A highway carbon emission simulation and deduction system based on digital twinning

Abstract

The application relates to the field of highway emission reduction technology and discloses a highway carbon emission simulation deduction system based on digital twinning. The system comprises a carbon emission data acquisition layer, a digital twinning modeling layer, a multi-dimensional carbon emission calculation layer, a simulation deduction optimization layer and an execution feedback adjustment layer. The carbon emission data acquisition layer obtains multi-source traffic parameters through a distributed sensing network, generates a dynamic carbon emission factor matrix and performs sensitivity grading; the digital twinning modeling layer constructs a road network twin body, generates a road three-dimensional topological structure and superimposes a vehicle energy consumption model to form a dynamic twinning scene; the multi-dimensional carbon emission calculation layer establishes a space-time mapping relationship, integrates emission data to generate a road section level carbon emission intensity atlas; the simulation deduction optimization layer converts the atlas into a set of control strategies, predicts a carbon emission change trend and outputs a Pareto optimal strategy combination; and the execution feedback adjustment layer monitors data, calculates a deviation rate, generates an adaptation index and dynamically corrects model parameters until the index is stable.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of highway emission reduction technology, in particular to a highway carbon emission simulation and deduction system based on digital twinning. BACKGROUND

[0002] Current highway carbon emission control faces many challenges. The dynamic change characteristics of traffic flow make it difficult to accurately capture real-time carbon emission data. Traditional data collection methods rely on fixed monitoring points, which have limited coverage and are difficult to integrate vehicle driving state, road environment and other multi-source information, resulting in large errors in the calculation of carbon emission factors.

[0003] In terms of modeling, existing road network models are mostly static or semi-static, and cannot real-time integrate high-precision maps and real-time traffic data, making it difficult to truly reflect the dynamic changes of road three-dimensional topological structure and vehicle energy consumption, resulting in a large deviation between carbon emission simulation scenarios and actual road conditions.

[0004] In terms of carbon emission calculation, the emission data of mobile sources and fixed sources are often separated from each other, lacking effective spatio-temporal mapping relationship, and unable to form an accurate road section-level carbon emission intensity map, making it difficult to support fine-grained carbon emission control.

[0005] In the simulation and deduction process, traditional methods mostly develop control strategies based on experience, lack data-driven scientific analysis, and are difficult to predict the carbon emission trend under different strategies, resulting in insufficient pertinence and effectiveness of control measures. At the same time, the feedback mechanism after the implementation of the strategy is not perfect, and the model parameters cannot be corrected in time, so that the deviation between the simulation results and the actual situation gradually increases, affecting the continuous optimization of the control effect.

[0006] Different factors have different degrees of influence on carbon emissions, and existing technologies do not classify carbon emission factors by sensitivity, making it difficult to highlight key influencing factors in the modeling and calculation process, resulting in resource waste and low efficiency. These problems jointly restrict the accuracy and efficiency of highway carbon emission control, and a new system is needed to break through the limitations of existing technologies. SUMMARY

[0007] The purpose of the present application is to provide a highway carbon emission simulation and deduction system based on digital twinning to solve the problems raised in the background art.

[0008] To achieve the above purpose, the present application provides a highway carbon emission simulation and deduction system based on digital twinning, which comprises:

[0009] A carbon emission data acquisition layer acquires multi-source traffic parameters through a distributed sensing network, processes vehicle driving state data, road environment data and meteorological data using a spatio-temporal fusion algorithm, generates a dynamic carbon emission factor matrix, and classifies the carbon emission factor matrix by sensitivity.

[0010] The digital twin modeling layer constructs a road network twin based on the sensitivity classification results, integrates high-precision maps and real-time traffic flow data, generates a three-dimensional road topology through multi-scale grid division, and overlays a vehicle energy consumption model to form a dynamic twin scene.

[0011] A multi-dimensional carbon emission calculation layer establishes a spatiotemporal mapping relationship between twin scenarios and carbon emission factors, uses an event-triggered mechanism to synchronize traffic flow status, integrates mobile source and stationary source emission data through a carbon emission accounting model, and generates a road segment-level carbon emission intensity map.

[0012] The simulation and optimization layer transforms the carbon emission intensity map into a set of control strategies based on reinforcement learning algorithms. It then uses Monte Carlo simulation to predict the carbon emission change trend under different strategies and outputs the Pareto optimal strategy combination.

[0013] The execution feedback adjustment layer monitors carbon emission data in real time after the implementation of the control strategy, calculates the deviation rate between the actual carbon emission and the simulation prediction value, generates a strategy fit index, and dynamically corrects the parameters of the carbon emission accounting model until the strategy fit index reaches a stable threshold.

[0014] Preferably, the method for acquiring multi-source traffic parameters through a distributed sensor network includes:

[0015] Three types of data acquisition terminals are deployed along the highway: millimeter-wave radar, video surveillance equipment, and environmental sensors. These terminals collect data on vehicle speed, traffic flow, vehicle type classification, road surface temperature, and real-time wind direction and speed. Multi-source traffic parameters include dynamic traffic flow data, static road attribute data, and meteorological interference data.

[0016] An adaptive sampling mechanism is adopted to dynamically adjust the data collection frequency according to traffic flow density. Based on historical congestion data, toll station entrances and exits, bridge and tunnel areas, and interchange areas are marked as high-frequency collection areas. Straight road sections are marked as low-frequency collection areas. A sampling frequency function is established to divide high-frequency and low-frequency collection areas by traffic flow density threshold. High-frequency collection areas use millisecond-level sampling intervals, while low-frequency collection areas use second-level sampling intervals.

[0017] The Kalman filter algorithm is used to correct noise interference in multi-source traffic parameters in real time. The input of the Kalman filter algorithm is the original collected data, and the output is the denoised traffic parameters. The deployment location of the collection terminals is dynamically adjusted according to the denoised data. If a new high-congestion area is detected, a new collection terminal is added to the area. If the traffic flow in the area covered by the original collection terminal is continuously lower than the threshold, the number of collection terminals in the area is reduced. The collected data from different terminals is integrated through a data synchronization protocol to obtain complete multi-source traffic parameters.

[0018] Preferably, the method for processing vehicle driving status data, road environment data, and meteorological data using the spatiotemporal fusion algorithm to generate a dynamic carbon emission factor matrix includes:

[0019] Multi-source traffic parameters are normalized, and the normalized parameters are arranged into a three-dimensional array according to time series and spatial location. The dimensions of the three-dimensional array include road segment number, collection time and parameter type. A spatiotemporal fusion network structure is constructed, which includes a temporal convolution module, a spatial attention module and a feature splicing layer.

