An elevation-curvature coupled fuel consumption model method for vehicle path optimization

Abstract

This invention relates to a large-scale model method for elevation-curvature coupled fuel consumption for vehicle route optimization, belonging to the field of vehicle route planning technology. The method includes: acquiring multi-source heterogeneous vehicle route data and performing standardization processing to generate a multi-dimensional standardized dataset; using the multi-dimensional standardized dataset as input features, fusing measured data from actual heavy-load vehicle conditions and engine bench test data to construct a training sample set, completing the training and verification of the elevation-curvature coupled fuel consumption prediction sub-model; constructing a multi-objective optimization function, fusing multi-dimensional constraints, constructing a route planning sub-model based on a graph neural network and generating alternative routes, and performing weighted calculation of the comprehensive cost of each alternative route through a dynamic weight allocation mechanism; selecting the optimal route, monitoring the road network status in real time and triggering iterative updates of the large model, dynamically adjusting the optimal route, and quantitatively outputting energy-saving and emission-reduction benefit data.

Description

Technical Field

[0001] This invention belongs to the field of vehicle routing technology, specifically relating to a method for a large fuel consumption model using an elevation-curvature coupled system for vehicle routing optimization. Background Technology

[0002] With the rapid development of the transportation industry, the energy efficiency, economy and safety of vehicle route planning are receiving increasing attention, especially for heavy-duty vehicles, whose fuel consumption is significantly affected by road terrain and geometric features. Inappropriate route planning can not only increase fuel consumption and pollutant emissions, but may also reduce traffic efficiency and increase driving risks.

[0003] Existing vehicle route optimization methods primarily focus on minimizing distance or time, prioritizing traffic efficiency while neglecting the coupled impact of road elevation changes and curvature on fuel consumption. This is especially true in complex terrains like mountains and hills, where elevation fluctuations and road curvature significantly alter engine load and driving resistance, directly affecting fuel consumption. Current methods, failing to adequately consider this coupling, suffer from low fuel consumption prediction accuracy, hindering the achievement of green and energy-efficient route optimization goals. Some route planning methods that do consider fuel consumption merely introduce elevation or curvature alone, without modeling their coupling. This fails to reflect fuel consumption variations in curved and sloping road sections. Furthermore, these methods often rely on theoretical fuel consumption models, neglecting real-world heavy-load test data and engine bench test data. They ignore the influence of vehicle load, engine performance, and road condition fluctuations during actual driving, resulting in significant discrepancies between predicted and actual fuel consumption and failing to provide accurate fuel consumption references for route planning. Meanwhile, existing methods mostly employ traditional algorithms such as Dijkstra's algorithm and genetic algorithms for path planning. These algorithms suffer from low traversal efficiency and incomplete path search when dealing with complex topologies of the entire road network, making them difficult to adapt to the needs of the entire road network that integrates multiple scenarios such as urban areas, mountainous areas, and highways. Furthermore, most methods lack a dynamic weight allocation mechanism, making it impossible to adjust the optimization strategy according to the real-time status of the road network (such as the degree of road congestion and changes in weather conditions). When the road network experiences congestion, construction, or severe weather, the optimized path is prone to problems such as increased fuel consumption and decreased traffic efficiency, resulting in insufficient flexibility and adaptability of path optimization. This makes it difficult to meet the vehicle path planning needs in different scenarios, especially the core needs of heavy-duty vehicles for fuel consumption control and driving safety. Heavy-duty vehicles have a large load and are more sensitive to road slope and curvature, and their fuel consumption is more significantly affected by terrain factors. Existing methods are difficult to match their driving characteristics.

[0004] Furthermore, existing route planning methods are rather crude in quantifying energy conservation and emission reduction benefits. Most simply compare the fuel consumption differences between different routes without incorporating data such as actual vehicle driving parameters and pollutant emission factors. Therefore, they cannot accurately output the fuel savings, CO2, and other pollutants (such as NO) of a route. x The current methods for reducing emissions (PM) are not conducive to users' intuitive understanding of the green and energy-saving value of routes, and are also insufficient to support the low-carbon management and assessment requirements of the transportation industry. Furthermore, existing methods handle multi-source heterogeneous data in a relatively simple manner, lacking a comprehensive standardized processing workflow. Data from different sources and in different formats (such as vehicle parameter data, road network GIS data, and meteorological data) suffers from inconsistencies in coordinates, formats, and data anomalies, resulting in poor model training data quality and further affecting the accuracy and stability of route planning. In summary, existing vehicle route optimization methods suffer from low fuel consumption prediction accuracy, lack of elevation-curvature coupling modeling, insufficient adaptability and flexibility, coarse quantification of energy-saving and emission-reduction benefits, and non-standardized data processing. These shortcomings fail to meet the green, efficient, and accurate route planning requirements of scenarios such as heavy-duty vehicles. Therefore, a vehicle route optimization method is needed that can couple altitude and curvature factors, predict fuel consumption, adapt to complex road network scenarios, standardize data processing, and quantify energy-saving and emission-reduction benefits to address the deficiencies of existing technologies. Summary of the Invention

[0005] To address the aforementioned problems in the existing technology, this invention provides a method for a high fuel consumption model using an elevation-curvature coupled system for vehicle path optimization. The objective of this invention can be achieved through the following technical solutions: A method for a large fuel consumption model using elevation-curvature coupling for vehicle route optimization includes: S1: Acquire multi-source heterogeneous data on vehicle routes, complete standardization processing, and generate a multi-dimensional standardized dataset; S2: Using the multidimensional standardized dataset as input features, a training sample set is constructed by fusing real vehicle heavy-load working condition test data and engine bench test data to train and verify the elevation-curvature coupled fuel consumption prediction sub-model. S3: Construct a multi-objective optimization function with the core objectives of minimizing fuel consumption and carbon emissions, integrate multi-dimensional constraints, construct a path planning sub-model based on graph neural network, traverse the entire road network from the starting point to the end point to generate alternative paths, introduce a dynamic weight allocation mechanism, and complete the weighted calculation of the comprehensive cost of each alternative path. S4: Based on the multi-objective optimization function and the comprehensive cost weighted calculation results, sort the candidate paths and select the optimal path; monitor the dynamic changes in the road network status in real time, trigger the large model iterative update mechanism, dynamically adjust the optimal path, and synchronously quantify and output energy-saving and emission-reduction benefit data.

