Urban road carbon emission prediction and optimization control method and system

Abstract

The invention relates to an urban road carbon emission prediction and optimization control method and system, and belongs to the technical field of road traffic carbon emission monitoring, and the method comprises the steps: obtaining real-time monitoring data of a to-be-detected road, and carrying out the preprocessing; performing space-time alignment on the preprocessed real-time monitoring data, extracting dynamic traffic features and static road features, and generating a space-time feature matrix; inputting the spatial-temporal characteristic matrix into a pre-trained carbon emission prediction model to obtain a prediction result matrix of the to-be-detected road; on the basis of the prediction result matrix, identifying a road grid of which the carbon emission intensity prediction value exceeds a preset threshold value as a high emission area; acquiring real-time traffic state data of the high emission area; and based on the real-time traffic state data and the prediction result matrix of the high emission area, constructing a multi-objective optimization function and carrying out solving to obtain an optimization control strategy of the high emission area. According to the invention, traffic carbon emission optimization control and real-time traffic condition synchronization can be realized.

Description

Technical Field

[0001] The present application relates to the technical field of road traffic carbon emission monitoring, and in particular to a method and system for predicting and optimizing carbon emissions from urban roads. Background Art

[0002] With the accelerating pace of global urbanization, urban transportation faces unprecedented pressure. Rising urban populations and booming economic activity have driven a sharp increase in traffic volume, which not only exacerbates traffic congestion but also worsens environmental pollution, particularly the impact of carbon emissions on the atmospheric environment. Transportation, as a major source of energy consumption, and especially greenhouse gases like carbon dioxide (CO2) emitted by vehicles, are a major factor in deteriorating urban air quality and climate change.

[0003] In many cities, especially those undergoing rapid development, transportation emissions have become a major source of carbon emissions. Furthermore, with the increasing number of vehicles, particularly those powered by traditional fuel, emissions are becoming increasingly severe, leading to a continuous increase in energy consumption within urban transportation systems and posing a significant challenge to sustainable urban development.

[0004] To more accurately predict and effectively control transportation carbon emissions, intelligent transportation systems (ITS) have garnered increasing attention in recent years. ITS integrates advanced communications, sensor, computer science, and data analytics technologies, aiming to improve transportation efficiency and reduce energy consumption and pollution emissions. However, despite the promise of ITS and big data technologies for carbon emission control, common carbon emission prediction methods in practice often rely on simple linear or static statistical models. These methods are unable to accurately capture the complex nonlinear relationships between traffic flow, vehicle behavior, and environmental factors, and lack dynamic adaptability. This makes it difficult to synchronize optimized control of transportation carbon emissions with real-time traffic conditions and to rapidly respond to abnormal changes in carbon emissions. Summary of the Invention

[0005] In order to achieve synchronization between optimized control of traffic carbon emissions and real-time traffic conditions, the present application provides a method and system for predicting and optimizing carbon emissions from urban roads.

[0006] In the first aspect, the present application provides a method for predicting and optimizing carbon emissions from urban roads, which adopts the following technical solutions: A method for predicting and optimizing carbon emissions from urban roads, characterized in that the method comprises: Acquire real-time monitoring data of the road to be inspected and perform preprocessing; wherein the real-time monitoring data includes real-time traffic flow data, environmental data, and vehicle emission data; Performing spatiotemporal alignment on the pre-processed real-time monitoring data, extracting dynamic traffic features and static road features, and generating a spatiotemporal feature matrix; Inputting the spatiotemporal feature matrix into a pre-trained carbon emission prediction model to obtain a prediction result matrix for the road to be detected; Based on the prediction result matrix, identifying road grids where the predicted carbon emission intensity exceeds a preset threshold as high emission areas; Obtaining real-time traffic status data of the high-emission area; Based on the real-time traffic status data and the prediction result matrix of the high-emission area, a multi-objective optimization function is constructed and solved to obtain an optimized control strategy for the high-emission area.

[0007] By adopting the above technical solutions, multi-source data can be collected in real time, and the spatiotemporal changes in traffic operation and carbon emissions can be fully understood. Carbon emission intensity can be predicted in real time through deep learning models, and high-emission areas can be identified. Through reinforcement learning and multi-objective optimization methods, traffic signals and path guidance can be precisely controlled, achieving accurate identification and dynamic regulation of high-risk areas for urban carbon emissions. This not only improves the intelligence and accuracy of urban traffic management, but also provides effective technical support for achieving sustainable urban development.

[0008] Optionally, the steps of performing spatiotemporal alignment on the pre-processed real-time monitoring data, extracting dynamic traffic features and static road features, and generating a spatiotemporal feature matrix include: Dividing the pre-processed real-time monitoring data into continuous data and discrete data; Perform Z-score normalization on the continuous data and one-hot encoding on the discrete data; Using the traffic signal controller clock as a benchmark, the timestamps of each data source are aligned through Kalman filtering; Based on the time-space aligned data, the average vehicle idling time and acceleration variance are calculated according to the pre-divided road grid as dynamic traffic characteristics; Extract road slope weight coefficient and grid adjacency relationship from pre-stored high-precision maps as static road features; The dynamic traffic features and static road features of each road grid are spliced into a complete feature vector; Based on the complete feature vectors, they are sorted according to timestamps and road grid IDs to construct a three-dimensional spatiotemporal feature matrix.

[0009] By employing this technical solution, we can effectively process traffic data from various sensors and data sources, extracting both dynamic and static traffic characteristics. These characteristics are integrated into a high-dimensional spatiotemporal feature matrix, which serves as input for the carbon emissions prediction model. This technical solution not only monitors and predicts traffic conditions in real time but also adjusts traffic management strategies based on real-time data, thereby reducing traffic carbon emissions and achieving the goal of intelligent traffic management.

[0010] Optionally, the method further includes a step of training a carbon emission prediction model, wherein the training step includes: Obtain a historical spatiotemporal feature matrix and real carbon emission label data of the corresponding grid to construct a sample data set; wherein the historical spatiotemporal feature matrix includes dynamic traffic characteristics and static road characteristics of the historical road grid; Dividing the sample data set into a training set, a validation set, and a test set; Inputting the training set into a pre-built hybrid neural network model for training, calculating the error between the carbon emission intensity prediction value and the actual carbon emission label data through a backpropagation algorithm, calculating the gradient layer by layer, and updating the weights and bias of the hybrid neural network model; Validating the hybrid neural network model based on the validation set, evaluating the performance of the hybrid neural network model and adjusting model hyperparameters according to the validation results; The prediction capability of the adjusted hybrid neural network model is tested based on the test set to obtain the trained carbon emission prediction model.

[0011] By adopting the aforementioned technical solution and introducing a hybrid neural network model structure combining graph neural networks and long-short-term memory networks, this approach achieves the collaborative modeling of spatial and temporal features for road carbon emissions prediction. Furthermore, a multi-objective joint training strategy and a validation-testing hierarchical evaluation process enhance the model's generalization performance and deployment stability. Ultimately, this model can be deployed on edge computing devices, achieving high-precision, low-latency predictions of traffic carbon emissions, providing strong technical support for urban green transportation management.