[0020] A three-dimensional array is input into a spatiotemporal fusion network structure. In the temporal convolution module, a convolutional neural network is used to extract temporal features from the input data, generating temporal feature sequences of different durations, including short-term, medium-term, and long-term feature sequences. In the spatial attention module, a self-attention mechanism is used to perform weighted fusion of spatial features. A feature interaction channel is set between the temporal convolution module and the spatial attention module to concatenate the temporal and spatial features of the same road segment.

[0021] A fully connected layer is applied to the output layer of the spatiotemporal fusion network structure to compress the dimension, transforming the high-dimensional feature vector of the network output into a two-dimensional matrix, thereby generating a dynamic carbon emission factor matrix.

[0022] Preferably, the method for sensitivity classification of the carbon emission factor matrix includes:

[0023] The sensitivity coefficient of each element in the carbon emission factor matrix is ​​calculated by performing partial derivative calculations on vehicle speed data, road surface temperature data, and wind direction and speed data from multiple traffic parameters.

[0024] A first threshold and a second threshold for sensitivity coefficients are preset. The sensitivity coefficient is compared with the first threshold and the second threshold, respectively. If the sensitivity coefficient is less than the first threshold, the parameter corresponding to the element is marked as a low sensitivity factor; if the sensitivity coefficient is greater than the first threshold and less than the second threshold, the parameter corresponding to the element is marked as a medium sensitivity factor; if the sensitivity coefficient is greater than the second threshold, the parameter corresponding to the element is marked as a high sensitivity factor.

[0025] Different labels are used to classify the sensitivity of the carbon emission factor matrix. Different labels represent different sensitivity levels. Low-sensitivity factors are defined as routine monitoring items, medium-sensitivity factors as key monitoring items, and high-sensitivity factors as core monitoring items.

[0026] Preferably, the method for generating the three-dimensional topology of the road includes:

[0027] Based on the sensitivity classification results, high-precision road baseline data is loaded into the digital twin platform. This high-precision data includes road design parameters, pavement material properties, and the distribution of roadside facilities. Differentiated modeling precision is applied to parameters with different sensitivity levels: centimeter-level precision for high-sensitivity factors, decimeter-level precision for medium-sensitivity factors, and meter-level precision for low-sensitivity factors. A grid subdivision algorithm is applied to refine the road model, and terrain elevation data is introduced to correct the road's 3D coordinates. The grid density is continuously adjusted through iterative optimization algorithms, stopping when the model error is less than a preset threshold. A 3D road topology containing lane lines, medians, and traffic signs is constructed using 3D modeling software, ultimately generating a 3D road topology marked with sensitivity levels.

[0028] Preferably, the method for establishing the spatiotemporal mapping relationship between twin scenarios and carbon emission factors includes:

[0029] A unified scene coordinate system is defined with the starting point of the highway as the origin, the X-axis parallel to the road centerline, the Y-axis perpendicular to the road cross-section, and the Z-axis representing the elevation direction. A GPS positioning device is used to calibrate the data acquisition terminal and obtain its position coordinates. A coordinate transformation algorithm is used to transform the parameter coordinates of the dynamic carbon emission factor matrix to the scene coordinate system, thereby obtaining the carbon emission factor distribution in the scene coordinate system. Using the origin of the scene coordinate system as a reference, the digital twin scene is aligned to obtain its absolute coordinates.

[0030] The three-dimensional topology of the road is transformed into the scene coordinate system through a coordinate mapping algorithm, thereby obtaining a twin scene model in the scene coordinate system. In the scene coordinate system, the carbon emission factor distribution and the twin scene model are spatiotemporally registered to establish a spatial dimension mapping relationship. The timestamp of the acquisition terminal is used as the reference clock, and the time axis of the twin scene is unified through a time synchronization protocol to establish a time dimension mapping relationship.

[0031] Preferably, the method for integrating mobile and stationary emission data through a carbon emission accounting model includes:

[0032] Based on the carbon emission factor distribution in the scene coordinate system and the twin scene model, an emission source network is constructed, and a carbon emission accounting model based on graph neural network is used to integrate the emission data of mobile sources and stationary sources. The carbon emission accounting model includes an input layer, a feature extraction layer, a source term fusion layer, a spatiotemporal correction layer, and an output layer. The emission source network is used as the input of the input layer of the carbon emission accounting model, and a road segment-level carbon emission intensity map is generated through the output layer.

[0033] Preferably, the method for constructing the emission source network includes:

[0034] Each vehicle in the mobile source emission data under the scene coordinate system is treated as a mobile node, and the instantaneous emission rate of each vehicle is extracted as the mobile node feature; each facility in the stationary source emission data is treated as a fixed node, and the total hourly emission of each facility is extracted as the fixed node feature; all mobile nodes and fixed nodes are collected to obtain the node set;

[0035] Iterate through all moving nodes and calculate the relative distance between any two moving nodes in the scene coordinate system. Set a preset moving node distance threshold. If the relative distance between any two moving nodes in the scene coordinate system is less than the preset moving node distance threshold, add a directed edge between the two moving nodes. If the relative distance between any two moving nodes in the scene coordinate system is greater than or equal to the preset moving node distance threshold, do not add an edge.

[0036] Traverse all fixed nodes and calculate the straight-line distance between any two fixed nodes in the scene coordinate system. Set a fixed node distance threshold; if the straight-line distance between any two fixed nodes in the scene coordinate system is less than the threshold, add an undirected edge between them; if the straight-line distance between any two fixed nodes in the scene coordinate system is greater than or equal to the threshold, do not add an edge. Using spatial indexing, match the nearest fixed node to each mobile node and add a directed edge between the mobile node and its nearest fixed node. Collect all directed and undirected edges to obtain an edge set. Construct the emission source network based on the obtained node and edge sets.

[0037] Preferably, the method for converting carbon emission intensity maps into a set of management strategies based on reinforcement learning algorithms includes:

[0038] The highway is divided into several control units, each of which records the current carbon emission intensity value and traffic flow data; at the same time, all adjustable control parameters are listed, including speed limits, number of lanes, signal timing schemes and incentive measures for new energy vehicles.

[0039] Three optimization objectives are established, including carbon emission reduction, traffic efficiency, and implementation cost targets; two types of constraints are set, including safety constraints and feasibility constraints.