[0006] Specifically, the standardization process includes: Acquire multi-source heterogeneous data on vehicle routes, store it in a preset database, classify and summarize it according to data type, divide the data into categories and establish data association indexes; Initiate a standardized processing flow, perform preprocessing operations on various types of data in sequence, unify the coordinate system of spatial data in multi-source heterogeneous data, and convert spatial coordinate data under different geographic coordinate systems into a preset unified geographic coordinate system; perform format normalization processing on various types of data, and convert unstructured data into structured data. The data is traversed step by step to identify and remove abnormal data caused by equipment failure or transmission interference during the data acquisition process, thereby generating the multidimensional standardized dataset.

[0007] Specifically, the process of constructing the training sample set includes: The input features for elevation-curvature coupled fuel consumption prediction are extracted from the multidimensional standardized dataset, normalized, and the feature values ​​are mapped to a preset range. By integrating real vehicle heavy-load operating condition test data and engine bench test data, the two types of data are format matched and consistency verified, and the training sample set is constructed by integrating them according to a preset ratio.

[0008] Specifically, the training and validation process of the elevation-curvature coupled fuel consumption prediction sub-model includes: The training sample set is randomly divided into a training set and a validation set according to a preset ratio; an initial elevation-curvature coupled fuel consumption prediction sub-model is constructed, the model input layer, hidden layer and output layer structure are defined, and the model iteration parameters are set. The preprocessed input features are used as model input, and the vehicle's fuel consumption per unit mileage and carbon emissions are used as model output. The model training process is started, and iterative training is performed using the training set. The weight parameters are adjusted in real time, and the training error is calculated. After each round of training iterations, a validation set is input, and the prediction error is calculated. Training continues iteratively until the prediction error falls below a preset error threshold, at which point model training stops.

[0009] Specifically, the process of constructing the multi-objective optimization function includes: The core optimization objectives are minimizing total fuel consumption and total carbon emissions, with travel time and toll costs as auxiliary optimization objectives; weighting coefficients are set according to the importance of the optimization objectives. The multi-objective optimization problem is transformed into a single-objective optimization problem by using the weighted summation method, clarifying the mathematical expression of the optimization function, and defining the optimization variables, objective function, and boundary conditions.

[0010] Specifically, the process of constructing a path planning sub-model based on a graph neural network includes: The entire road network is defined as a graph structure, road intersections in the road network are defined as nodes in the graph structure, and road segments between adjacent intersections are defined as edges in the graph structure. The fuel consumption, travel time, travel cost, and quantified constraints of each road segment are extracted as attribute information of the graph structure edges and associated with the corresponding road segments. Initialize the graph neural network model parameters, input the global road network graph structure, start the graph neural network algorithm, perform a depth-first traversal of the global road network from the starting point to the end point, and search for all feasible paths that meet the constraints; perform deduplication and filtering on the searched feasible paths, remove duplicate paths and paths that do not meet the core constraints, and generate alternative paths.

[0011] Specifically, the dynamic weight allocation mechanism includes the following process: The elevation-curvature coupled fuel consumption prediction sub-model is invoked to obtain the fuel consumption prediction value of the corresponding road segment of the alternative route, which is used as the core weight factor for comprehensive cost calculation. The initial weight of the core weight factor is set. Based on the travel time, toll, and driving safety factor of each road segment, the corresponding auxiliary weight factor is set, and the initial weight of each auxiliary weight factor is clarified. Real-time road network status data is acquired, and the values ​​of core weight factors and auxiliary weight factors are dynamically adjusted according to the degree of road network congestion and weather conditions. The comprehensive cost value of each alternative route is generated by weighted summation.

[0012] Specifically, the process of sorting the candidate paths and selecting the optimal path includes: The parameters of each candidate path are input into the multi-objective optimization function to calculate the optimization objective value of each candidate path. The optimization objective value is then integrated with the comprehensive cost value to comprehensively rank all candidate paths. Based on the ranking results, the path with the lowest fuel consumption, optimal carbon emissions, and lowest overall cost is selected, and the feasibility of the selected path is verified.

[0013] Specifically, the monitoring process for the dynamic changes in the road network status includes: The acquired road network status data is analyzed in real time, compared with the preset normal range, and if a change in the road network status is detected, the large model iterative update mechanism is triggered, and the process of adjusting the multi-objective optimization function, traversing the road network of the path planning sub-model, generating alternative paths, and calculating the comprehensive cost weighting is re-executed.

[0014] Specifically, the process of quantifying and outputting the energy conservation and emission reduction benefit data includes: Retrieve fuel consumption and carbon emission prediction data for the selected optimal route, and also retrieve historical fuel consumption and carbon emission reference data for the shortest distance and shortest time routes. Based on the output of the elevation-curvature coupled fuel consumption prediction sub-model, the fuel savings and pollutant emission reductions of the optimal path compared to the traditional shortest distance path and shortest time path are quantitatively calculated, and the data on energy saving and emission reduction benefits are generated according to a preset format.

[0015] The beneficial effects of this invention are as follows: This invention constructs an elevation-curvature coupled fuel consumption prediction sub-model, integrates real vehicle heavy-load operating condition test data and engine bench test data for training and verification, and realizes coupled quantitative prediction of vehicle fuel consumption by altitude change and road curvature. Compared with the existing single-factor fuel consumption prediction method, it improves the accuracy of fuel consumption prediction, provides data support for green path optimization, and effectively reduces vehicle fuel consumption and pollutant emissions.

[0016] This invention constructs a multi-objective optimization function with the core objectives of minimizing fuel consumption and carbon emissions. It integrates multi-dimensional constraints such as vehicles, roads, weather, and traffic control, and combines graph neural networks to construct a path planning sub-model. This enables deep traversal of the entire road network and selection of feasible paths. Compared with traditional path planning algorithms, it has stronger adaptability and can meet the vehicle path planning needs in different scenarios, especially adapting to the driving characteristics of heavy-duty vehicles.

