executed by the processor, a method for predicting and visualizing carbon emissions of motor vehicles on urban streets is implemented.

[0018] The beneficial effects brought about by the technical solution provided by the present invention are:

[0019] This method uses features extracted from urban streetscape data and road network structures to construct input data. Graph neural networks can discover complex relationships and characteristics in urban transportation carbon emissions, thereby predicting and optimizing urban transportation efficiency and carbon emissions. Predicting street taxi carbon emissions through graph neural networks can promote the application and development of deep learning in the field of urban transportation carbon emissions. It also provides a reliable method for sensing carbon emissions at the scale of urban streets and road sections, achieving refined segment-level carbon emissions perception. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a flow chart of a method for predicting and visualizing carbon emissions of motor vehicles on urban streets implemented by the present invention;

[0021] Figure 2 This is a flowchart of a method for predicting and visualizing carbon emissions of motor vehicles on urban streets according to an embodiment of the present invention;

[0022] Figure 3 Schematic diagram of data collection and preprocessing according to an embodiment of the present invention;

[0023] Figure 4This is a diagram of a street view feature extraction module according to an embodiment of the present invention;

[0024] Figure 5 This is a diagram showing the specific structural process information of the model of the embodiment of the present invention;

[0025] Figure 6 This is a design diagram of a system for predicting carbon emissions from motor vehicles on urban streets according to an embodiment of the present invention;

[0026] Figure 7 This is a diagram of the road network structure of Wuhan City according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0028] A flowchart of a method for predicting and visualizing carbon emissions of motor vehicles on urban streets in this embodiment is shown in FIG. Figure 1 The flowchart of a method for predicting and visualizing carbon emissions of motor vehicles on urban streets according to an embodiment of the present invention is as follows: Figure 2 , specifically including the following steps:

[0029] S1. Calculate the true carbon emissions of motor vehicles on urban streets based on GPS trajectory data, obtain urban street images and urban road network structures, and preprocess the true carbon emissions, urban street images, and urban road network structures. Data preprocessing methods include cleaning, converting, and standardizing the data to improve its accuracy and consistency.

[0030] Among them, the GPS trajectory data of motor vehicles on urban streets includes motor vehicle type, motor vehicle speed, motor vehicle operation trajectory on the street, and street motor vehicle traffic volume.

[0031] The true value of carbon emissions from motor vehicles on urban streets is calculated based on the GPS trajectory data of motor vehicles on urban streets:

[0032] Combined with the GPS trajectory data of motor vehicles on urban streets, the COPERT model is used to calculate the true value of carbon emissions from motor vehicles on urban streets. First, the carbon emission factor is calculated:

[0033]

[0034] Where V is the speed of the vehicle, the values ​​of a, b, c, d, and e are determined by different types of vehicles, and E HOT(C) is the carbon emission factor;

[0035] The obtained carbon emission factor is multiplied by the total length of the street and the motor vehicle flow on the street to obtain the total carbon emissions of all motor vehicles on the street. Based on the operation trajectory of motor vehicles on the street, the total carbon emissions of all motor vehicles on all streets are obtained.

[0036] S2. Convert the preprocessed urban road network structure into a graph structure, with urban streets as nodes of the graph structure.

[0037] S3. Use convolutional neural network to extract features from the preprocessed urban street images as node features of the graph structure, and encode the extracted features into feature vectors.

[0038] This embodiment uses a convolutional neural network (CNN) image semantic segmentation model (such as DeepLabV3, EfficientNet, SegNet, etc.) to extract features. The image semantic segmentation model (mainly DeepLabV3) can effectively capture the object, background and scene information in the image and encode it into feature vectors. These feature vectors can represent the physical and environmental attributes of the street, such as road conditions, traffic conditions, surrounding buildings, etc. This embodiment uses the mature existing technology DeepLabV3 for semantic segmentation reference. Figure 4 , Figure 4 2 is a diagram of a street view feature extraction module according to an embodiment of the present invention.