[0040] Using a reinforcement learning algorithm, the policy network parameters are initialized. Multiple sets of control policy samples are generated through interaction with the environment. The achievement degree of three optimization objectives is evaluated for each sample, and the achievement degree scores of the three optimization objectives are obtained. The achievement degree scores of the three optimization objectives are weighted and calculated to obtain a comprehensive performance score. The set with the highest comprehensive performance score is selected from multiple sets of samples as the optimal control policy. The selected optimal control policy is transformed into specific executable control instructions. The executable control instructions include speed limit instructions, lane control instructions, and signal timing instructions.

[0041] The methods for obtaining the achievement scores of the three optimization objectives include:

[0042] The achievement score for the carbon emission reduction target is calculated by measuring the percentage decrease in carbon emission intensity after control measures; the achievement score for the traffic efficiency target is calculated by measuring the traffic volume per unit time; and the achievement score for the implementation cost target is calculated by calculating the total human and material resources required for the control measures.

[0043] Preferably, the method for generating the strategy fit index includes:

[0044] The system collects actual carbon emission data after the execution of control commands, synchronously records changes in traffic operation status, calculates the deviation rate between actual carbon emissions and simulation predictions, and generates a standardized strategy fit index with a value range of 0 to 1. It presets a first threshold and a second threshold for the standardized strategy fit index. If the strategy fit index is greater than the second threshold, the control strategy is deemed to be suitable. If the strategy fit index is between the first and second thresholds, the control strategy needs fine-tuning. If the strategy fit index is less than the first threshold, the control strategy is deemed unsuitable, the carbon emission accounting model parameters are re-optimized, and the strategy update process is triggered.

[0045] Compared with the prior art, the beneficial effects of the present invention are:

[0046] By acquiring multi-source traffic parameters through distributed sensor networks, processing relevant data using spatiotemporal fusion algorithms, and generating a dynamic carbon emission factor matrix, while simultaneously performing sensitivity classification, key factors affecting carbon emissions can be captured more accurately, making the determination of carbon emission factors more consistent with actual conditions.

[0047] The digital twin modeling layer constructs a road network twin based on the sensitivity classification results, integrates high-precision maps and real-time traffic flow data, generates a three-dimensional road topology through multi-scale grid division, and overlays a vehicle energy consumption model to form a dynamic twin scene. This allows the constructed scene to reflect the dynamic changes of roads and vehicles in real time, with a higher degree of fit with actual road conditions, providing a more realistic basic scene for subsequent carbon emission calculations and simulations.

[0048] The multi-dimensional carbon emission calculation layer establishes a spatiotemporal mapping relationship between twin scenarios and carbon emission factors, adopts an event-triggered mechanism to synchronize traffic flow status, integrates mobile source and stationary source emission data to generate road segment-level carbon emission intensity maps, and realizes the effective fusion of data from different emission sources, making the presentation of carbon emission intensity more refined and targeted, and clearly showing the carbon emission status of different road segments.

[0049] The simulation and optimization layer transforms the carbon emission intensity map into a set of control strategies based on reinforcement learning algorithms. It predicts the carbon emission change trend under different strategies through Monte Carlo simulation and outputs the Pareto optimal strategy combination, making the formulation of control strategies more scientific and systematic, and providing diversified strategy options according to different carbon emission situations.

[0050] The execution feedback adjustment layer monitors carbon emission data in real time after the implementation of the control strategy, calculates the deviation rate between the actual carbon emission and the simulation prediction value, generates a strategy fit index, and dynamically corrects the parameters of the carbon emission accounting model until the strategy fit index reaches a stable threshold, forming a closed-loop optimization process. This enables the model to continuously adapt to changes in the actual situation and improve the fit between the strategy and actual needs.

[0051] The collaborative work among different layers, from data collection, modeling, calculation, and extrapolation to feedback and adjustment, forms a complete system covering all aspects of highway carbon emission control. This helps to achieve full-process, refined management of carbon emissions, making carbon emission control more dynamic and adaptable, and better able to cope with complex and ever-changing highway traffic conditions. Attached Figure Description

[0052] Figure 1 This is a schematic diagram illustrating the working principle of the highway carbon emission simulation and extrapolation system based on digital twins as described in this invention.

[0053] Figure 2 A flowchart for data acquisition in a distributed sensor network;

[0054] Figure 3 A flowchart for generating the 3D topology of a road;

[0055] Figure 4 A flowchart for integrating emission data into a carbon emission accounting model;

[0056] Figure 5 A flowchart for enhancing learning to generate control strategies. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] Please see Figure 1 This invention provides a highway carbon emission simulation and extrapolation system based on digital twins, the system comprising:

[0059] The carbon emission data acquisition layer acquires multi-source traffic parameters through a distributed sensor network, applies a spatiotemporal fusion algorithm to process vehicle driving status data, road environment data, and meteorological data, generates a dynamic carbon emission factor matrix, and performs sensitivity classification on the carbon emission factor matrix.

[0060] The digital twin modeling layer constructs a road network twin based on the sensitivity classification results, integrates high-precision maps and real-time traffic flow data, generates a three-dimensional road topology through multi-scale grid division, and overlays a vehicle energy consumption model to form a dynamic twin scene;

[0061] The multi-dimensional carbon emission calculation layer establishes a spatiotemporal mapping relationship between twin scenarios and carbon emission factors, adopts an event-triggered mechanism to synchronize traffic flow status, integrates mobile source and stationary source emission data through a carbon emission accounting model, and generates a road segment-level carbon emission intensity map.

[0062] The simulation and optimization layer uses reinforcement learning algorithms to transform the carbon emission intensity map into a set of control strategies. It then uses Monte Carlo simulation to predict the carbon emission change trend under different strategies and outputs the Pareto optimal strategy combination.

[0063] The execution feedback adjustment layer monitors carbon emission data after the implementation of control strategies in real time, calculates the deviation rate between actual carbon emissions and simulation predictions, generates a strategy fit index, and dynamically corrects the parameters of the carbon emission accounting model until the strategy fit index reaches a stable threshold.

[0064] Example 1: See Figure 2In the carbon emission data acquisition layer, the process of acquiring multi-source traffic parameters through a distributed sensor network requires a combination of hardware deployment and algorithm optimization. Specifically, three types of acquisition terminals are deployed along the highway according to functional requirements: millimeter-wave radar, video surveillance equipment, and environmental sensors. Millimeter-wave radar is installed on gantry frames above the road or on side poles. Its emitted electromagnetic waves can penetrate weather interference such as rain and fog, continuously capturing real-time speed and traffic flow data of passing vehicles. It also distinguishes different vehicle types, such as large trucks and small passenger cars, based on the characteristics of the radar echoes. Video surveillance equipment uses high-definition cameras, installed in areas with obstructed visibility, such as curves and slopes. Image recognition technology further refines vehicle type classification, supplementing the millimeter-wave radar's detailed information in vehicle type identification. Environmental sensors are distributed in the median strips on both sides of the road, and spaced out at bridge and tunnel entrances and exits, collecting real-time data such as road surface temperature, atmospheric humidity, and wind direction and speed. Multi-source traffic parameters thus encompass dynamic traffic flow data (vehicle speed, traffic volume, vehicle type classification), static road attribute data (road grade, number of lanes, pavement material), and meteorological interference data (wind direction and speed, pavement temperature).