[0017] This invention introduces a dynamic weight allocation mechanism, using the elevation-curvature coupled fuel consumption prediction value as the core weight factor. Combined with the real-time status of the road network, the weight parameters are dynamically adjusted. This allows for flexible adjustment of the comprehensive cost calculation strategy based on changes in road congestion, weather conditions, and other factors. This makes the comprehensive ranking of alternative routes more aligned with actual driving needs, improving the rationality and flexibility of route optimization.

[0018] This invention establishes a real-time road network status monitoring and large model iterative update mechanism, which can capture changes in road network status in real time, trigger the route replanning process in a timely manner, dynamically adjust the optimal route, ensure that the optimized route always adapts to the current road network status, avoid route failure or increased fuel consumption due to road network changes, and improve the timeliness and reliability of route planning. Attached Figure Description

[0019] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0020] Figure 1 This is a diagram illustrating the overall technical architecture of the elevation-curvature coupled fuel consumption model method for vehicle path optimization according to the present invention. Figure 2 This is a flowchart of the training and verification process for the elevation-curvature coupled fuel consumption prediction sub-model in this invention. Detailed Implementation

[0021] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0022] Please see Figures 1-2 A method for a large fuel consumption model using elevation-curvature coupling for vehicle path optimization, comprising: S1: Acquire multi-source heterogeneous data on vehicle routes, complete standardization processing, and generate a multi-dimensional standardized dataset; S2: Using the multidimensional standardized dataset as input features, a training sample set is constructed by fusing real vehicle heavy-load working condition test data and engine bench test data to train and verify the elevation-curvature coupled fuel consumption prediction sub-model. S3: Construct a multi-objective optimization function with the core objectives of minimizing fuel consumption and carbon emissions, integrate multi-dimensional constraints, construct a path planning sub-model based on graph neural network, traverse the entire road network from the starting point to the end point to generate alternative paths, introduce a dynamic weight allocation mechanism, and complete the weighted calculation of the comprehensive cost of each alternative path. S4: Based on the multi-objective optimization function and the comprehensive cost weighted calculation results, sort the candidate paths and select the optimal path; monitor the dynamic changes in the road network status in real time, trigger the large model iterative update mechanism, dynamically adjust the optimal path, and synchronously quantify and output energy-saving and emission-reduction benefit data.

[0023] Specifically, the standardization process includes: First, multi-source heterogeneous data on vehicle routes is synchronously acquired through a multi-source data acquisition interface. This multi-source heterogeneous data includes vehicle basic parameter data, road network GIS spatial data, road geometric feature data, terrain elevation data, meteorological environmental data, and traffic control data. Vehicle basic parameter data includes vehicle model, curb weight, load parameters, and engine performance parameters. Road network GIS spatial data includes road segment coordinates, road grade, and road network connectivity. Road geometric feature data includes road curvature, slope, and number of lanes. Terrain elevation data includes road segment elevation changes and elevation gradients. Meteorological environmental data includes temperature, precipitation, wind speed, and visibility. Traffic control data includes traffic restriction rules, toll fees, and speed limits. The collected data undergoes integrity verification, and invalid data with a missing data rate exceeding a preset threshold is removed. Verified data is then stored in a preset database. Finally, the data is categorized and summarized according to data type, and a unique identifier and associated index are established for each data category. This system enables rapid access and related queries for various types of data. A standardized processing flow is initiated, performing preprocessing operations on each type of data sequentially. A coordinate transformation algorithm is invoked to unify the coordinate system of spatial data from multi-source heterogeneous data, converting spatial coordinate data from different geographic coordinate systems into a preset unified geographic coordinate system, eliminating coordinate deviations caused by different coordinate systems. Data format normalization is performed on various types of data, unifying data encoding formats, numerical units, and data storage formats. Data structuring algorithms are used to convert unstructured data into structured data, ensuring a unified and identifiable data format. An outlier detection algorithm is used to systematically detect each type of data, setting outlier judgment thresholds to identify and remove abnormal data generated during data collection due to equipment failure, transmission interference, or environmental interference. Missing items after removing outlier data are filled using an interpolation completion algorithm. Finally, a multidimensional standardized dataset with unified format, meeting accuracy standards, and free of outlier data is generated and synchronously stored in a dedicated database for model training.

[0024] Specifically, the process of constructing the training sample set includes: A feature extraction algorithm is invoked to accurately extract input features suitable for elevation-curvature coupled fuel consumption prediction from the multidimensional standardized dataset. These input features include altitude change parameters, road curvature features, vehicle load, and vehicle speed. The altitude change parameters are the elevation difference and rate of elevation change of road segments, and the road curvature features are the curvature value and rate of curvature change of road segments. After extraction, the features are validated, and redundant features irrelevant to fuel consumption prediction are removed. The validated input features are normalized using a Min-Max normalization algorithm to map feature values ​​to a preset range, eliminating dimensional differences between different features and preventing interference from feature value differences in model training. Simultaneously, real-world heavy-load test data and engine bench test data are retrieved. The verification data includes actual vehicle heavy-load operating condition test data such as vehicle speed, fuel consumption, engine speed, and throttle opening data under different altitudes and road curvatures, and engine bench test data such as engine fuel consumption and emissions data under different loads and speeds. The two types of data are processed for format matching, unifying data encoding, numerical units, and timestamp formats. The logical consistency of the two types of data is compared through a consistency check algorithm, and data conflicts, duplicates, and invalid data are eliminated. The preprocessed input features, actual vehicle heavy-load operating condition test data, and engine bench test data are integrated according to a preset ratio, and the training sample set is divided into a training part and a verification part. The sample labels are defined, the training sample set is completed, and it is stored in the model training sample library.