[0012] Optionally, the step of pre-building the hybrid neural network model includes: Obtain the adjacency matrix of the static road features of the historical road grid; Adding self-connection to the adjacency matrix to generate an extended adjacency matrix, and calculating the degree matrix of the extended adjacency matrix; The input layer structure is defined based on the dynamic traffic feature dimensions of the historical road grid, and a hidden layer consisting of two layers of long short-term memory network units is constructed, with a preset number of neurons in each layer. Configuring an activation function of the long short-term memory network unit to output a temporal context feature vector; Combining the extended adjacency matrix and the degree matrix into a normalized adjacency matrix; Constructing a graph convolution layer based on the normalized adjacency matrix, configuring a trainable weight matrix and an activation function of the graph convolution layer, and outputting a spatial topological feature vector; Concatenating the temporal context feature vector and the spatial topology feature vector to obtain a fused feature vector; Performing dimensionality reduction processing on the fused feature vector through a fully connected layer to generate a joint feature vector; Define probability distribution parameters for the fully connected layer weights and initialize the mean and standard deviation of the probability distribution parameters; An output layer mapping relationship is constructed to generate a complete model architecture including predicted values and confidence intervals to obtain the hybrid neural network model.

[0013] By employing these technical solutions and integrating static road topology with dynamic traffic time series characteristics, a hybrid neural network model based on graph convolutional networks and long-short-term memory networks was constructed, enabling accurate prediction of road carbon emissions. This model not only captures the spatial dependencies between road grids but also provides a deep understanding of the dynamic patterns of traffic characteristics over time. Furthermore, by incorporating probability distribution modeling to provide prediction confidence intervals, it significantly improves the accuracy, robustness, and interpretability of the predictions.

[0014] Optionally, the step of constructing and solving a multi-objective optimization function based on the real-time traffic status data and the prediction result matrix of the high-emission area to obtain an optimized control strategy for the high-emission area includes: Obtain real-time traffic status data and prediction result matrix of high emission areas; Extracting a set of intersections associated with a road grid in the high-emission area and a set of vehicles traveling toward the high-emission area; Based on the intersection set and the vehicle set, an objective function of minimizing regional total carbon emissions is defined, and traffic delay upper limit, signal cycle boundary and path connectivity constraints are added to the objective function; Combining the prediction result matrix and the real-time traffic light state into a state vector and performing normalization processing; Generate signal control actions and path induction actions based on proximal policy optimization algorithm; Defining a signal control action for the intersection set and encoding the signal control action into a signal light control instruction; wherein the signal control action includes a cycle adjustment amount and a green-to-signal ratio adjustment amount; defining a path induction action for the vehicle set and specifying a target node selection range; According to the target node output by the path induction action, a vehicle detour path is calculated based on a shortest path algorithm and encapsulated into a vehicle path induction instruction; The signal light control instructions and the vehicle path guidance instructions are combined to obtain an optimized control strategy for the high emission area.

[0015] By employing this technical solution, combined with real-time traffic data and predictions of high-emission areas, and using a reinforcement learning algorithm to dynamically optimize traffic signals and routing, this approach reduces carbon emissions while ensuring smooth traffic flow. By defining precise control actions and optimization targets, the entire system effectively achieves the dual goals of low carbon emissions and high-efficiency transportation, making it suitable for intelligent transportation and smart city development.

[0016] Optionally, after obtaining the optimized control strategy for the high-emission area, the method further includes: Sending the optimized control strategy to the roadside execution unit in the high-emission area; wherein the optimized control strategy includes signal light control instructions and vehicle path guidance instructions; Obtaining a prediction result matrix generated by the carbon emission prediction model based on real-time traffic status data and the optimization control strategy; Collecting actual traffic flow data and vehicle emission data after the optimization control strategy is executed; Calculating a deviation matrix between predicted values and actual values based on the predicted values in the prediction result matrix and the actual traffic flow data and vehicle emission data; Parameters of the carbon emission prediction model are corrected according to the deviation matrix to obtain an updated carbon emission prediction model.

[0017] By employing this technical solution and using the deviation between predicted results and actual data to calibrate the model, the accuracy of the carbon emission prediction model is ensured. This closed-loop optimization method not only optimizes carbon emission control but also ensures smooth traffic flow, ultimately providing strong technical support for smart transportation and green travel.

[0018] In a second aspect, the present application provides an urban road carbon emission prediction and optimization control system, which adopts the following technical solutions: An urban road carbon emission prediction and optimization control system, the system comprising: A data monitoring and processing module is used to obtain real-time monitoring data of the road to be inspected and perform preprocessing; wherein the real-time monitoring data includes real-time traffic flow data, environmental data and vehicle emission data; A spatiotemporal feature matrix generation module is used to perform spatiotemporal alignment on the pre-processed real-time monitoring data, extract dynamic traffic features and static road features, and generate a spatiotemporal feature matrix; A carbon emission prediction module, configured to input the spatiotemporal feature matrix into a pre-trained carbon emission prediction model to obtain a prediction result matrix for the road to be inspected; A carbon emission intensity identification module, configured to identify, based on the prediction result matrix, road grids whose predicted carbon emission intensity values exceed a preset threshold as high emission areas; A traffic status acquisition module, configured to acquire real-time traffic status data of the high-emission area; The optimization control module is used to construct and solve a multi-objective optimization function based on the real-time traffic status data and the prediction result matrix of the high-emission area to obtain an optimization control strategy for the high-emission area.

[0019] In a third aspect, the present application provides a computer device that adopts the following technical solution: A computer device comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method according to the first aspect.

[0020] In a fourth aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium stores a computer program capable of being loaded by a processor and executing any one of the methods in the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a first flow chart of a method for predicting and optimizing carbon emissions from urban roads according to one of the embodiments of the present application.

[0022] Figure 2 This is a second flow chart of a method for predicting and optimizing carbon emissions from urban roads according to one of the embodiments of the present application.

[0023] Figure 3 This is a third flow chart of a method for predicting and optimizing carbon emissions from urban roads according to one of the embodiments of the present application.

[0024] Figure 4 This is a fourth flow chart of a method for predicting and optimizing carbon emissions from urban roads according to one of the embodiments of the present application.

[0025] Figure 5 This is the fifth flow chart of a method for predicting and optimizing carbon emissions from urban roads according to one of the embodiments of the present application.

[0026] Figure 6 This is the sixth flow chart of a method for predicting and optimizing carbon emissions from urban roads according to one of the embodiments of the present application. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solutions and advantages of this application more clear, the following Figure 1-6It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application.

[0028] The embodiments of the present application disclose a method for predicting and optimizing carbon emissions from urban roads.