[0065] An adaptive sampling mechanism is employed to dynamically adjust the data collection frequency. Its core principle is to divide the collection area based on traffic flow density and set sampling intervals. The system first accesses the historical traffic database, filtering out the frequency and duration of congestion at tollbooth entrances / exits, bridge / tunnel areas, and interchanges over the past year, marking these areas as high-frequency collection zones. Straight road sections, due to their relatively stable traffic flow, are marked as low-frequency collection zones. A sampling frequency function is established, using a traffic flow density threshold as the dividing criterion. When the number of vehicles in a certain area exceeds the threshold per unit time, the system automatically switches to high-frequency collection mode, employing millisecond-level sampling intervals to ensure the capture of instantaneous changes in vehicle states such as rapid acceleration and deceleration. When the number of vehicles is below the threshold, the system switches to second-level sampling intervals to reduce unnecessary data redundancy.

[0066] Noise correction is applied to multi-source traffic parameters using a Kalman filter algorithm. This algorithm takes the raw data output from each data acquisition terminal as input and processes it through two steps: prediction and update. In the prediction phase, the traffic parameters for the next moment are estimated based on a physical model of vehicle motion. In the update phase, the estimated values ​​are corrected by incorporating newly acquired actual data, thereby filtering out outliers caused by equipment vibration and electromagnetic interference, and outputting denoised traffic parameters. Based on the denoised data, the system periodically analyzes the traffic flow trends in each area. If a road segment experiences congestion for a week and the original data acquisition terminals have insufficient coverage, an automatic deployment command for additional data acquisition terminals is triggered, and equipment is temporarily installed using mobile brackets. If the traffic flow in a certain area remains below a set threshold for a month, and historical data shows that this state is stable, a command to reduce the number of data acquisition terminals is issued, and some equipment is removed to reduce energy consumption. Finally, the data collected from different terminals is integrated through a data synchronization protocol to form a multi-source traffic parameter set with unified timestamps and standardized format.

[0067] When using a spatiotemporal fusion algorithm to process vehicle driving status data, road environment data, and meteorological data to generate a dynamic carbon emission factor matrix, the multi-source traffic parameters are first normalized. Data such as vehicle speed and traffic flow are mapped to a range of 0 to 1 according to their value range, eliminating dimensional differences between different parameters. The processed parameters are arranged into a three-dimensional array according to road segment number, collection time, and parameter type. For example, the coordinates of a certain element in the array (3, 15:30, 2) can represent the second type of parameter (such as road surface temperature) collected at 15:30 on road segment 3.

[0068] The constructed spatiotemporal fusion network structure comprises a temporal convolutional module, a spatial attention module, and a feature concatenation layer. After inputting a three-dimensional array into the network, the convolutional neural network in the temporal convolutional module performs sliding computation on the data using convolutional kernels of different sizes, generating short-term feature sequences (e.g., traffic flow fluctuations within 5 minutes), medium-term feature sequences (e.g., traffic flow trends within 1 hour), and long-term feature sequences (e.g., traffic cycle patterns over 24 hours). The spatial attention module weights the importance of parameters for different road segments; for example, in bridge areas, road surface temperature has a higher weight than in straight road segments. A self-attention mechanism automatically adjusts the proportion of parameters for each road segment in the overall calculation. A feature interaction channel is established between the temporal convolutional module and the spatial attention module, concatenating the temporal features (e.g., speed changes during morning rush hour) and spatial features (e.g., slope and curvature of the road segment) of the same road segment to form a comprehensive feature vector with both spatiotemporal attributes.

[0069] In the output layer of the spatiotemporal fusion network structure, the fully connected layer compresses the high-dimensional feature vectors, transforming the multidimensional data into a two-dimensional matrix through matrix operations. Rows in the matrix represent different road segments, columns represent different times, and the values ​​in the matrix reflect the carbon emission influencing factors of that road segment at the corresponding time, thus generating a dynamic carbon emission factor matrix. This matrix is ​​dynamically updated over time; when significant changes in traffic flow or weather conditions are detected, the values ​​in the matrix are adjusted in real time to reflect the distribution of carbon emission influencing factors under the current conditions.

[0070] Example 2: See Figure 3 When classifying the sensitivity of the carbon emission factor matrix, three steps are required: sensitivity coefficient calculation, threshold classification, and level labeling. First, the sensitivity coefficient of each element in the carbon emission factor matrix is ​​calculated. Specifically, partial derivatives are calculated from vehicle speed data, road surface temperature data, and wind direction and speed data from multiple traffic sources. By analyzing the impact of small changes in these parameters on the carbon emission factor values, the sensitivity coefficient corresponding to each element is obtained.

[0071] A first and a second threshold for sensitivity coefficients are preset, determined based on extensive historical carbon emission data and correlation analysis of traffic parameters. The calculated sensitivity coefficients are compared to these two thresholds to classify the sensitivity level. If the sensitivity coefficient is less than the first threshold, it indicates that the parameter corresponding to that element has a relatively small impact on the carbon emission factor, and it is marked as a low-sensitivity factor. If the sensitivity coefficient is greater than the first threshold but less than the second threshold, it indicates that the parameter has a certain impact on the carbon emission factor, and it is marked as a medium-sensitivity factor. If the sensitivity coefficient is greater than the second threshold, it means that the parameter has a significant impact on the carbon emission factor, and it is marked as a high-sensitivity factor.

[0072] Different identifiers are used to classify the sensitivity of the carbon emission factor matrix, for example, using different colors or numerical codes as identifiers, with different identifiers corresponding to different sensitivity levels. Low-sensitivity factors are defined as routine monitoring items, meaning data is collected and monitored at a regular frequency; medium-sensitivity factors are key monitoring items, requiring increased monitoring frequency to more accurately capture their changes; high-sensitivity factors are core monitoring items, employing the highest monitoring frequency and the most stringent monitoring standards to ensure real-time monitoring of their dynamics.