[0025] Specifically, the training and validation process of the elevation-curvature coupled fuel consumption prediction sub-model includes: The training sample set is randomly divided into a training set and a validation set according to a preset ratio. This ratio ensures uniform sample distribution and avoids sample bias affecting model training performance. An initial elevation-curvature coupled fuel consumption prediction sub-model is constructed based on a deep learning algorithm. The network structure of the input layer, hidden layer, and output layer is defined, and model iteration parameters such as the number of neurons in the hidden layer, activation function type, iteration count, and learning rate are set. The preprocessed input features are used as model input, and the vehicle's fuel consumption per unit mileage and carbon emissions are used as model output. The model training process is then initiated, and the training set is used to train the model. Iterative training is performed, with the model training error calculated in real time during each iteration. The training error is calculated using mean squared error (MSE) or mean absolute error (MAE). The backpropagation algorithm is used to adjust the model weight parameters and bias terms to optimize the model fitting effect. After each iteration, the validation set is input into the model being trained, the model prediction error is calculated, and the prediction error is compared with a preset error threshold, which is no greater than 5%. Iterative training continues until the prediction error is lower than the preset error threshold. At this point, model training stops, the parameters of the trained model are fixed, and the model is stored in a model library for subsequent steps.

[0026] Specifically, the process of constructing the multi-objective optimization function includes: The core requirements of multi-objective optimization are clearly defined, with minimizing total vehicle fuel consumption and total carbon emissions as the core optimization objectives. Travel time and toll costs are introduced as auxiliary optimization objectives. The definitions and calculation methods for each optimization objective are clarified. Total vehicle fuel consumption is the sum of fuel consumption across all road segments; total carbon emissions are calculated based on fuel consumption and carbon emission factors; travel time is the sum of travel time across all road segments; and toll costs include road tolls, fuel costs, and other related expenses. Based on the importance of each optimization objective and the actual needs of vehicle route optimization, weighting coefficients are set for each objective. The weighting coefficient for the core optimization objective is higher than that for the auxiliary optimization objectives to ensure... The optimization direction aligns with green and energy-saving requirements. A weighted summation method is used to transform the multi-objective optimization problem into a single-objective optimization problem. The mathematical expression of the optimization function is clearly defined, along with the optimization variables, objective function, and boundary conditions. The optimization variables are path-related parameters, including path segment composition, travel speed, and travel time. The boundary conditions are the quantification thresholds for various constraints, including vehicle constraints, road constraints, weather constraints, and traffic control constraints. The constructed optimization function is validated for rationality, verifying the correctness and feasibility of its expression. The optimization variables and weighting coefficients are adjusted to ensure the optimization function accurately quantifies the optimization effect of each alternative path.

[0027] Specifically, the process of constructing a path planning sub-model based on a graph neural network includes: The entire road network data is preprocessed to remove invalid and impassable road segments, and the processed road network is defined as a graph structure. Key nodes such as road intersections and entrances / exits are defined as nodes in the graph structure, and each node is assigned a unique identifier and coordinate information. These key nodes also include service areas, toll stations, and turning points at intersections. Road segments between adjacent key nodes are defined as edges in the graph structure. Fuel consumption, travel time, toll fees, and quantified constraints are extracted for each road segment. These quantified constraints include vehicle weight limits, road speed limits, restricted driving periods, and weather restrictions. These parameters are used as the graph structure. Edge attribute information is associated with corresponding road segments to establish the relationship between nodes and edges; the graph neural network model parameters are initialized, and key parameters such as the number of training iterations and learning rate are set. The completed global road network graph structure is input into the graph neural network model; the graph neural network algorithm is started to perform a depth-first traversal of the global road network from the starting point to the ending point, and a path search algorithm is used to search for all feasible paths that meet the constraints; the searched feasible paths are deduplicated, duplicate paths are removed, and paths that do not meet the core constraints are filtered out. Paths that meet the requirements are retained as candidate paths, and the edge attribute information and overall path parameters of each candidate path are stored simultaneously.

[0028] Specifically, the dynamic weight allocation mechanism includes the following process: The elevation-curvature coupled fuel consumption prediction sub-model is invoked, and relevant parameters of each candidate path segment are input to obtain the predicted fuel consumption value for each segment. This predicted fuel consumption value is used as the core weight factor for comprehensive cost calculation, and an initial weight value is set for the core weight factor. Based on the travel time, toll cost, and driving safety factor of each segment, corresponding auxiliary weight factors are set, and the initial weight values ​​of each auxiliary weight factor are clarified to ensure that the sum of the weights of the core weight factor and the auxiliary weight factors is 1. Road network status data, including parameters such as road segment congestion level and weather conditions, are obtained in real time through the road network status monitoring interface. The road segment congestion level is divided into smooth and lightly congested. The system employs four levels of congestion: moderate, severe, and heavy. Weather conditions include sunny, cloudy, rainy, snowy, and foggy weather. The acquired data is analyzed in real-time to assess changes in the road network status. Based on the degree of congestion and changes in weather conditions, the values ​​of core and auxiliary weight factors are dynamically adjusted. The weight of travel time is appropriately increased in congested sections and under severe weather conditions, while the weight of core weight factors is increased in energy-sensitive scenarios. A weighted summation algorithm is used to calculate the comprehensive cost value of each alternative route by weighting the fuel consumption, travel time, toll, and driving safety coefficient according to the adjusted weights. This comprehensive cost value is then stored in the route cost database.

[0029] Specifically, the process of sorting the candidate paths and selecting the optimal path includes: The relevant parameters of each candidate route are retrieved, including fuel consumption, carbon emissions, travel time, toll fees, and driving safety factors for each road segment. These parameters are then input into the multi-objective optimization function to calculate the optimization target values ​​for each candidate route, including fuel consumption, carbon emissions, travel time, and toll fees. The comprehensive cost values ​​of each candidate route are retrieved from the route cost database. The optimization target values ​​and comprehensive cost values ​​are then merged, and a multi-objective ranking algorithm is used to comprehensively rank all candidate routes, generating a route ranking list. Based on the ranking list, the route with the lowest fuel consumption, optimal carbon emissions, and lowest comprehensive cost is selected as the candidate optimal route. The feasibility of the candidate optimal route is verified by calling real-time road network status data and constraint parameters to verify whether the candidate optimal route meets all constraints and is suitable for the current road network status. If any non-compliance is found, the next route is selected from the ranking list until the optimal route that meets all requirements is determined.