[0029] Reference Figure 1 , a method for predicting and optimizing carbon emissions from urban roads, the method comprising: Step S101, obtaining real-time monitoring data of the road to be detected and performing preprocessing; Among them, real-time monitoring data includes real-time traffic flow data, environmental data and vehicle emission data; Specifically, traffic flow data is primarily acquired through geomagnetic sensors, video detectors, or floating vehicle systems. It includes key indicators such as vehicle speed, traffic density, lane occupancy, and the proportion of new energy vehicles. These indicators are direct factors influencing changes in carbon emissions per unit mileage. Environmental data focuses on the indirect regulatory effects of external physical conditions on emissions. For example, temperature changes can alter engine efficiency, humidity affects road friction, and slope determines load energy consumption trends. Vehicle emissions data relies on real-time readings of instantaneous CO2 emissions from vehicles using on-board diagnostics (OBD) equipment. This high-frequency and high-granularity data is a crucial foundation for building micro-emission estimation models.

[0030] Furthermore, data preprocessing steps primarily include outlier removal, missing value filling, and format standardization. Outlier removal typically utilizes statistical distribution models combined with spatiotemporal constraints, such as the 3σ rule to filter illogical velocity fluctuations. Missing value filling typically utilizes time series interpolation (such as linear or Lagrange interpolation) or spatial neighboring interpolation (KNN) to enhance spatiotemporal data integrity. Data format standardization ensures consistency across multiple sources through unit standardization and field structure alignment in subsequent processing. This step transforms raw, multi-source, heterogeneous data into standardized, structured input data, providing reliable data support for the model.

[0031] Step S102: performing spatiotemporal alignment on the pre-processed real-time monitoring data, extracting dynamic traffic features and static road features, and generating a spatiotemporal feature matrix; Among them, the spatiotemporal feature matrix includes timestamp, road grid ID and feature vector; Specifically, heterogeneous data is uniformly mapped across time and space, and a feature matrix containing complete semantic information is constructed. In terms of spatiotemporal alignment, the time dimension is typically discretized using fixed time windows (e.g., 5 minutes, 15 minutes), and data from different sources must be aligned via timestamps to achieve event synchronization. The spatial dimension, based on a grid-based partitioning model, divides the urban road network into several geographically continuous grid cells. Common methods include static regular partitioning (e.g., 0.01° longitude and latitude grids) or dynamic adaptive partitioning based on road density, thereby providing a geographic index for subsequent spatial modeling.

[0032] Furthermore, when extracting features, dynamic traffic features focus on reflecting the statistical laws of time-varying traffic behavior. For example, "average idling time" can reflect road congestion trends, and "acceleration variance" is used to characterize the severity of vehicle acceleration behavior and is often used to characterize the impact of driving behavior on emissions. Static road features such as the "slope weight coefficient" are factors that influence the slope on vehicle load, calculated using road elevation data. This calculation can be performed by numerically integrating the slope differences between grids based on the digital elevation model (DEM). The resulting spatiotemporal feature matrix structure is in the form of a three-dimensional tensor, with index dimensions including timestamp (t), road grid number (g), and feature vector (F). This matrix serves as the core input to the prediction model and comprehensively expresses the operating status and physical environment of the road at a certain point in time.

[0033] Step S103: input the spatiotemporal feature matrix into a pre-trained carbon emission prediction model to obtain a prediction result matrix for the road to be tested; The prediction result matrix includes road grid ID, predicted carbon emission intensity and confidence interval; Specifically, the system incorporates a hybrid neural network architecture from deep learning, combining the strengths of long short-term memory (LSTM) and graph neural networks (GNN) to jointly model time series characteristics and spatial topological features. LSTM uses a gating mechanism to capture long-term temporal dependencies in traffic characteristics, such as periodic variations between morning and evening rush hours and short-term spikes caused by emergencies. GNN, on the other hand, models the propagation of traffic states between adjacent grids based on the road network's adjacency matrix, thereby capturing, for example, the transmission effect of upstream road congestion on downstream emissions.

[0034] Furthermore, to quantify the uncertainty of the prediction interval, the model incorporates a Bayesian neural network (BNN) structure at the output. By introducing probability distribution parameters into the neuron weights, the output not only includes the predicted value but also estimates its confidence interval. In this solution, the confidence interval is set to 95%, meaning there is a 95% probability that the predicted value will be included within the prediction interval. This mechanism effectively enhances the model's credibility and facilitates risk control in subsequent optimization strategies. The final output is presented as a prediction matrix, with each row containing the road grid ID, the predicted carbon emission intensity value, and its upper and lower confidence bounds, forming a high-confidence prediction basis that can be used to control inputs.

[0035] Step S104: Based on the prediction result matrix, identify road grids whose predicted carbon emission intensity exceeds a preset threshold as high emission areas; Based on a city's carbon emission control targets, a risk threshold for emission intensity (e.g., 150 g / km) is set to screen out grids exceeding the standard. It should be noted that simple single-point identification can be subject to noise. Therefore, density-based spatial clustering algorithms (e.g., DBSCAN) or morphological expansion methods are often used to merge geographically adjacent high-emission grids into a connected region, facilitating systematic regulation.

[0036] Furthermore, the identification process combines the predicted value and its confidence interval. When the lower confidence bound also exceeds the threshold, the area is considered high-confidence, high-emission area; otherwise, it is marked as medium risk, thus implementing a dynamic graded response mechanism. This identification strategy has a certain degree of "error tolerance" and avoids overreacting to areas with large prediction errors.

[0037] Step S105, obtaining real-time traffic status data of high-emission areas; Traffic guidance parameters and control constraints within the identified areas are perceived in real time to ensure that optimization strategies align with current traffic conditions. Real-time traffic status data includes, but is not limited to, current intersection signal timing parameters (green-to-signal ratio, cycle, and phase sequence), lane flow, headway, average queue length, and path distribution probabilities and turn ratios in guide lanes within high-emission zones.

[0038] Step S106 : Based on the real-time traffic status data and the prediction result matrix of the high-emission area, a multi-objective optimization function is constructed and solved to obtain an optimized control strategy for the high-emission area.

[0039] The optimized control strategy includes signal control instructions (adjusting the signal cycle and green-to-red ratio at intersections associated with high-emission areas) and route guidance instructions (generating detour routes for vehicles heading towards high-emission areas).

[0040] Specifically, based on real-time traffic status data, a multi-objective optimization model integrating traffic efficiency and emission reduction is constructed. The objective of the optimization function is to minimize the total carbon emissions of the entire area, while the constraints include the maximum traffic delay time, signal cycle boundaries, path accessibility, etc. To solve this optimization problem, a policy gradient-based reinforcement learning algorithm, Proximal Policy Optimization (PPO), can be used. This method constructs an environment-state-action model, defines the prediction result matrix and the traffic light state as the "state space", and defines the signal control instructions (such as cycle adjustment, green light ratio) and path induction instructions (such as target node redirection) as the "action space". Through continuous interaction and reward function iteration, the optimal strategy is approached.

[0041] In one embodiment of the present application, the path guidance scheme introduces an improved Dijkstra path search algorithm and uses emission intensity and path length weighted as the path cost function, namely: Cost=γ·Epred+δ·Dpath, where γ>δ; This function design ensures that the detour plan prioritizes emission reduction while taking into account traffic efficiency and has strategic flexibility.