[0073] When generating the 3D topology of the road, high-precision road base data is loaded into the digital twin platform based on the sensitivity classification results. This high-precision road base data covers road design parameters, such as road width, design speed, and curve radius; pavement material properties, such as asphalt or cement pavement and pavement roughness; and data on the distribution of roadside facilities, such as the location of traffic signs, the type and length of median strips, and the distribution of streetlights.

[0074] Different modeling precision is adopted for parameters with different sensitivity levels. For road sections corresponding to high sensitivity factors, since the changes in parameters have a significant impact on carbon emission simulation results, centimeter-level modeling precision is used to meticulously represent the minute undulations of the road surface and the precise position of lane lines; for road sections corresponding to medium sensitivity factors, decimeter-level modeling precision is used to clearly display the overall structure and main facilities of the road; for road sections corresponding to low sensitivity factors, meter-level modeling precision is used to mainly present the general direction and basic layout of the road.

[0075] A grid subdivision algorithm is applied to refine the road model, dividing it into grids of varying sizes based on the sensitivity level and terrain characteristics of different road sections. Terrain elevation data is incorporated to correct the road's 3D coordinates, ensuring the model matches the actual terrain. An iterative optimization algorithm continuously adjusts the grid density, comparing the model with actual measurement data during iteration to calculate model error. Iteration stops when the model error falls below a preset threshold.

[0076] A 3D road topology is constructed using 3D modeling software, accurately reproducing the number and location of lane lines, the shape and extent of medians, and the style and location of various traffic signs. Sensitivity level markers are integrated into the 3D topology; for example, specific textures or markings are added to road sections corresponding to high sensitivity factors. The final result is a 3D road topology with sensitivity level markers, providing an accurate road model foundation for subsequent carbon emission simulation.

[0077] Example 3: See Figure 4 When establishing the spatiotemporal mapping relationship between twin scenes and carbon emission factors, a unified scene coordinate system needs to be defined first. Taking the starting point of the highway as the origin, the X-axis is parallel to the road centerline and set along the road's extension direction; the Y-axis is perpendicular to the road cross-section and points outwards; the Z-axis is the elevation direction, perpendicular to the horizontal plane and upwards. This coordinate system allows for the quantitative description of the position of all elements along the highway. GPS positioning equipment is used to calibrate the data collection terminals deployed along the route, obtaining the precise position coordinates of each terminal in this coordinate system, accurate to the centimeter level, ensuring the spatial accuracy of subsequent data mapping.

[0078] A coordinate transformation algorithm is used to convert the parameter coordinates of the dynamic carbon emission factor matrix to the scene coordinate system. Each parameter in the dynamic carbon emission factor matrix, originally based on the local coordinate records of the acquisition terminal, needs to be mapped to a unified scene coordinate system through rotation, translation, and scaling operations, thereby generating a carbon emission factor distribution in the scene coordinate system. The distribution result is presented in a grid format, with each grid corresponding to a specific spatial range, recording the carbon emission factor values ​​within that range. Using the origin of the scene coordinate system as a reference, the digital twin scene is aligned with the coordinates. By adjusting the spatial position parameters of the twin scene, the positions of elements such as roads and vehicles in the twin scene are made to correspond one-to-one with the positions of actual highways, obtaining the absolute coordinates of the twin scene.

[0079] The 3D road topology is transformed into the scene coordinate system using a coordinate mapping algorithm. Originally constructed based on the relative coordinates of the design drawings, the road's 3D topology needs to have its absolute position in the scene coordinate system calculated using the coordinate mapping algorithm, thus generating a twin scene model in the scene coordinate system. The model includes details such as the road's 3D geometry, lane divisions, and the spatial structure of bridges and tunnels. In the scene coordinate system, the carbon emission factor distribution and the twin scene model are spatiotemporally registered. Spatial registration adjusts the grid position of the carbon emission factor distribution by calculating the spatial overlap between the two, ensuring it precisely corresponds to the road segments in the twin scene model. Temporal registration synchronizes the timestamps of both, ensuring that changes in the carbon emission factor's value are consistent with changes in the vehicle's driving status in the twin scene, thereby establishing a mapping relationship between the spatial and temporal dimensions. Using the timestamp of the acquisition terminal as the reference clock, the timeline of the twin scene is unified through a network time protocol, ensuring that the time records of all dynamic elements in the twin scene (such as vehicle movement and traffic light changes) are synchronized with the acquisition terminal, with time errors controlled at the millisecond level.

[0080] When integrating mobile and stationary emission data through a carbon emission accounting model, an emission source network is constructed based on the carbon emission factor distribution in the scene coordinate system and a twin scene model. Mobile sources include all vehicles traveling on highways, while stationary sources include office buildings at toll stations, charging facilities in service areas, and road lighting equipment. The graph neural network-based carbon emission accounting model comprises five layers: an input layer, a feature extraction layer, a source term fusion layer, a spatiotemporal correction layer, and an output layer. The input layer receives data from the emission source network and converts the emission parameters of mobile and stationary sources into vector forms recognizable by the neural network. The feature extraction layer extracts instantaneous emission features of mobile sources (such as the surge in emissions during vehicle acceleration) and stable emission features of stationary sources (such as the continuous energy consumption emissions of lighting equipment) through convolution operations. The source term fusion layer calculates the mutual influence between mobile and stationary sources through an attention mechanism. For example, vehicle emissions on a certain road segment may be affected by the superimposed emissions from stationary sources in nearby service areas; the fusion layer quantifies this superimposed effect. The spatiotemporal correction layer corrects the fused emission data based on spatial distance and time intervals in the scene coordinate system. For example, emission sources that are farther apart have less mutual influence, and emission data from different time points are not directly superimposed. The output layer uses a fully connected neural network to convert the processed feature vectors into a road segment-level carbon emission intensity map. The map displays the differences in carbon emission intensity between different road segments in a color gradient, with darker colors indicating higher carbon emission intensity. The final generated map covers the entire highway network, and the carbon emission intensity data for each road segment is updated every 5 minutes.

[0081] The core calculation formula of the carbon emission accounting model is:

[0082]

[0083] in, This indicates the total carbon emissions of a certain road section. This indicates the carbon emissions from mobile sources along this road section. This indicates the carbon emissions from stationary sources along this road section. This represents the interactive emissions between mobile and stationary sources. , , These are the weighting coefficients for the three factors, which are dynamically adjusted based on the road segment type (such as urban or suburban road segments).