[0030] Specifically, the monitoring process for the dynamic changes in the road network status includes: The real-time road network status monitoring module is activated, continuously collecting road network status data through traffic monitoring data interfaces, meteorological data interfaces, and traffic control update interfaces. This includes data related to road segment congestion status, road construction status, changes in weather conditions, and updates to traffic control rules. The road segment congestion status data includes average vehicle speed, traffic volume, and congestion duration. The road construction status data includes the scope of the construction section, construction duration, and detour suggestions. The traffic control rule updates include adjustments to traffic restrictions, speed limits, and toll fees. The collected data is analyzed in real time to extract key status parameters. These key status parameters are compared one by one with preset normal ranges to determine whether the road network status has changed. If a change in the road network status is detected, and this change affects the overall cost or feasibility of alternative routes, the large model iterative update mechanism is triggered. After the iterative update mechanism is activated, the parameter adjustment of the multi-objective optimization function, the full-domain road network traversal of the path planning sub-model, the generation of alternative routes, and the weighted calculation of overall costs are re-executed to complete the path update.

[0031] Specifically, the process of quantifying and outputting the energy conservation and emission reduction benefit data includes: Retrieve the fuel consumption and carbon emission prediction data of the optimal route from the model prediction result database to determine the total fuel consumption and total carbon emissions of the optimal route. The carbon emissions include CO2 and NO. x The system calculates emissions of pollutants such as PM2.5; it also retrieves historical fuel consumption and carbon emission reference data for the shortest distance and shortest time paths to ensure consistency between the reference data and the driving conditions and vehicle parameters of the optimal path. Driving conditions include driving time, weather conditions, and vehicle load; vehicle parameters include vehicle model and engine performance. Based on the output of the elevation-curvature coupled fuel consumption prediction sub-model, a difference calculation algorithm is used to quantify the fuel savings, CO2 and other pollutant emission reductions of the optimal path compared to the traditional shortest distance and shortest time paths. Fuel savings are calculated based on the difference in fuel consumption per 100 kilometers and the total path mileage; pollutant emission reductions are calculated based on the fuel consumption difference and the corresponding pollutant emission factors. The calculated energy-saving and emission-reduction data are then verified for accuracy, data with calculation errors exceeding a preset range are removed, and calculation deviations are corrected. The verified energy-saving and emission-reduction data are then organized according to a preset format, clarifying the specific values ​​and calculation basis of each emission reduction indicator, generating standardized energy-saving and emission-reduction benefit data, and simultaneously outputting it to the user terminal and data storage module.

[0032] This embodiment provides a method for a high fuel consumption model using elevation-curvature coupling for vehicle routing optimization, and describes the model construction process. The specific implementation process is as follows: S1: Acquisition and Standardization Processing of Multi-Source Heterogeneous Data Multi-source heterogeneous data on vehicle routes are synchronously acquired through a multi-source data acquisition interface. This multi-source heterogeneous data includes vehicle basic parameter data, road network GIS spatial data, road geometric feature data, terrain elevation data, meteorological environment data, and traffic control data. Among them, vehicle basic parameter data includes vehicle model, curb weight m, load parameter m_z, and engine performance parameters (power P_f, torque T_f); road network GIS spatial data includes road segment coordinates (x,y,z), road grade L_d, and road network connection relationship M_l; road geometric feature data includes road curvature k, slope θ, and number of lanes N_c; terrain elevation data includes road segment elevation change ΔH and elevation gradient G_h; meteorological environment data includes temperature t_q, precipitation intensity I_j, wind speed v_f, and visibility V_n; and traffic control data includes restricted driving periods T_j, toll fees C_t, and speed limit requirements v_x.

[0033] The collected data undergoes integrity verification, discarding invalid data with a missing data rate exceeding a preset threshold P1 (0 < P1 < 1), and storing the verified data in a preset database. The data is then categorized and summarized according to data type, with each category having a unique identifier and associated index to enable rapid retrieval and related queries. A standardized processing flow is initiated, performing preprocessing operations on each data category sequentially: a coordinate transformation algorithm is invoked to unify the coordinate system of spatial data from multi-source heterogeneous data, converting spatial coordinate data from different geographic coordinate systems to a preset unified geographic coordinate system (such as the WGS84 coordinate system) to eliminate coordinate biases caused by different coordinate systems; and further processing is performed on each data category. The data format is normalized to unify the data encoding format, numerical units, and data storage format. Unstructured data (such as meteorological text data) is converted into structured data using data structuring algorithms to ensure that the data format is uniform and recognizable. Outlier detection algorithms (such as the 3σ criterion) are used to traverse and detect various types of data one by one. An outlier judgment threshold T1 is set to identify and remove abnormal data caused by equipment failure, transmission interference, and environmental interference during the data acquisition process. For missing items after removing abnormal data, interpolation completion algorithms (such as linear interpolation) are used to fill in the missing items. Finally, a multidimensional standardized dataset D_b with uniform format, qualified accuracy, and no abnormal data is generated and synchronously stored in a dedicated database for model training.

[0034] S2: Construction of training sample set and training and validation of elevation-curvature coupled fuel consumption prediction sub-model 2.1 Construction of Training Sample Set A feature extraction algorithm (such as PCA feature extraction algorithm) is called to accurately extract input features for elevation-curvature coupled fuel consumption prediction from the multidimensional standardized dataset D_b generated by S1. The input features include altitude change parameters, road curvature features, vehicle load and driving speed. The altitude change parameters are the road segment altitude difference ΔH and the altitude change rate v_H (v_H=ΔH / Δt, where Δt is the driving time of the road segment). The road curvature features are the road segment curvature value k and the curvature change rate v_k (v_k=Δk / Δt). After extraction, the features are validated. The correlation coefficient r between each feature and fuel consumption is calculated using a correlation analysis algorithm. Redundant features with a correlation coefficient |r| < r0 (r0 is a preset correlation coefficient threshold) are removed. The validated input features are normalized. The Min-Max normalization algorithm is used to map the feature values ​​to the preset interval [0,1]. The normalization process is as follows: x_norm=(x_raw-xmin) / (xmax-xmin) Where x_norm is the normalized feature value, x_raw is the original feature value, xmin is the minimum value of the feature, and xmax is the maximum value of the feature, in order to eliminate the difference in units between different features and avoid the interference of feature numerical differences on model training.