[0042] It is understandable that building an intelligent control system that can dynamically adapt to different emission scenarios and traffic demands, and generating traffic operation and control strategies that take into account both low carbon and high efficiency through algorithm optimization, can improve the overall traffic sustainability of the city.

[0043] In the above implementation method, multi-source data is collected in real time to fully understand the spatiotemporal changes in traffic operation and carbon emissions. Carbon emission intensity is predicted in real time through deep learning models to identify high-emission areas. Traffic signals and path induction are precisely controlled through reinforcement learning and multi-objective optimization methods to achieve accurate identification and dynamic regulation of high-risk areas for urban carbon emissions. This not only improves the intelligence and accuracy of urban traffic management, but also provides effective technical support for achieving sustainable urban development.

[0044] Reference Figure 2 As an implementation method of step S102, the steps of performing spatiotemporal alignment on the pre-processed real-time monitoring data, extracting dynamic traffic features and static road features, and generating a spatiotemporal feature matrix include: Step S201, dividing the pre-processed real-time monitoring data into continuous data and discrete data; The pre-processed real-time monitoring data is categorized into continuous and discrete data. Continuous data generally refers to data with quantitative relationships that can be mathematically calculated, such as vehicle speed and traffic density. Discrete data typically includes data with categorical labels, such as vehicle type (fuel vehicle, electric vehicle) and weather conditions.

[0045] Step S202, Z-score normalization is performed on continuous data, and one-hot encoding is performed on discrete data; For continuous data, the Z-score normalization method can be used. This method eliminates the influence of different dimensions by converting the data to a standard normal distribution with zero mean and unit variance, ensuring that each dimension of the feature has the same contribution during training, ensuring that the model training is not biased towards features with larger values. For discrete data (such as vehicle type), the one-hot encoding method can be used to convert each category into a binary vector. This ensures that the discrete data can be correctly understood and processed by machine learning models such as neural networks, avoiding the bias that may be caused when categorical data is directly input into the model as numerical data.

[0046] Step S203, using the traffic signal controller clock as a reference, aligning the timestamps of each data source through Kalman filtering; Data from different sources (such as traffic flow, environmental data, and vehicle emissions data) may have certain delays or different time bases due to different collection devices. Therefore, Kalman filtering can be used to align the timestamps of different data sources. Kalman filtering is a recursive estimation method based on a state-space model. It can obtain optimal estimates from multiple data sources in the presence of noise. It can be used to compensate for time errors caused by signal delays between devices, ensuring that all data is synchronized based on the traffic signal controller's clock (which serves as a unified time base).

[0047] Step S204: Based on the time-space aligned data, the average idling time and acceleration variance of the vehicles are calculated according to the pre-divided road grids as dynamic traffic features; Average idling time refers to the duration of time vehicles in a specific area spend below a certain threshold (e.g., 5 km / h). This characteristic reflects the degree of traffic congestion. A high average idling time in an area indicates high traffic volume and the potential for severe congestion. Acceleration variance describes traffic flow stability by calculating the variance of vehicle acceleration (speed change) over time. A high variance indicates unstable traffic flow with large speed fluctuations, often associated with traffic accidents, sudden braking, and other situations.

[0048] Specifically, these dynamic features can be calculated using a sliding window statistical method (for example, using a 5-minute time window), and these features can be calculated for each grid based on a pre-divided road grid (such as a 500m×500m grid).

[0049] Step S205: extracting the road slope weight coefficient and grid adjacency relationship from the pre-stored high-precision map as static road features; The road slope weight coefficient indicates the significant impact of the slope on vehicle driving. The greater the slope, the greater the vehicle load, resulting in more fuel consumption and carbon emissions. The slope weight coefficient is usually calculated based on the road slope angle (θ), where θ is the slope angle. The common empirical coefficient is 6%, that is: W slope =1+0.06×|θ|. Grid adjacency can be constructed by analyzing the topological structure of the road grid and constructing a grid adjacency matrix. If two grids are adjacent, the matrix element is 1, otherwise it is 0. This adjacency matrix is used to describe the spatial relationship between road grids and helps analyze the spatial diffusion characteristics of traffic flow.

[0050] Step S206: combining the dynamic traffic features and static road features of each road grid into a complete feature vector; All extracted dynamic traffic features (such as idling time and acceleration variance) and static road features (such as slope weight coefficient and adjacency matrix) are combined to form a complete feature vector for each road grid. This feature vector comprehensively reflects the traffic status and road structure characteristics of the area and serves as the basic data input for the model.

[0051] Step S207 : Based on the complete feature vector, the three-dimensional spatiotemporal feature matrix is constructed by sorting according to the timestamp and the road grid ID.

[0052] Among them, the three-dimensional spatiotemporal feature matrix has three dimensions: time step, road grid ID and feature, forming a high-dimensional data representation to provide input for the subsequent carbon emission prediction model.

[0053] The above implementation effectively processes traffic data from various sensors and data sources, extracting both dynamic and static traffic characteristics. These characteristics are integrated into a high-dimensional spatiotemporal feature matrix, which serves as input for the carbon emissions prediction model. This technical solution not only enables real-time monitoring and prediction of traffic conditions but also adjusts traffic management strategies based on real-time data, thereby reducing traffic carbon emissions and achieving the goal of intelligent traffic management.

[0054] Reference Figure 3 As an implementation of the carbon emission prediction model, the training steps include: Step S301: Obtain the historical spatiotemporal feature matrix and the real carbon emission label data of the corresponding grid to construct a sample data set; Among them, the historical spatiotemporal feature matrix includes the dynamic traffic features and static road features of the historical road grid; Specifically, the method integrates static and dynamic features. Static features, such as road length, width, speed limit, and number of lanes, reflect road structural properties. Dynamic features, such as traffic volume, average speed, and road congestion index, reflect temporal fluctuations. Carbon emission labels are derived from traffic monitoring equipment or model estimation results to calibrate the prediction targets.

[0055] Step S302, dividing the sample data set into a training set, a validation set, and a test set; Splitting by time series prevents future information leakage, the validation set is used to monitor overfitting risks, and the test set measures generalization ability. For example, the data for a month can be divided into the first 20 days as the training set, the 21st to 25th days as the validation set, and the 26th to 30th days as the test set.

[0056] Step S303: Input the training set into the pre-built hybrid neural network model for training. The error between the carbon emission intensity prediction value and the actual carbon emission label data is calculated through the back propagation algorithm. The gradient is calculated layer by layer, and the weights and biases of the hybrid neural network model are updated. In one embodiment of this application, the hybrid neural network model structure consists of a graph neural network (GNN) and a long short-term memory (LSTM). The GNN uses an adjacency matrix to model spatial topology (such as the influence of upstream and downstream roads), while the LSTM processes time series information (such as changes in traffic flow over time). The outputs of the two are fused for prediction, and weights and biases can be optimized using error backpropagation and gradient descent.

[0057] Step S304, verifying the hybrid neural network model based on the validation set, evaluating the performance of the hybrid neural network model and adjusting the model hyperparameters according to the validation results; The validation set results are used to evaluate model performance indicators (such as MAE and RMSE) and adjust hyperparameters (such as learning rate, hidden layer dimension, and dropout rate). Grid search or Bayesian optimization can be used.