[0084] Example 4: When constructing the emission source network, mobile and stationary sources are first processed as nodes. Each vehicle in the mobile source emission data under the scene coordinate system is treated as a mobile node. Data such as instantaneous engine fuel injection quantity and speed are collected through the on-board OBD interface. Combined with vehicle type (e.g., heavy truck, small car) and real-time speed, the instantaneous emission rate of each vehicle is calculated as a feature of the mobile node. For example, the instantaneous emission rate of a heavy diesel truck traveling at 90 km / h on a highway will be updated in real time according to changes in engine load. This data will be continuously recorded as the core feature of the corresponding mobile node. Each facility in the stationary source emission data is treated as a fixed node. For the air conditioning system of a toll station, the hourly energy consumption data is read through electricity and gas meters and converted into the total hourly emissions. For charging piles in service areas, indirect carbon emissions are calculated based on charging power and usage time, and used as a feature of the fixed node. All mobile nodes corresponding to all vehicles in motion and all fixed nodes corresponding to facilities along the route are aggregated to form a node set. Each node in the set contains a unique identifier, spatial coordinates, and emission characteristic data.

[0085] Iterate through all moving nodes and calculate the relative distance between any two moving nodes in the scene coordinate system. The relative distance is calculated using the difference in spatial coordinates between two points. For example, in the scene coordinate system, if the coordinates of moving node A are (x1, y1, z1) and the coordinates of moving node B are (x2, y2, z2), then the relative distance between them is the straight-line distance between the two points in 3D space. A preset moving node distance threshold is set based on the width of highway lanes and vehicle safety distances, for example, 50 meters. When the relative distance between two moving vehicles is less than 50 meters, a directed edge is added between the corresponding two moving nodes, with the edge pointing from the front vehicle to the rear vehicle, because the exhaust emissions of the front vehicle directly affect the driving environment of the rear vehicle. If the relative distance is greater than or equal to 50 meters, no edge is added, indicating that the emission impact between the two vehicles is negligible.

[0086] Iterate through all fixed nodes and calculate the straight-line distance between any two fixed nodes in the scene coordinate system. For example, the straight-line distance between a group of charging piles in a service area and a group of nearby streetlights is calculated using their coordinates. A preset fixed node distance threshold is set, based on the emission impact range of the facility, for example, 100 meters. When the straight-line distance between two fixed nodes is less than 100 meters, an undirected edge is added between the two fixed nodes because their emissions will have a cumulative effect in the local space; if the straight-line distance is greater than or equal to 100 meters, no edge is added. Using spatial indexing techniques, such as R-tree indexing, all fixed nodes are spatially partitioned, and the nearest fixed node is quickly matched for each mobile node. For example, a vehicle driving near a service area finds the nearest charging pile facility using spatial indexing, and a directed edge is added between the mobile node and this fixed node, pointing from the fixed node to the mobile node, because emissions from stationary sources will spread to the area where the vehicle is located.

[0087] All directed and undirected edges are collected to form an edge set. Each node in the node set contains a unique identifier, spatial coordinates, and emission characteristics. Each edge in the edge set contains the connected pair of nodes and the edge type (directed or undirected). An emission source network is constructed based on the node and edge sets, presented graphically. Nodes represent mobile and stationary sources, and edges represent the emission impact relationships between them. This network is dynamically updated. When a vehicle's location changes, the coordinates of mobile nodes are updated in real time, and the corresponding edges are added or removed accordingly. When the operational status of a stationary source changes (e.g., a facility ceases operation), its node characteristics are updated, and the attributes of related edges are adjusted. The emission source network comprehensively presents the spatial distribution of all emission sources on the highway and their interactions, providing a structured data foundation for subsequent carbon emission accounting.

[0088] Example 5: See Figure 5 When transforming carbon emission intensity maps into a set of control strategies using reinforcement learning algorithms, highways are first divided into several control units. The division is based on factors including natural road segmentation (e.g., road segments between different interchanges), traffic flow characteristics (e.g., congested and uncongested sections during peak hours), and differences in carbon emission intensity distribution. The length of each control unit can be adjusted according to actual conditions. Each unit records its current carbon emission intensity value, derived from the corresponding area data in the road segment-level carbon emission intensity map, and also records traffic flow data, including the number of vehicles passing through per unit time and average vehicle speed. All adjustable control parameters are listed, including speed limits (e.g., adjusting the speed limit of a road segment from 120 km / h to 100 km / h), number of lanes (e.g., temporarily opening emergency lanes as through lanes to increase the number of lanes), traffic light timing schemes (e.g., adjusting the green light duration at toll station entrances), and incentives for new energy vehicles (e.g., toll exemptions for new energy vehicles).

[0089] Three optimization objectives are established: carbon emission reduction, traffic efficiency, and implementation cost. Two types of constraints are set: safety constraints, including that speed limits must not be lower than the minimum safe speed and that lane number adjustments must not affect emergency passage; and feasibility constraints, including that adjustments to traffic light timing schemes must comply with traffic signal control regulations and that incentive measures for new energy vehicles must be within the existing policy framework.

[0090] When using reinforcement learning algorithms, the policy network parameters are first initialized. The network input consists of the carbon emission intensity value and traffic flow data of the control unit, and the output consists of possible control strategies. Multiple sets of control strategy samples are generated through interaction with the environment. Each sample contains a specific set of control parameter adjustment schemes. The achievement degree of three optimization objectives is evaluated for each sample, resulting in achievement scores for the three optimization objectives. The achievement score for the carbon emission reduction objective is calculated by the percentage decrease in carbon emission intensity after control measures; the achievement score for the traffic efficiency objective is calculated by statistically analyzing the road segment traffic volume per unit time; and the achievement score for the implementation cost objective is calculated by calculating the total human and material resources required for control measures (such as the number of traffic guidance personnel and equipment deployment costs required for temporarily opening emergency lanes). The achievement scores of the three optimization objectives are then weighted and calculated, with weights determined according to the current traffic management priority, to obtain a comprehensive effectiveness score. The set with the highest comprehensive effectiveness score from multiple samples is selected as the optimal control strategy. The selected optimal control strategy is transformed into specific, executable control instructions, such as speed limit instructions issued through variable speed limit signs, lane control instructions implemented through traffic cones and guidance signs, and signal timing instructions executed through the traffic signal control system.