[0035] Simultaneously retrieve the actual vehicle heavy-load operating condition measured data D_s and the engine bench test data D_t. The actual vehicle heavy-load operating condition measured data D_s includes vehicle speed v_s, fuel consumption f_s, engine speed n_s, and throttle opening α_y under different altitudes and road curvatures. The engine bench test data D_t includes engine fuel consumption f_t under different loads L_f and different engine speeds n_t, and emission data E_t (including CO2 and NO). x (Emissions); Format matching processing is performed on the two types of data to unify data encoding, numerical units, and timestamp formats. The logical consistency of the two types of data is compared through a consistency check algorithm (such as the deviation of engine fuel consumption at the same speed not exceeding Δf_y), and data conflicts, duplicates, and invalid data are eliminated. The preprocessed input features, actual vehicle heavy-load working condition measured data D_s, and engine bench test data D_t are integrated according to a preset ratio R1 (0.7≤R1≤0.8). The training sample set is divided into a training part D_x and a verification part D_j. The sample labels are defined (labels are vehicle fuel consumption per unit mileage f_d and carbon emissions E_tan). The construction of the training sample set D_y is completed and stored in the model training sample library.

[0036] 2.2 Construction, Training and Validation of Elevation-Curvature Coupled Fuel Consumption Prediction Sub-model 2.2.1 Model Construction An initial elevation-curvature coupled fuel consumption prediction sub-model M_y is constructed based on a deep learning algorithm (CNN algorithm is used in this embodiment). The network structure of the input layer, hidden layer and output layer is defined: the number of neurons in the input layer is equal to the dimension N_t of the preprocessed input features, the hidden layer has N_y neurons (N_y is a preset positive integer), the activation function is the ReLU function, and the output layer has 2 neurons, corresponding to the fuel consumption f_d per unit mileage and carbon emissions E_tan, respectively. The model iteration parameters are set, including the number of iterations N_d (100≤N_d≤500), the learning rate η (0.001≤η≤0.01), and the regularization coefficient λ (0<λ<1), to avoid model overfitting.

[0037] The model loss function L_s is constructed and calculated using mean squared error (MSE) to measure the deviation between the model's predicted values ​​and the actual values. The process is as follows: L_s=(1 / N_y)×Σ[(f_y_i-f_s_i) 2 +(E_y_i-E_tan_i) 2 ] Where N_y is the total number of training samples, f_y_i is the predicted fuel consumption of the i-th sample, f_s_i is the actual fuel consumption of the i-th sample, E_y_i is the predicted carbon emission of the i-th sample, and E_tan_i is the actual carbon emission of the i-th sample.

[0038] 2.2.2 Model Training The training portion D_x of the training sample set is input into the initial elevation-curvature coupled fuel consumption prediction sub-model M_y to start the model training process: In each iteration, the model input layer receives the preprocessed input features, performs feature extraction and coupling operations through the hidden layer, and the output layer outputs the predicted values ​​of fuel consumption and carbon emissions; the model training loss L_s is calculated in real time, and the backpropagation algorithm (BP algorithm) is used to adjust the model weight parameters W and the bias term b to optimize the model fitting effect; after each N_j iteration (N_j is the preset interval number of iterations), the rate of decrease of the current training loss v_s (v_s=|L_s_prev-L_s_curr| / N_j) is calculated. If v_s < v_s_th (v_s_th is the preset loss decrease rate threshold), the learning rate η is appropriately increased to accelerate model convergence; the training continues iteratively until the preset number of iterations N_d is reached, or the training loss L_s < L_s_th (L_s_th is the preset loss threshold), at which point the model training stops.

[0039] 2.2.3 Model Validation The validation portion D_j of the training sample set is input into the trained elevation-curvature coupled fuel consumption prediction sub-model M_y, and the fuel consumption prediction value f_y and carbon emission prediction value E_y of each validation sample are output; the model prediction error ε is calculated using the mean absolute error (MAE), as follows: Where N_j is the total number of validation samples; if the prediction error ε≤ε_th (ε_th is the preset error threshold, ε_th≤5%), the model validation is successful, the model parameters are solidified, and the final elevation-curvature coupled fuel consumption prediction sub-model My_g is obtained and stored in the model library for subsequent steps; if ε>ε_th, the model iteration parameters are adjusted (such as increasing the number of iterations N_d, adjusting the learning rate η), and the model training process is re-executed until the validation is successful.

[0040] S3: Construction of multi-objective optimization function, construction of path planning sub-model and calculation of overall cost. 3.1 Construction of Multi-Objective Optimization Function The core and auxiliary objectives of the multi-objective optimization are clearly defined. The core optimization objectives are minimizing the total fuel consumption F_z and the total carbon emissions E_z. The auxiliary optimization objectives are travel time T_z and travel cost C_z. According to the importance of each optimization objective, weighting coefficients are set for each optimization objective. The sum of the weighting coefficients of the core optimization objective is ω_h (0.6≤ω_h≤0.8), and the sum of the weighting coefficients of the auxiliary optimization objectives is ω_f (0.2≤ω_f≤0.4), and ω_h+ω_f=1. The specific weighting coefficients are set as follows: fuel consumption weight ω_y, carbon emissions weight ω_t, travel time weight ω_T, and travel cost weight ω_C, satisfying ω_y+ω_t=ω_h and ω_T+ω_C=ω_f.