[0039] DeepLabV3 includes an encoder and a decoder. The encoder uses atrous convolution to control the resolution of the features calculated in the DCNN. The atrous convolution includes convolutional and pooling layers. The input (street image) passes through a 1×1 Conv, a 3×3 Conv (rate=6), a 3×3 Conv (rate=12), a 3×3 Conv (rate=18), and an image pooling layer for fusion. Then, it passes through another 1×1 Conv. The result is input to a 4x upsampling module of the decoder to obtain result A. The low-level features of the input (street image) after atrous convolution are input to the 1×1 Conv of the decoder to obtain result B. Result A and result B are spliced ​​through a Concat, and after splicing, they pass through a 3×3 Conv and a 4x upsampling to obtain the final extracted image feature vector.

[0040] Here combined Figure 4 To illustrate the specific form of street features and feature vectors, the features extracted from a street image are [taxi, road, tree, building, sky, ..., wall, bus, bicycle], which are encoded as the feature vector [0.369, 0.008, 0.055, 0.017, 0.043, ..., 0.001, 0.003, 0, 0.00062]. Of course, this is only an example for ease of understanding and does not limit the feature encoding method of the present invention.

[0041] S4. Construct a graph neural network model, match the street motor vehicle GPS trajectory data and street view image feature vectors into the graph structure, use the street motor vehicle GPS trajectory data and street view image feature vectors matched into the graph structure as the input of the graph neural network, use the matched graph structure as the graph structure of the graph neural network, and use the true value of carbon emissions calculated in S1 as the label data for graph neural network training; feed back the output value of the model and the error of the label data, continuously optimize the model, and evaluate the graph neural network model.

[0042] The specific steps for matching urban street motor vehicle carbon emissions and street image feature vectors into the graph structure are:

[0043] The urban road network structure is used as a graph structure, the urban streets are nodes of the graph structure, the feature vectors of the urban street images are used as feature vectors of the nodes, and the GPS trajectory data of street motor vehicles are encoded and added to the feature vectors of the nodes.

[0044] For example, if the motor vehicle type is a taxi, which is coded as 1, and the traffic volume in a certain time period on a certain street is 8, which is also coded as 8, the code [1.8] is added to the feature vector of the street node.

[0045] The method for evaluating the graph neural network model is: divide the matched urban street motor vehicle GPS trajectory data, street view image feature vectors, and graph structure data into different test sets and validation sets, use different test sets and validation sets to train and verify the graph neural network model, obtain multiple model evaluation results, and calculate the average of multiple model evaluation results.

[0046] Use parameter tuning techniques to optimize model hyperparameters:

[0047] The hyperparameters that need to be set in the model are learning rate, regularization parameter, hidden layer size, etc. By systematically trying different hyperparameter combinations, comparing model performance, and selecting the best hyperparameter combination, model performance can be improved.

[0048] Analyze and improve the model errors:

[0049] For data with large errors, conduct a detailed analysis to identify model weaknesses and areas for improvement. For example, you can use visualization methods to view the differences between the model's predictions and the actual results to identify common error patterns. You can also try adding more features or refactoring the model to address specific types of errors.

[0050] S5. Establish network services for encapsulating, calling and displaying the graph neural network model, and construct a platform for predicting and visualizing carbon emissions from motor vehicles on urban streets. By inputting GPS trajectory data of motor vehicles on urban streets, urban street view images and urban road network structures, the output results are displayed.

[0051] This visualization platform consists of two parts: user interface and backend services, such as Figure 6 As shown in the figure, the user interface provides real-time carbon emission prediction and route planning capabilities. The backend service is responsible for receiving user input, such as street view imagery, vehicle model, and street number information, passing this information to the model for prediction, and finally returning the prediction results to the user. The visualization platform is capable of receiving and processing taxi routes and street view imagery data in real time, and then using a graph neural network model to make predictions. When integrating the model and data processing pipeline, it is necessary to ensure the efficiency and stability of the entire visualization platform. The visualization platform is capable of collecting user feedback and data, and continuously optimizes and updates the model based on this feedback and data. This improves the accuracy and reliability of the model while enhancing the user experience of the prototype system. In addition to the above functions, other system functions related to predicting carbon emissions of motor vehicles on urban streets can be added as needed.