[0058] Step S305 , testing the prediction capability of the adjusted hybrid neural network model based on the test set to obtain a trained carbon emission prediction model.

[0059] Among them, completely unseen test set data is used to verify the model's adaptability to real scenarios, and finally a deployable model is generated.

[0060] The aforementioned implementation introduces a hybrid neural network model structure combining graph neural networks and long-short-term memory networks, enabling the coordinated modeling of spatial and temporal features for road carbon emissions prediction. Furthermore, a multi-objective joint training strategy and a validation-testing hierarchical evaluation process enhance the model's generalization performance and deployment stability. Ultimately, this model can be deployed on edge computing devices to achieve high-precision, low-latency predictions of traffic carbon emissions, providing strong technical support for urban green transportation management.

[0061] Reference Figure 4 As an implementation of step S303, the step of pre-building a hybrid neural network model includes: Step S401, obtaining an adjacency matrix in static road features of a historical road grid; Specifically, in urban road networks, the structural relationship of roads is one of the key factors influencing traffic behavior prediction. The adjacency matrix is a basic representation of graph structure, which can encode the topological relationship between roads in matrix form. Each road grid can be regarded as a node in the graph, and the physical or logical connection between roads is reflected in the edge weight information in the adjacency matrix. If there is a connection between two nodes, the corresponding position is 1, otherwise it is 0. Historical static road characteristics such as road length, number of lanes, and intersection density do not change over time and are an important basis for forming adjacency relationships. For example, if there is a main road connecting two grids, the corresponding element of the adjacency matrix is set to 1, otherwise it is 0. In this way, the spatial structure of the road grid can be clearly modeled, providing a structural prior for the subsequent construction of graph neural networks.

[0062] Step S402, adding self-connection to the adjacency matrix to generate an extended adjacency matrix, and calculating the degree matrix of the extended adjacency matrix; Specifically, in traditional graph neural networks, if self-loops are not added, the node's own information will not be included in the aggregation calculation, resulting in the inability to preserve the original node state during feature updates. By adding the identity matrix to the adjacency matrix, it is ensured that the node can consider both its own information and the information of its neighboring nodes during graph convolution operations. Subsequently, by calculating the degree matrix corresponding to the expanded adjacency matrix, the total number of connections for each node is counted as the degree value (in diagonal matrix form) for subsequent graph convolution normalization operations.

[0063] Step S403: defining an input layer structure based on the dynamic traffic feature dimension of the historical road grid, constructing a hidden layer comprising two layers of long short-term memory network units, with a preset number of neurons in each layer; Dynamic traffic features refer to time-varying traffic status information, such as historical vehicle speeds, traffic volume, and congestion levels, exhibiting typical temporal characteristics. To capture this temporal dependency, the input layer establishes data channels based on these feature dimensions, typically of the dimensions T×N×F, where T is the time step, N is the number of nodes, and F is the feature dimension of each node. Subsequently, a two-layer long short-term memory (LSTM) network is employed as the hidden layer. LSTM effectively mitigates the vanishing gradient problem of traditional RNNs during long sequence training. It controls information flow through input, forget, and output gates, enabling the model to learn long-term dependencies. Each layer is configured with a number of neurons (e.g., 64 or 128) to enhance feature extraction.

[0064] Step S404: configuring the activation function of the long short-term memory network unit and outputting the temporal context feature vector; Specifically, the LSTM module uses activation functions such as Sigmoid and Tanh in its internal calculations, controlling the gating mechanism and updating the cell state, respectively. This design allows information at each time step to be selectively retained or forgotten over multiple cycles. Ultimately, the hidden state outputs a contextual feature vector representing the time series trend. This vector is a high-level abstraction of historical traffic dynamics, encompassing time series statistical features such as speed variability and the frequency of traffic pattern changes.

[0065] Step S405, combining the extended adjacency matrix and the degree matrix into a normalized adjacency matrix; Specifically, graph convolution operations in graph neural networks require weighted aggregation of node information. Directly using an unnormalized adjacency matrix can lead to numerical instability and training difficulties. Therefore, a symmetric normalized adjacency matrix is constructed. Symmetric normalization on both sides of the matrix means that information aggregation takes into account both the connection strength of the sending node and the connection density of the receiving node, thereby achieving more stable graph representation learning.

[0066] Step S406: construct a graph convolution layer based on the normalized adjacency matrix, configure the trainable weight matrix and activation function of the graph convolution layer, and output a spatial topological feature vector; The graph convolution layer generalizes traditional convolution to graph structures. Its core is to propagate features through adjacency relationships. This operation can be viewed as a weighted fusion and transformation of neighboring node features. Finally, nonlinearity is introduced through activation functions such as ReLU. Stacking multiple graph convolution layers can extract deeper spatial topological relationships, such as identifying the potential impact of important intersections or bridge nodes on carbon emissions.

[0067] Step S407, concatenating the temporal context feature vector and the spatial topology feature vector to obtain a fused feature vector; Temporal and spatial features characterize the state of the transportation system from different perspectives. The former reflects dynamic trends, while the latter reflects topological structure and neighborhood influences. Concatenating the two into a fused feature vector is a key step in achieving spatiotemporal collaborative modeling. This concatenation preserves the integrity of both features, enabling subsequent neural networks to extract correlated features within a unified representation space. For example, carbon emissions on certain road sections surge during peak hours as traffic capacity decreases in adjacent areas.

[0068] Step S408, performing dimensionality reduction processing on the fused feature vector through a fully connected layer to generate a joint feature vector; Because the concatenated features have high dimensionality, directly inputting them into the prediction layer can lead to parameter redundancy and overfitting. Therefore, a fully connected layer is required to reduce their dimensionality. This dimensionality reduction not only compresses the feature space but also strengthens the nonlinear interactions between different features. This processing preserves the most representative composite information, making it a key input for constructing output predictions.

[0069] Step S409: define probability distribution parameters for the fully connected layer weights, and initialize the mean and standard deviation of the probability distribution parameters; To enhance the interpretability and credibility of the model output, we employ the principles of Bayesian neural networks and define the weights of the fully connected layers as probabilistic variables. The mean is initialized using a Xavier normal distribution to maintain consistent input and output variances and prevent exploding or vanishing gradients. The standard deviation is initialized to a small constant, such as 0.01, to ensure initial model stability. This initialization scheme helps stabilize the model's convergence trajectory during the initial stages of training and lays the foundation for estimating confidence intervals for the model's predictions.

[0070] Step S410: construct an output layer mapping relationship, generate a complete model architecture including predicted values and confidence intervals, and obtain a hybrid neural network model.

[0071] The output layer constructs a complete prediction architecture by mapping the joint feature vector to the target space (i.e., the predicted carbon emissions value and its confidence interval). Using methods such as Gaussian distribution prediction (mean + variance), the model not only outputs the expected value but also quantifies its uncertainty, providing confidence intervals for the results. Ultimately, the model integrates LSTM and GNN structures, resulting in a spatiotemporal hybrid neural network model.