[0041] The multi-objective optimization problem is transformed into a single-objective optimization problem using a weighted summation method, and the multi-objective optimization function F_opt is constructed as follows: The process for each optimization objective is as follows: 1. Total fuel consumption of vehicle driving F_z: Call the elevation-curvature coupled fuel consumption prediction sub-model My_g to obtain the fuel consumption per unit mileage f_d_j of each road segment (j is the segment number, j=1,2,...,N_sec), and combine it with the length of each road segment L_d_j to calculate F_z=Σ(f_d_j×L_d_j); 2. Total carbon emissions E_z: Based on total fuel consumption F_z and carbon emission factor k_E (carbon emissions per unit of fuel consumption), calculate E_z = F_z × k_E; 3. Travel time T_z: Combining the length of each road segment L_d_j and the travel speed of each road segment v_d_j, calculate the travel time t_d_j for each road segment = L_d_j / v_d_j, and sum them up to get T_z = Σt_d_j; 4. Toll C_z: Summarize the toll C_d_j for each segment of the route, and calculate C_z = ΣC_d_j + F_z × k_y (k_y is the fuel cost per unit of fuel consumption).

[0042] The optimization variables are defined as path-related parameters X=(L_d_j,v_d_j,C_d_j,f_d_j), and the constraint boundary conditions are the quantization thresholds of various constraints. The multidimensional constraints include: vehicle constraints (load m_z≤m_z_th, vehicle speed v_d_j≤v_x), road constraints (slope θ≤θ_th, curvature k≤k_th), meteorological constraints (wind speed v_f≤v_f_th, visibility V_n≥V_n_th), and traffic control constraints (the travel time is not in T_j). The rationality of the constructed optimization function F_opt is verified to confirm the correctness and feasibility of the function expression. The optimization variables and weighting coefficients are adjusted to ensure that the optimization function can quantify the optimization effect of each alternative path.

[0043] 3.2 Construction of a Path Planning Sub-model Based on Graph Neural Networks The path planning sub-model M_l is constructed based on the graph neural network algorithm (GAT algorithm is used in this embodiment). The specific construction process is as follows: 1. Road Network Graph Structure Construction: The global road network data after S1 processing is preprocessed to remove invalid and impassable road segments. The processed global road network is defined as a graph structure G=(V,E), where V is the set of nodes and E is the set of edges. Key nodes such as road intersections, entrances / exits, service areas, and toll stations in the road network are defined as nodes v_i (i=1,2,...,N_node) of the graph structure, and each node is assigned a unique identifier and coordinate information (x_i,y_i,z_i). The road segments between adjacent key nodes are defined as edges e_ij (connecting nodes v_i and v_j) of the graph structure, and each edge is assigned a unique identifier.

[0044] 2. Edge Attribute Assignment: Extract relevant parameters for each road segment (edge ​​e_ij), including road segment fuel consumption f_d_ij, road segment travel time t_d_ij, and road segment travel cost C_d_ij obtained through the elevation-curvature coupled fuel consumption prediction sub-model My_y_g, as well as quantified constraints (slope threshold θ_th_ij, curvature threshold k_th_ij, speed limit v_x_ij). Use these parameters as attribute information A_ij for the graph structure edge e_ij, associate them with the corresponding road segment, and establish the association between nodes and edges.

[0045] Model initialization and training: Initialize the graph neural network model parameters, set the model iteration count N_iter_l and learning rate η_l, and input the constructed global road network graph structure G=(V,E) and edge attribute information A_ij into the graph neural network model; with the path optimization objective (minimizing F_opt) as the training objective, iteratively train the model, adjust the model weight parameters, until the model can accurately identify the road network structure and edge attribute relationships, and complete the construction of the path planning sub-model M_l.

[0046] 4. Alternative Path Generation: Input the coordinates (x_S, y_S, z_S) of the starting point S and the coordinates (x_E, y_E, z_E) of the ending point E into the path planning sub-model M_l, start the graph neural network algorithm, and use the A* path search algorithm to perform a depth-first traversal of the entire road network from the starting point to the ending point to search for all feasible paths that meet the constraints; perform deduplication on the searched feasible paths, remove duplicate paths, and filter out paths that do not meet the core constraints (such as speed exceeding the speed limit or gradient exceeding the threshold), retaining N_b paths (N_b is a preset positive integer) that meet the requirements as alternative paths P_b={P_b1,P_b2,...,P_bN_b}, and simultaneously store the edge attribute information and overall path parameters of each alternative path.

[0047] 3.3 Dynamic weight allocation and comprehensive cost calculation A dynamic weight allocation mechanism is introduced, and the specific process is as follows: 1. Initial Weight Setting: The elevation-curvature coupled fuel consumption prediction sub-model My_y_g is invoked to obtain the fuel consumption prediction value f_d_ij for each candidate path P_bj corresponding to the road segment. This fuel consumption prediction value is used as the core weight factor ω_hz for comprehensive cost calculation, and the initial weight value ω_hc (0.5≤ω_hc≤0.7) of the core weight factor is set. Based on the travel time t_d_ij, toll cost C_d_ij, and driving safety factor S_a for each road segment... _ij (S_a_ij is calculated based on slope, curvature, and weather conditions, S_a_ij∈[0,1]), and set corresponding auxiliary weight factors respectively: travel time weight ω_Tc (0.15≤ω_Tc≤0.25), travel cost weight ω_Cc (0.1≤ω_Cc≤0.2), and driving safety factor weight ω_Sc (0.05≤ω_Sc≤0.1), to ensure ω_hc+ω_Tc+ω_Cc+ω_Sc=1.

[0048] Dynamic weight adjustment: Real-time road network status data D_l is acquired through the road network status monitoring interface, including road segment congestion level K_y (K_y∈[0,1], the larger the value, the more severe the congestion) and meteorological condition level K_q (K_q∈[1,5], the higher the level, the worse the meteorological conditions). The acquired data is analyzed in real time, and the values ​​of each weight factor are dynamically adjusted according to changes in road network congestion level and meteorological conditions: When K_y>K_y_th (K_y_th is the preset congestion threshold) or K_q≥K_q_th (K_q_th is the preset meteorological level threshold), the travel time weight ω_T and driving safety factor weight ω_S are increased, and the core weight factor ω_hz is decreased; when K_y≤K_y_th and K_q<K_q_th, the core weight factor ω_hz is increased, and the auxiliary weight factors remain unchanged; after adjustment, it is ensured that the sum of all weight factors is still 1.