[0052] The embodiment also includes a device for predicting and visualizing carbon emissions of motor vehicles on urban streets, comprising:

[0053] processor;

[0054] a memory having stored thereon a computer program executable on the processor;

[0055] Wherein, when the computer program is executed by the processor, a method for predicting and visualizing carbon emissions of motor vehicles on urban streets is implemented.

[0056] In a further embodiment, Wuhan City is selected as the research area, and reference is made to Figure 3 , Figure 3 This is a schematic diagram of data collection and preprocessing for an embodiment of the present invention. The taxi GPS data used comes from the Wuhan Municipal Transportation Administration. The total CO emissions from motor vehicles on each street in Wuhan are calculated using the vehicle CO emission factor based on the taxi GPS trajectories. The calculated data is then divided into seven levels using the natural discontinuity method as label data. The Wuhan road network structure is obtained from OSM. The Wuhan OSM street network structure is used as the graph structure for the GNN network. Each street is a node in the graph, for a total of 5075 nodes. Figure 7 As shown, Figure 7This is a diagram of the Wuhan road network structure according to an embodiment of the present invention. The street view images were obtained from 56,461 street panorama data downloaded from Tencent Maps. A pre-trained DeepLabV3 model was used for semantic segmentation feature extraction, which segmented the images into 19 feature categories. A two-layer SAGE network was designed using the DGL framework for node classification tasks. The street features and graph structure were combined as input to the SAGE model. The resulting input data and label data were trained using the SAGE model. The dataset was divided into a training set and a validation set. The model was trained and optimized, ultimately yielding predicted CO emission levels.

[0057] Based on the CO emission factors of motor vehicles, the CO emissions of motor vehicles on Wuhan streets can be calculated. Based on the CO emissions of taxi streets, the CO emissions are divided into seven different levels using the natural interval method to form the true value of taxi street CO emissions, which is consistent with the CO emissions statistics in the "2017 Wuhan Motor Vehicle Pollution Prevention and Control Annual Report".

[0058]

[0059] Where V is the taxi speed, E HOT(C) is the carbon emission factor;

[0060] The obtained carbon emission factor is multiplied by the total length of the street and the taxi flow on the street to obtain the total carbon emissions of all taxis on the street. Based on the operation trajectory of taxis on the street, the total carbon emissions of all taxis on all streets are obtained.

[0061] refer to Figure 5 , determine the input and output of the SAGE model, set parameters, and train the model. Determine the model structure and hierarchy, design a two-layer SAGE network, each node is sufficient to obtain the characteristics of all other nodes, and select LSTM as the aggregation function of SAGE for experiments, because LSTM has stronger expression ability and better ability to process time series data, which can better capture the front-end dependency and dynamics between nodes. Set the order of the neighbor nodes to be aggregated (that is, the distance extending outward from the neighbor nodes of each node) to 2. The model will aggregate the information of the neighbor nodes of each node and the neighbor nodes of the neighbor nodes. The ReLu activation function is used to improve the expression ability and computational efficiency of the model. The loss function is the cross entropy function, and the optimizer is the Adam optimization estimate. In this example, the SAGE model, reference Figure 5The input data is the Wuhan road network structure, which is converted into a large undirected graph structure with a total of 5075 nodes. The circles in the figure represent nodes, the numbers in them represent street numbers, and the lines between nodes represent the connections between streets. The connections between all streets constitute the road network graph structure. In the road network structure diagram, [] represents a feature vector. Each node represents a street. The node (street) contains 19 basic features: car, road, building, wall, railing, tree, lawn, sign, bus, motorcycle, truck, sky, pole, light, pedestrian, cyclist, train, bicycle, and sidewalk. The output value is the CO2 emission level of taxis on the street, with a total of 7 levels. The numbers in [] in the carbon emission level diagram of each street represent the carbon emission level. The true label value is the CO2 emission of taxis on the street calculated using the CO2 emission factor.