[0072] The above implementation integrates static road topology with dynamic traffic time series characteristics to construct a hybrid neural network model based on a graph convolutional network and a long short-term memory network, achieving accurate prediction of road carbon emissions. This model not only captures the spatial dependencies between road grids but also provides a deep understanding of the dynamic patterns of traffic characteristics over time. Furthermore, it incorporates probability distribution modeling to provide prediction confidence intervals, significantly improving the accuracy, robustness, and interpretability of the predictions.

[0073] Reference Figure 5 As an implementation of step S106, based on the real-time traffic status data and the prediction result matrix of the high-emission area, a multi-objective optimization function is constructed and solved to obtain the optimized control strategy for the high-emission area, including the following steps: Step S501, obtaining real-time traffic status data and a prediction result matrix of high emission areas; Among them, real-time traffic status data includes vehicle speed, density and real-time traffic light status; the prediction result matrix includes road grid identification, prediction value and confidence interval; It is understandable that through real-time traffic status data (such as vehicle speed, density and traffic light status), the system can perceive the current traffic conditions; and the prediction matrix of high-emission areas (including road grid identification, emission prediction values and confidence intervals) provides a quantitative basis for the spatial distribution of carbon emissions.

[0074] Step S502, extracting a set of intersections associated with the road grid of the high-emission area and a set of vehicles heading towards the high-emission area; The intersection set refers to all intersections connected to the road network within high-emission zones. These intersections are important traffic flow control points. By analyzing the road network topology, the system can determine which intersections are closely associated with carbon emissions in high-emission zones. The vehicle set refers to all vehicles about to enter high-emission zones. By monitoring real-time traffic data, the system can identify vehicles heading into high-emission zones and target them for control.

[0075] By mapping the grids to road topology, the system extracts key control points (intersections) directly associated with high-emission grids and the vehicle groups approaching these areas. This is the prerequisite for localized control and directional guidance. For example, if the intersections associated with area A12 are X1, X2, and X3, and vehicles currently approaching this area include V101 through V120, the system will target these intersections as control targets for the optimization strategy.

[0076] Step S503: Based on the intersection set and the vehicle set, define an objective function for minimizing the total carbon emissions of the region, and add traffic delay upper limit, signal cycle boundary and path connectivity constraints to the objective function; Specifically, the objective function is designed based on the relationship between traffic flow and carbon emissions. By defining an objective function that minimizes carbon emissions, the system can minimize carbon emissions when controlling traffic lights and route guidance. The objective function typically considers the relationship between vehicle carbon emissions and travel time, as well as the impact of different control measures on emissions.

[0077] For example, if vehicles take too long to pass through a particular intersection, this results in higher emissions. However, by adjusting the traffic light cycle and the green-to-signal ratio, vehicle efficiency can be improved, thereby reducing carbon emissions. Therefore, the objective function not only considers carbon emissions, but also minimizes them while maximizing traffic flow.

[0078] Furthermore, to achieve this objective function, certain constraints must be imposed to ensure traffic stability and realism. These constraints include: maximum intersection delays cannot exceed a certain value, signal cycles must be kept within a reasonable range, and routes must maintain road network connectivity. These constraints ensure that the system doesn't simply ignore the normal operation of traffic while optimizing carbon emissions, but rather that traffic flow remains stable while reducing emissions.

[0079] Step S504: combining the prediction result matrix and the real-time traffic light state into a state vector and performing normalization processing; Specifically, in traffic optimization control, the state vector represents a specific description of the current traffic and emissions state. This state vector combines predicted carbon emissions with real-time traffic light status to form a multidimensional state description. This allows the system to fully understand the current traffic and carbon emissions status, enabling it to optimize traffic light control strategies.

[0080] For example, a state vector might contain information such as carbon emission predictions for multiple intersections, current traffic light status, and green-to-signal ratios. To make this information suitable for reinforcement learning algorithms, it's necessary to normalize the data. This involves mapping data from different ranges to a unified numerical range to facilitate machine learning model training.

[0081] Step S505: generating a signal control action and a path induction action based on a proximal strategy optimization algorithm; Specifically, the policy is updated using the Proximal Policy Optimization (PPO) algorithm. PPO is a reinforcement learning algorithm that continuously optimizes its decision-making strategy through interactions with the system within the environment. After each interaction, the algorithm updates the parameters of the policy network based on a reward function. This reward function is typically a weighted function of carbon emission reductions and traffic delay reductions, aiming to reduce emissions while maintaining smooth traffic flow.

[0082] In the embodiments of this application, a reinforcement learning algorithm dynamically controls traffic signals, adjusting the signal cycle and green-to-signal ratio based on real-time traffic conditions, thereby continuously optimizing the balance between traffic flow and carbon emissions. This algorithm has significant advantages in adaptability and flexibility, and can cope with complex and dynamic traffic environments.

[0083] Step S506, defining a signal control action for the intersection set, and encoding the signal control action into a signal light control instruction; Among them, the signal control action includes the period adjustment amount and the green-to-signal ratio adjustment amount; Specifically, these signal control actions are derived based on an analysis of traffic conditions and carbon emission forecasts. The goal is to maximize traffic flow while simultaneously reducing carbon emissions by optimizing the signal cycle and green-to-signal ratio. Once these control actions are defined, they need to be encoded into a format suitable for parsing and execution by the signal controller. For example, control instructions can use JSON format to represent adjustments to the signal cycle and green-to-signal ratio. This allows for standardized and modular control instructions, facilitating real-time regulation and management of the system.

[0084] Step S507, defining a path guidance action for the vehicle set and specifying a target node selection range; In the optimization control strategy, route guidance directs vehicle routes to avoid high-emission areas or alleviate traffic pressure within them. By setting a range of target nodes that vehicles can choose from, the system dynamically adjusts vehicle routes to avoid excessive pollution or congestion caused by overly concentrated traffic flows.

[0085] In practice, the system calculates the optimal detour route based on factors such as carbon emissions in high-emission areas, traffic flow, and signal control status, and assigns a target node to each vehicle. The system also ensures the connectivity of the detour route with the road network to avoid excessive detours that increase traffic burdens.

[0086] Step S508 , calculating a detour path for the vehicle based on the shortest path algorithm according to the target node output by the path guidance action, and encapsulating the detour path into a vehicle path guidance instruction; Based on the previously defined target node, the system calculates a detour using a modified shortest path algorithm. To achieve low carbon emissions and minimize delays, carbon emission intensity and path length are factored into the cost function. This approach ensures that the detour not only avoids high-emission areas but also keeps the route as short as possible, reducing travel time and fuel consumption. The calculated detour is encapsulated as route guidance instructions and transmitted in a standard format for easy parsing and execution by the vehicle terminal.

[0087] Step S509 : combining the signal light control instruction and the vehicle path guidance instruction to obtain an optimized control strategy for the high emission area.

[0088] Among them, all generated signal control instructions and path induction instructions will be combined into a complete set of optimization control strategies, covering the entire high-emission area. These instructions will be sent to traffic signal controllers and vehicle terminals in real time to perform optimization control.