[0091] When generating the strategy fit index, actual carbon emission data after the execution of control commands is collected, which can be obtained through carbon emission monitoring equipment along the route. Changes in traffic operation status, including real-time changes in vehicle speed and traffic flow, are recorded simultaneously. The deviation rate between the actual carbon emissions and the simulation prediction is calculated. The deviation rate is calculated by dividing the difference between the actual and predicted values ​​by the absolute value of the predicted value. Then, a standardized strategy fit index with a value range of 0 to 1 is generated through standardization processing. A first threshold and a second threshold for the standardized strategy fit index are preset, for example, the first threshold is 0.6 and the second threshold is 0.8. If the strategy fit index is greater than 0.8, the control strategy is considered fit and execution can continue. If the strategy fit index is between 0.6 and 0.8, the control strategy needs fine-tuning, such as adjusting speed limits or traffic light timings. If the strategy fit index is less than 0.6, the control strategy is considered unfit, the carbon emission calculation model parameters are re-optimized (e.g., correcting parameter coefficients in the vehicle energy consumption model), and a strategy update process is triggered to regenerate a new control strategy. This process is repeated until the strategy fit index reaches a stable threshold, that is, the fluctuation range of the strategy fit index calculated in multiple consecutive calculations is less than the set value.

[0092] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0093] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A highway carbon emission simulation deduction system based on digital twinning, characterized in that, The application comprises: a carbon emission data collection layer, which acquires multi-source traffic parameters through a distributed sensing network, processes vehicle driving state data, road environment data and meteorological data using a space-time fusion algorithm, generates a dynamic carbon emission factor matrix, and classifies the sensitivity of the carbon emission factor matrix; a sensitivity coefficient of each element in the carbon emission factor matrix is calculated, and the sensitivity coefficient is obtained by performing partial derivative calculation on vehicle speed data, road surface temperature data and wind direction and speed data in the multi-source traffic parameters; a digital twin modeling layer, which constructs a road network twin according to the sensitivity classification result, fuses high-precision maps and real-time traffic flow data, generates a three-dimensional topological structure of the road through multi-scale grid division, and superimposes a vehicle energy consumption model to form a dynamic twin scene; a multi-dimensional carbon emission calculation layer, which establishes a space-time mapping relationship between the twin scene and the carbon emission factor, synchronizes the traffic flow state using an event triggering mechanism, integrates mobile source and fixed source emission data through a carbon emission accounting model, and generates a road section-level carbon emission intensity atlas; a simulation deduction optimization layer, which converts the carbon emission intensity atlas into a set of control strategies based on a reinforcement learning algorithm, predicts the carbon emission trend under different strategies through Monte Carlo simulation, and outputs a Pareto optimal strategy combination; an execution feedback adjustment layer, which monitors the carbon emission data after the implementation of the control strategy in real time, calculates the deviation rate of the actual carbon emission amount and the simulation prediction value, generates a strategy adaptation index, dynamically corrects the parameters of the carbon emission accounting model until the strategy adaptation index reaches a stable threshold; The method for processing vehicle driving state data, road environment data and meteorological data using a space-time fusion algorithm to generate a dynamic carbon emission factor matrix comprises: normalizing the multi-source traffic parameters, arranging the normalized parameters into a three-dimensional array according to time sequence and spatial position, and constructing a space-time fusion network structure containing a time convolution module, a spatial attention module and a feature concatenation layer; inputting the three-dimensional array into the space-time fusion network structure, performing time dimension feature extraction on the input data using a convolutional neural network in the time convolution module to generate time feature sequences of different lengths, the time feature sequences of different lengths including short-term feature sequences, medium-term feature sequences and long-term feature sequences, performing weighted fusion of spatial dimension features through a self-attention mechanism in the spatial attention module, and setting a feature interaction channel between the time convolution module and the spatial attention module to concatenate the time features and spatial features of the same road section; applying a fully connected layer to the output layer of the space-time fusion network structure to compress the dimension, converting the high-dimensional feature vector output by the network into a two-dimensional matrix, and generating a dynamic carbon emission factor matrix.

2. The highway carbon emission simulation and deduction system based on digital twinning of claim 1, wherein The method for acquiring multi-source traffic parameters through a distributed sensing network comprises: deploying three kinds of collection terminals, i.e., millimeter wave radars, video monitoring devices and environmental sensors, along the expressway to collect vehicle speed, traffic volume, vehicle type classification data, road surface temperature and real-time wind direction and speed data; the multi-source traffic parameters include dynamic traffic flow data, static road attribute data and meteorological interference data; Adopting an adaptive sampling mechanism, the data collection frequency is dynamically adjusted according to the traffic flow density; based on historical congestion data, the entrance and exit of toll stations, bridge and tunnel areas and interchanges are marked as high-frequency collection areas; flat road segments are marked as low-frequency collection areas; a sampling frequency function is established, and the high-frequency collection area and the low-frequency collection area are divided by the traffic flow density threshold; the high-frequency collection area adopts a millisecond-level sampling interval, and the low-frequency collection area adopts a second-level sampling interval; The noise interference in the multi-source traffic parameters is corrected in real time through the Kalman filtering algorithm, the input of the Kalman filtering algorithm is the original collected data, and the output is the denoised traffic parameters; the deployment position of the collection terminal is dynamically adjusted according to the denoised data, if a new congestion-prone area is detected, a new collection terminal is added in the area; if it is detected that the traffic flow in the coverage area of the original collection terminal is continuously below the threshold, the number of collection terminals in the area is reduced; the collection data of different terminals is integrated through a data synchronization protocol to obtain complete multi-source traffic parameters.

3. The expressway carbon emission simulation and deduction system based on digital twinning according to claim 2, characterized in that, The method for classifying the sensitivity of the carbon emission factor matrix comprises: A first threshold of the sensitivity coefficient and a second threshold of the sensitivity coefficient are preset, and the sensitivity coefficient is compared with the first threshold of the sensitivity coefficient and the second threshold of the sensitivity coefficient respectively; if the sensitivity coefficient is less than the first threshold of the sensitivity coefficient, the parameter corresponding to the element is marked as a low-sensitive factor; if the sensitivity coefficient is greater than the first threshold of the sensitivity coefficient and less than the second threshold of the sensitivity coefficient, the parameter corresponding to the element is marked as a medium-sensitive factor; if the sensitivity coefficient is greater than the second threshold of the sensitivity coefficient, the parameter corresponding to the element is marked as a high-sensitive factor; The carbon emission factor matrix is classified in different sensitivities using different identifiers, different identifiers represent different sensitivity levels, the low-sensitive factor is defined as a regular monitoring item, the medium-sensitive factor is a key monitoring item, and the high-sensitive factor is a core monitoring item.