[0062] The basic features of the graph structure are the result of matching the street view image feature vectors to the graph structure. Based on the basic features, the street motor vehicle GPS trajectory data and the street view image feature vectors are matched to the graph structure. Features are further added based on the basic features. For example, the motor vehicle type is a taxi, coded as 1, and the traffic volume in a certain time period on a certain street is 8, coded as 8. The feature vector coded [1.8] is added to the feature vector of the street node, and the number of features increases to 20. Data such as street length and motor vehicle speed can also be added to feature vectors using this method. Other features, such as weather conditions, can also be added as needed.

[0063] The accuracy of CO2 predictions for taxis on streets in Wuhan was verified using both validation and test datasets. The model's predictions were largely consistent with the true values ​​on the test set, achieving an accuracy of 81.4%, with the exception of a few discrepancies. The model was able to accurately predict CO2 emissions from taxis on the test set, with mean square error (MSE), relative mean error (RMSE), magnitude of error (MAE), mapelement of error (MAPE), mean average (ME), and mean sum (MSL) values ​​of 0.340, 0.583, 0.254, 0.167, 4.0, and 0.042, respectively. The model exhibited no significant overfitting issues, demonstrated high generalization and scalability for carbon emissions prediction, and demonstrated excellent performance across all evaluation metrics, demonstrating its high application value.

[0064] Step 5: Establish a network service for graph neural network model encapsulation, calling services and interface display, build a platform for predicting and visualizing carbon emissions of motor vehicles on urban streets, and return the prediction results by inputting carbon emissions data of motor vehicles on urban streets, urban street scene images, and urban road network structure.

[0065] To further test the feasibility of the urban vehicle street carbon emissions prediction method used in this paper and verify its effectiveness and performance, we selected 10 common graph neural network models—GCN, GCN2, GAT, GAT2, Edge, SG, Cheb, PNA, DGN, and TAG—for modeling and comparison on training and test sets. The optimal hyperparameter combinations for each model under the same input conditions were selected, and the performance of different models in predicting urban taxi street carbon emissions was compared, as shown in Table 1. Table 1 compares the CO prediction accuracy of different GNN models.

[0066] Table 1 Comparison of CO prediction accuracy of different GNN models

[0067]

[0068] The prediction results, as shown in Table 1, show that the SAGE model outperformed all other models in most metrics. Specifically, while the PNA model achieved the lowest performance in all metrics except ME and exhibited better prediction performance in terms of error metrics, it had the worst accuracy in predicting CO emissions and was therefore excluded from the comparison with other models. The SAGE model, however, achieved the lowest MSE, RMSE, MAE, MAPE, ME, and MSL values, indicating that it possessed the lowest overall error compared to the other models. The GCN, GCN2, GAT, GAT2, Edge, SG, Cheb, DGN, and TAG models performed relatively similarly, with their best accuracy ranging from 81.1% to 81.5%, and their error metrics also being relatively close. The TAG model's MSE, RMSE, MAE, MAPE, and MSL values ​​were second only to the SAGE model, but its best accuracy was lower than that of the SAGE model.

[0069] Step 5: Develop a carbon emission prediction service system, such as Figure 5 The system adopts a microservices architecture. Technically, the front-end utilizes commonly used front-end development tools such as Vue, ElementUI, and Echarts, while the back-end utilizes the Spring Cloud system, including microservice components such as Nacos, Sentinel, RabbitMQ, and Dubbo, as well as databases such as MySQL. Business-wise, the system implements basic business functions such as permission control and logging, as well as specialized business functions such as user registration, data upload, model prediction, and result presentation.