[0089] In this implementation, a reinforcement learning algorithm is employed to dynamically optimize traffic signals and route guidance, combining real-time traffic data with predictions of high-emission areas. This approach reduces carbon emissions while ensuring smooth traffic flow. By defining precise control actions and optimization targets, the entire system effectively achieves the dual goals of low carbon emissions and high-efficiency transportation, making it suitable for intelligent transportation and smart city development.

[0090] Reference Figure 6 As a further implementation of the urban road carbon emission prediction and optimization control method, after obtaining the optimization control strategy for high emission areas, the method further includes: Step S601: Send the optimized control strategy to the roadside execution unit in the high emission area; Among them, the optimization control strategy includes signal light control instructions and vehicle path guidance instructions; Specifically, traffic light control instructions are usually sent to the signal controllers at the associated intersections, which control parameters such as the traffic light cycle and green-to-signal ratio at the intersections; while vehicle path induction instructions are sent to the on-board terminal through vehicle-road cooperative communication technology, informing the vehicle of the path or detour route to be selected when passing through high-emission areas.

[0091] For example, in a real-world application, suppose the traffic light cycle in a high-emission area needs to be adjusted from the current 120 seconds to 100 seconds, and vehicles need to detour from Intersection A to Intersection B. The traffic light control command will indicate the adjustment of the traffic light cycle at the corresponding intersection, and the route guidance command will inform vehicles heading into the area how to choose a detour route.

[0092] Step S602: Obtain a prediction result matrix generated by a carbon emission prediction model based on real-time traffic status data and an optimized control strategy; The currently implemented optimized control strategy parameters and real-time traffic status data are input into the prediction model to generate emission prediction results that strictly match the strategy execution conditions, providing a homologous reference benchmark for subsequent deviation analysis. Step S603, collecting actual traffic flow data and vehicle emission data after the optimization control strategy is executed; To evaluate the effectiveness of optimized control strategies, real-time traffic flow and vehicle emissions data must be collected. Actual traffic flow data typically includes lane-level vehicle density and speed, reflecting the smoothness of traffic flow. Vehicle emissions data is collected through the vehicle's onboard diagnostics (OBD) system, which collects engine operating parameters such as fuel consumption and engine load. Roadside pollutant concentration sensors can also monitor air pollutant concentrations (such as NOx and CO2) in real time. This information helps assess actual carbon emissions.

[0093] Step S604, calculating a deviation matrix between the predicted values and the actual values based on the predicted values in the prediction result matrix and the actual traffic flow data and vehicle emission data; The deviation between the predictions and actual values is calculated by comparing the traffic emission prediction model's results with actual traffic flow data and vehicle emissions data. The calculation of the deviation matrix primarily involves three aspects: absolute deviation, relative deviation rate, and confidence weight. Absolute deviation refers to the direct difference between the predicted value and the actual value. The relative deviation rate evaluates the magnitude of the prediction error relative to the actual value by calculating the ratio of the absolute deviation to the actual value. The confidence weight is calculated based on the width of the confidence interval of the prediction result and reflects the reliability of the model's prediction. The higher the confidence level, the greater the weight.

[0094] Step S605 , performing parameter correction on the carbon emission prediction model according to the deviation matrix to obtain an updated carbon emission prediction model.

[0095] When the deviation matrix shows a large deviation between the prediction and the actual, the carbon emission prediction model needs to be calibrated. In order to improve the prediction accuracy, the Bayesian update method can be used to calibrate the model parameters.

[0096] Specifically, the Bayesian method uses prior knowledge and observed data to calculate the posterior distribution and update model parameters. The specific calculation formula is: p(θ|B)∝p(B|θ)⋅p(θ), where θ is the model parameter and B is the bias matrix. p(θ|B) represents the posterior probability distribution of the model parameter θ given the bias matrix B, p(B|θ) represents the probability of observing the bias matrix B under the parameters θ, and p(θ) is the prior distribution of the parameter θ. The Bayesian updating method can dynamically adjust the parameters of the carbon emission prediction model based on the information in the bias matrix, thereby optimizing the model's predictive ability.

[0097] In this implementation, the deviation between the predicted results and actual data is used to calibrate the model, ensuring the accuracy of the carbon emission prediction model. This closed-loop optimization method not only optimizes carbon emission control but also ensures smooth traffic flow, ultimately providing strong technical support for smart transportation and green travel.

[0098] The embodiments of the present application also disclose an urban road carbon emission prediction and optimization control system.

[0099] An urban road carbon emission prediction and optimization control system, the system comprising: A data monitoring and processing module is used to obtain real-time monitoring data of the road to be inspected and perform preprocessing; wherein the real-time monitoring data includes real-time traffic flow data, environmental data and vehicle emission data; The spatiotemporal feature matrix generation module is used to perform spatiotemporal alignment on the pre-processed real-time monitoring data, extract dynamic traffic features and static road features, and generate a spatiotemporal feature matrix; The carbon emission prediction module is used to input the spatiotemporal feature matrix into the pre-trained carbon emission prediction model to obtain the prediction result matrix of the road to be tested; A carbon emission intensity identification module is used to identify road grids whose carbon emission intensity prediction values exceed a preset threshold as high-emission areas based on the prediction result matrix; Traffic status acquisition module, used to obtain real-time traffic status data in high-emission areas; The optimization control module is used to construct and solve a multi-objective optimization function based on real-time traffic status data and the prediction result matrix of high-emission areas to obtain the optimized control strategy for high-emission areas.

[0100] An urban road carbon emission prediction and optimization control system in an embodiment of the present application can implement any of the above-mentioned urban road carbon emission prediction and optimization control methods, and the specific working processes of each module in the urban road carbon emission prediction and optimization control system can refer to the corresponding processes in the above-mentioned method embodiments.

[0101] In the several embodiments provided in this application, it should be understood that the provided methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for example, the division of a module is merely a logical functional division, and in actual implementation, other division methods may be used, such as combining or integrating multiple modules into another system, or ignoring or not implementing certain features.

[0102] The embodiment of the present application also discloses a computer device.

[0103] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for predicting and optimizing carbon emissions from urban roads as described above is implemented.

[0104] The embodiment of the present application also discloses a computer-readable storage medium.

[0105] A computer-readable storage medium stores a computer program that can be loaded by a processor and executed by any one of the above-mentioned methods for predicting and optimizing carbon emissions from urban roads.

[0106] Among them, computer-readable storage media can be any tangible medium that contains or stores a program that can be used by or in combination with an instruction execution system, apparatus or device; the program code contained on the computer-readable medium can be transmitted using any appropriate medium, including but not limited to wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0107] It should be noted that, in the above embodiments, the description of each embodiment has different emphases. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0108] The above are all preferred embodiments of the present application and are not intended to limit the scope of protection of this application. Unless otherwise specified, any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features. In other words, unless otherwise specified, each feature is merely an example of a series of equivalent or similar features.