4. The expressway carbon emission simulation and deduction system based on digital twinning of claim 3, wherein The method for generating a road three-dimensional topology structure comprises: According to the sensitivity classification result, load high-precision road basic data in the digital twin platform; the high-precision road basic data includes road design parameters, road surface material properties and along-line facility distribution data; different modeling accuracies are adopted for parameters of different sensitivity levels, a road section corresponding to a high-sensitive factor adopts a centimeter-level modeling accuracy, a road section corresponding to a medium-sensitive factor adopts a decimeter-level modeling accuracy, and a road section corresponding to a low-sensitive factor adopts a meter-level modeling accuracy; a grid subdivision algorithm is applied to finely process the road model, terrain elevation data is introduced to correct the road three-dimensional coordinates, and the grid density is continuously adjusted through an iterative optimization algorithm, and the processing is stopped when the model error is less than a preset threshold; a road three-dimensional topology structure containing lane lines, isolation belts and traffic signs is constructed based on a three-dimensional modeling software, and finally a road three-dimensional topology structure with a sensitivity level mark is generated.

5. The expressway carbon emission simulation and deduction system based on digital twinning according to claim 4, characterized in that, The method for establishing a space-time mapping relationship between a twin scene and a carbon emission factor comprises: A unified scene coordinate system is defined with the starting point of the expressway as the origin, the X-axis parallel to the road center line, the Y-axis perpendicular to the road cross section, and the Z-axis in the elevation direction; a GPS positioning device is used to calibrate the collection terminal to obtain the position coordinates of the collection terminal; a coordinate conversion algorithm is used to convert the parameter coordinates of the dynamic carbon emission factor matrix to the scene coordinate system, and then the carbon emission factor distribution in the scene coordinate system is obtained; the origin of the scene coordinate system is taken as the reference datum to perform coordinate alignment on the digital twin scene to obtain the absolute coordinates of the twin scene; A coordinate mapping algorithm is used to convert the three-dimensional topological structure of the road to the scene coordinate system, and then the twin scene model in the scene coordinate system is obtained; in the scene coordinate system, the carbon emission factor distribution and the twin scene model are spatio-temporally registered to establish a mapping relationship in the spatial dimension; the timestamp of the collection terminal is taken as the reference clock, and a time synchronization protocol is used to unify the time axis of the twin scene to establish a mapping relationship in the time dimension.

6. The expressway carbon emission simulation and deduction system based on digital twinning according to claim 5, characterized in that, The method for integrating mobile source and fixed source emission data through the carbon emission accounting model comprises: According to the carbon emission factor distribution in the scene coordinate system and the twin scene model, an emission source network is constructed, and a carbon emission accounting model based on a graph neural network is used to integrate the mobile source and fixed source emission data; the carbon emission accounting model comprises an input layer, a feature extraction layer, a source item fusion layer, a space-time correction layer, and an output layer; the emission source network is input to the input layer of the carbon emission accounting model, and a road section level carbon emission intensity map is generated through the output layer.

7. The expressway carbon emission simulation and deduction system based on digital twinning according to claim 6, characterized in that, The method for constructing the emission source network comprises: Each vehicle in the mobile source emission data in the scene coordinate system is taken as a mobile node, and the instantaneous emission rate of each vehicle is extracted as the feature of the mobile node; each facility in the fixed source emission data is taken as a fixed node, and the hourly total emission of each facility is extracted as the feature of the fixed node; all mobile nodes and fixed nodes are collected to obtain a node set; All mobile nodes are traversed, and the relative distance between any two mobile nodes in the scene coordinate system is calculated; a preset mobile node distance threshold is set, if the relative distance between any two mobile nodes in the scene coordinate system is less than the preset mobile node distance threshold, a directed edge is added between the corresponding two mobile nodes; if the relative distance between any two mobile nodes in the scene coordinate system is greater than or equal to the preset mobile node distance threshold, no edge is added; All fixed nodes are traversed, and the straight line distance between any two fixed nodes in the scene coordinate system is calculated; a preset fixed node distance threshold is set, if the straight line distance between any two fixed nodes in the scene coordinate system is less than the preset fixed node distance threshold, an undirected edge is added between the corresponding two fixed nodes; if the straight line distance between any two fixed nodes in the scene coordinate system is greater than or equal to the preset fixed node distance threshold, no edge is added; through a spatial indexing technology, the closest fixed node to each mobile node is matched, and a directed edge between the mobile node and the corresponding closest fixed node is added; all directed edges and undirected edges are collected to obtain an edge set; based on the obtained node set and edge set, an emission source network is constructed.

8. The expressway carbon emission simulation and deduction system based on digital twinning according to claim 7, characterized in that, The method for converting the carbon emission intensity map into the set of control strategies based on the reinforcement learning algorithm comprises: The expressway is divided into a plurality of control units, and the current carbon emission intensity value and traffic flow data are recorded in each unit. Meanwhile, all adjustable control parameters are listed, including speed limit value, lane number, signal timing scheme, and new energy vehicle incentive measures. Three optimization objectives are established, including carbon emission reduction target, traffic efficiency target, and implementation cost target. Two types of constraint conditions are set, including safety constraint and feasibility constraint. The reinforcement learning algorithm is used to initialize the strategy network parameters, generate a plurality of sets of control strategy samples by interacting with the environment, evaluate the achievement degree of the three optimization objectives for each sample, obtain the achievement degree scores of the three optimization objectives, and perform weighted calculation on the achievement degree scores of the three optimization objectives to obtain a comprehensive performance score. The set of control strategy with the highest comprehensive performance score is selected from the plurality of sets of samples as the optimal control strategy. The selected optimal control strategy is converted into specific executable control instructions. The executable control instructions include speed limit instruction, lane control instruction, and signal timing instruction. The method for obtaining the achievement degree scores of the three optimization objectives comprises: The decline rate of carbon emission intensity after control is calculated as the achievement degree score of the carbon emission reduction target. The road traffic volume per unit time is counted as the achievement degree score of the traffic efficiency target. The total manpower and material resources required for the control measures are calculated as the achievement degree score of the implementation cost target.

9. The expressway carbon emission simulation and deduction system based on digital twinning of claim 8, wherein, The method for generating the strategy adaptation index comprises: Actual carbon emission data after execution of the control instruction is collected, and traffic running state changes are recorded synchronously. The deviation rate of actual carbon emission and simulation prediction value is calculated to generate a standardized strategy adaptation index with a value range of 0 to 1. A first threshold value of the standardized strategy adaptation index and a second threshold value of the standardized strategy adaptation index are preset. If the strategy adaptation index is greater than the second threshold value, it is determined that the control strategy is adapted. If the strategy adaptation index is between the first threshold value and the second threshold value, it is determined that the control strategy needs to be fine-tuned. If the strategy adaptation index is less than the first threshold value, it is determined that the control strategy is not adapted, the carbon emission accounting model parameters are re-optimized, and a strategy updating process is triggered.

Images