[0070] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting and visualizing carbon emissions from motor vehicles on urban streets, characterized in that: The following steps are involved: S1. Calculate the true carbon emission value of motor vehicles on urban streets based on GPS trajectory data of motor vehicles on urban streets, obtain urban street images and urban road network structure, and pre-process the true carbon emission value, urban street images, and urban road network structure; S2, converting the pre-processed urban road network structure into a graph structure, with urban streets as nodes of the graph structure; S3. Extract features from the pre-processed urban street image using the CNN semantic segmentation model as node features of the graph structure, and encode the extracted node features into node feature vectors. S4. Construct a graph neural network model, match the street motor vehicle GPS trajectory data and street view image feature vectors into a graph structure, use the street motor vehicle GPS trajectory data and street view image feature vectors matched into the graph structure as the input of the graph neural network, use the matched graph structure as the graph structure of the graph neural network, and use the true value of carbon emissions calculated in S1 as label data for training the graph neural network; provide feedback on the output value of the model and the error of the label data, continuously optimize the model, and evaluate the graph neural network model; S5. Establish a network service for encapsulating, calling, and displaying the graph neural network model, and build a platform for predicting and visualizing carbon emissions from motor vehicles on urban streets. By inputting GPS trajectory data of motor vehicles on urban streets, urban street view images, and urban road network structures, the platform can display the output results. In step S4, the graph neural network model adopts the SAGE model, which adopts the ReLu activation function, the cross entropy function as the loss function, the Adam optimization estimation as the optimizer, and a two-layer SAGE network is designed. LSTM is used as the aggregation function of SAGE, and the order of the neighbor nodes to be aggregated is set to n. The information of the n-order neighbor nodes of each node is aggregated to capture the mutual connection relationship between nodes.

2. The method for predicting and visualizing carbon emissions of motor vehicles on urban streets according to claim 1, characterized in that: The GPS trajectory data of motor vehicles on urban streets in step S1 includes the type of motor vehicle, the speed of the motor vehicle, the running trajectory of the motor vehicle on the street, and the traffic volume of motor vehicles on the street.

3. The method for predicting and visualizing carbon emissions of motor vehicles on urban streets according to claim 2, characterized in that: The true value of carbon emissions from motor vehicles on urban streets is calculated based on the GPS trajectory data of motor vehicles on urban streets: Combined with the GPS trajectory data of motor vehicles on urban streets, the COPERT model is used to calculate the true value of carbon emissions from motor vehicles on urban streets. First, the carbon emission factor is calculated: Where V is the speed of the vehicle, 、 The values ​​of , c, d, and e are determined by different types of motor vehicles. is the carbon emission factor; The obtained carbon emission factor is multiplied by the total length of the street and the motor vehicle flow on the street to obtain the total carbon emissions of all motor vehicles on the street. Based on the operation trajectory of motor vehicles on the street, the total carbon emissions of all motor vehicles on all streets are obtained.

4. The method for predicting and visualizing carbon emissions of motor vehicles on urban streets according to claim 1, characterized in that: In step S1, the data preprocessing method is: cleaning, converting and standardizing the data to improve the accuracy and consistency of the data.

5. The method for predicting and visualizing carbon emissions of motor vehicles on urban streets according to claim 2, characterized in that: In step S4, the carbon emissions of motor vehicles on urban streets and the street image feature vectors are matched to the graph structure as follows: The urban road network structure is taken as a graph structure, and the urban streets are the nodes of the graph structure. The urban streets are numbered, and the urban streets with different numbers are connected with lines to represent the connection relationship between the nodes. The feature vector of the city street image is used as the feature vector of the node, and the GPS trajectory data of the street motor vehicle is encoded and added to the feature vector of the node to obtain the feature vector of the matched graph structure.

6. The method for predicting and visualizing carbon emissions of motor vehicles on urban streets according to claim 5, characterized in that: The specific encoding of the street motor vehicle GPS trajectory data is as follows: the street motor vehicle GPS trajectory data is encoded into floating point data [ab] and added as a feature vector to the graph node feature vector.

7. The method for predicting and visualizing carbon emissions of motor vehicles on urban streets according to claim 1, characterized in that: In step S5, the method for evaluating the graph neural network model is: dividing the matched urban street motor vehicle GPS trajectory data, street view image feature vectors, and graph structure data into different test sets and verification sets, using different test sets and verification sets to train and verify the graph neural network model, obtaining multiple model evaluation results, and calculating the average of the multiple model evaluation results.

8. A device for predicting and visualizing carbon emissions from motor vehicles on urban streets, characterized in that: The device comprises: processor; a memory having stored thereon a computer program executable on the processor; Wherein, when the computer program is executed by the processor, a method for predicting and visualizing carbon emissions of motor vehicles on urban streets as described in any one of claims 1 to 7 is implemented.

Images