Claims

1. A method for predicting and optimizing carbon emissions from urban roads, characterized in that: The method comprises: Acquire real-time monitoring data of the road to be inspected and perform preprocessing; wherein the real-time monitoring data includes real-time traffic flow data, environmental data, and vehicle emission data; Performing spatiotemporal alignment on the pre-processed real-time monitoring data, extracting dynamic traffic features and static road features, and generating a spatiotemporal feature matrix; Inputting the spatiotemporal feature matrix into a pre-trained carbon emission prediction model to obtain a prediction result matrix for the road to be detected; Based on the prediction result matrix, identifying road grids where the predicted carbon emission intensity exceeds a preset threshold as high emission areas; Obtaining real-time traffic status data of the high-emission area; Based on the real-time traffic status data and the prediction result matrix of the high-emission area, a multi-objective optimization function is constructed and solved to obtain an optimized control strategy for the high-emission area.

2. The urban road carbon emission prediction and optimization control method according to claim 1 is characterized in that: The steps of performing spatiotemporal alignment on the pre-processed real-time monitoring data, extracting dynamic traffic features and static road features, and generating a spatiotemporal feature matrix include: Dividing the pre-processed real-time monitoring data into continuous data and discrete data; Perform Z-score normalization on the continuous data and one-hot encoding on the discrete data; Using the traffic signal controller clock as a benchmark, the timestamps of each data source are aligned through Kalman filtering; Based on the time-space aligned data, the average vehicle idling time and acceleration variance are calculated according to the pre-divided road grid as dynamic traffic characteristics; Extract road slope weight coefficient and grid adjacency relationship from pre-stored high-precision maps as static road features; The dynamic traffic features and static road features of each road grid are spliced into a complete feature vector; Based on the complete feature vectors, they are sorted according to timestamps and road grid IDs to construct a three-dimensional spatiotemporal feature matrix.

3. The urban road carbon emission prediction and optimization control method according to claim 1 is characterized in that: The method further includes a step of training a carbon emission prediction model, wherein the training step includes: Obtain a historical spatiotemporal feature matrix and real carbon emission label data of the corresponding grid to construct a sample data set; wherein the historical spatiotemporal feature matrix includes dynamic traffic characteristics and static road characteristics of the historical road grid; Dividing the sample data set into a training set, a validation set, and a test set; Inputting the training set into a pre-built hybrid neural network model for training, calculating the error between the carbon emission intensity prediction value and the actual carbon emission label data through a backpropagation algorithm, calculating the gradient layer by layer, and updating the weights and bias of the hybrid neural network model; Validating the hybrid neural network model based on the validation set, evaluating the performance of the hybrid neural network model and adjusting model hyperparameters according to the validation results; The prediction capability of the adjusted hybrid neural network model is tested based on the test set to obtain the trained carbon emission prediction model.

4. The urban road carbon emission prediction and optimization control method according to claim 3 is characterized in that: The steps of pre-building the hybrid neural network model include: Obtain the adjacency matrix of the static road features of the historical road grid; Adding self-connection to the adjacency matrix to generate an extended adjacency matrix, and calculating the degree matrix of the extended adjacency matrix; The input layer structure is defined based on the dynamic traffic feature dimensions of the historical road grid, and a hidden layer consisting of two layers of long short-term memory network units is constructed, with a preset number of neurons in each layer. Configuring an activation function of the long short-term memory network unit to output a temporal context feature vector; Combining the extended adjacency matrix and the degree matrix into a normalized adjacency matrix; Constructing a graph convolution layer based on the normalized adjacency matrix, configuring a trainable weight matrix and an activation function of the graph convolution layer, and outputting a spatial topological feature vector; Concatenating the temporal context feature vector and the spatial topology feature vector to obtain a fused feature vector; Performing dimensionality reduction processing on the fused feature vector through a fully connected layer to generate a joint feature vector; Define probability distribution parameters for the fully connected layer weights and initialize the mean and standard deviation of the probability distribution parameters; An output layer mapping relationship is constructed to generate a complete model architecture including predicted values and confidence intervals to obtain the hybrid neural network model.

5. The urban road carbon emission prediction and optimization control method according to claim 1 is characterized in that: The steps of constructing and solving a multi-objective optimization function based on the real-time traffic status data and the prediction result matrix of the high-emission area to obtain an optimized control strategy for the high-emission area include: Obtain real-time traffic status data and prediction result matrix of high emission areas; Extracting a set of intersections associated with a road grid in the high-emission area and a set of vehicles traveling toward the high-emission area; Based on the intersection set and the vehicle set, an objective function of minimizing regional total carbon emissions is defined, and traffic delay upper limit, signal cycle boundary and path connectivity constraints are added to the objective function; Combining the prediction result matrix and the real-time traffic light state into a state vector and performing normalization processing; Generate signal control actions and path induction actions based on proximal policy optimization algorithm; Defining a signal control action for the intersection set and encoding the signal control action into a signal light control instruction; wherein the signal control action includes a cycle adjustment amount and a green-to-signal ratio adjustment amount; defining a path induction action for the vehicle set and specifying a target node selection range; According to the target node output by the path induction action, a vehicle detour path is calculated based on a shortest path algorithm and encapsulated into a vehicle path induction instruction; The signal light control instructions and the vehicle path guidance instructions are combined to obtain an optimized control strategy for the high emission area.

6. The urban road carbon emission prediction and optimization control method according to claim 5 is characterized in that: After obtaining the optimized control strategy for the high emission area, the method further includes: Sending the optimized control strategy to the roadside execution unit in the high-emission area; wherein the optimized control strategy includes signal light control instructions and vehicle path guidance instructions; Obtaining a prediction result matrix generated by the carbon emission prediction model based on real-time traffic status data and the optimization control strategy; Collecting actual traffic flow data and vehicle emission data after the optimization control strategy is executed; Calculating a deviation matrix between predicted values and actual values based on the predicted values in the prediction result matrix and the actual traffic flow data and vehicle emission data; Parameters of the carbon emission prediction model are corrected according to the deviation matrix to obtain an updated carbon emission prediction model.

7. An urban road carbon emission prediction and optimization control system, characterized in that: The system comprises: A data monitoring and processing module is used to obtain real-time monitoring data of the road to be inspected and perform preprocessing; wherein the real-time monitoring data includes real-time traffic flow data, environmental data and vehicle emission data; A spatiotemporal feature matrix generation module is used to perform spatiotemporal alignment on the pre-processed real-time monitoring data, extract dynamic traffic features and static road features, and generate a spatiotemporal feature matrix; A carbon emission prediction module, configured to input the spatiotemporal feature matrix into a pre-trained carbon emission prediction model to obtain a prediction result matrix for the road to be inspected; A carbon emission intensity identification module, configured to identify, based on the prediction result matrix, road grids whose predicted carbon emission intensity values exceed a preset threshold as high emission areas; A traffic status acquisition module, configured to acquire real-time traffic status data of the high-emission area; The optimization control module is used to construct and solve a multi-objective optimization function based on the real-time traffic status data and the prediction result matrix of the high-emission area to obtain an optimization control strategy for the high-emission area.

8. A computer device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 6 when executing the program.

9. A computer-readable storage medium, characterized in that: A computer program is stored which can be loaded by a processor and execute the method according to any one of claims 1 to 6.

Images