[0049] 3. Comprehensive Cost Calculation: A weighted summation algorithm is used to calculate the comprehensive cost C_zj for each alternative route P_bj by weighting the fuel consumption F_zj, travel time T_zj, toll C_zj, and driving safety factor S_aj according to the adjusted weights. The calculation process is as follows: (1-S_a_j) is used to convert the safety factor into safety cost. The larger the value, the higher the safety risk and the higher the overall cost. The calculated overall cost C_z_j of each alternative path is stored in the path cost database.

[0050] S4: Optimal route selection, road network status monitoring, and quantitative output of energy conservation and emission reduction benefits. 4.1 Ranking of candidate paths and selection of the optimal path 1. Path Comprehensive Ranking: Retrieve the relevant parameters of each candidate path P_bj, input the parameters of each candidate path one by one into the multi-objective optimization function F_opt constructed by S3, and calculate the optimization objective value F_optj corresponding to each candidate path; retrieve the comprehensive cost C_zj of each candidate path from the path cost database, merge the optimization objective value F_optj with the comprehensive cost C_zj, and use the TOPSIS multi-objective ranking algorithm to comprehensively rank all candidate paths, generating a path ranking list L_p. The ranking priority is that the smaller F_optj and the smaller C_zj, the higher the path ranking.

[0051] 2. Optimal Path Selection: Based on the path ranking list L_p, select the path with the smallest F_y_j and C_z_j as the candidate optimal path P_h; perform feasibility verification on the candidate optimal path P_h by calling the real-time road network status data D_l and constraint parameters to verify whether the candidate optimal path meets all constraints and is suitable for the current road network status (e.g., no construction, no severe congestion); if the candidate optimal path P_h meets all requirements, it is determined as the final optimal path P_y; if there are any non-compliance items, select the next path from the path ranking list L_p and repeat the verification process until the optimal path P_y that meets all requirements is determined.

[0052] 4.2 Dynamic monitoring of road network status and iterative updates of the large model 1. Road Network Status Monitoring: Activate the real-time road network status monitoring module. Continuously collect road network status data D_l through traffic monitoring data interfaces, meteorological data interfaces, and traffic control update interfaces. This includes road segment congestion status (congestion level K_y, congestion duration T_y), road construction status (construction section range S_s, construction duration T_s), changes in meteorological conditions (temperature t_q, wind speed v_f), and updates to traffic control rules (adjustments to traffic restrictions, speed limits, and toll fees). The collected data is analyzed in real time to extract key status parameters. These parameters are then compared one by one with preset normal ranges to determine whether the road network status has changed.

[0053] 2. Large Model Iterative Update: If a change in the road network status is detected, and this change affects the comprehensive cost C_z_j or feasibility of the alternative paths (e.g., construction makes the road section impassable, or congestion significantly increases travel time), the large model iterative update mechanism is triggered. After the iterative update mechanism is activated, the multi-objective optimization function parameter adjustment (adjustment of weighting coefficients) of S3, the full-domain road network traversal of the path planning sub-model M_l, the generation of alternative paths, and the comprehensive cost weighting calculation process are re-executed. Then, the alternative path sorting and optimal path selection process of S4.1 is executed to achieve dynamic adjustment of the optimal path and ensure that the output optimal path P_y always adapts to the current road network status.

[0054] 4.3 Quantitative Output of Energy Conservation and Emission Reduction Benefit Data 1. Data Retrieval: Retrieve the fuel consumption prediction data F_y_z and carbon emission prediction data E_y_z of the selected optimal path P_y from the model prediction result database; at the same time, retrieve the fuel consumption reference data F_j_z and carbon emission reference data E_j_z of the historical shortest distance path P_j, and the fuel consumption reference data F_s_z and carbon emission reference data E_s_z of the historical shortest time path P_s, to ensure that the reference data are consistent with the driving conditions (driving time period, weather conditions, vehicle load) and vehicle parameters (vehicle model, engine performance) of the optimal path.

[0055] 2. Quantitative Calculation: Based on the output of the elevation-curvature coupled fuel consumption prediction sub-model My_y_g, the difference calculation algorithm is used to quantify and calculate the energy saving and emission reduction data of the optimal path P_y compared to the traditional shortest distance path P_j and the shortest time path P_s. The specific calculation process is as follows: (1) Fuel savings ΔF_j=F_j_z-F_y_z, ΔF_s=F_s_z-F_y_z; (2) CO2 emission reduction ΔE_j_CO2=ΔF_j×k_E_CO2, ΔE_s_CO2=ΔF_s×k_E_CO2 (k_E_CO2 is the CO2 emission factor); (3) Other pollutant emission reductions ΔE_j_w = ΔF_j × k_E_w, ΔE_s_w = ΔF_s × k_E_w (k_E_w is the emission factor for other pollutants, such as NO) x PM).

[0056] Data Verification and Output: The calculated energy conservation and emission reduction data are verified for accuracy. The calculation error ε_j = |ΔF_calc - ΔF_meas| / ΔF_meas (ΔF_meas is a small number of actual vehicle measurement differences). Data with calculation errors ε_j > ε_j_th (ε_j_th ≤ 3%) are removed to correct the calculation deviation. The verified energy conservation and emission reduction data are organized according to a preset format, clarifying the specific values ​​and calculation basis of each emission reduction indicator, generating standardized energy conservation and emission reduction benefit data D_j, and synchronously output to the user terminal and data storage module for user viewing and industry management.

[0057] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for a high-fuel-consumption model using elevation-curvature coupling for vehicle path optimization, characterized in that, include: S1: Acquire multi-source heterogeneous data on vehicle routes, complete standardization processing, and generate a multi-dimensional standardized dataset; S2: Using the multidimensional standardized dataset as input features, a training sample set is constructed by fusing real vehicle heavy-load working condition test data and engine bench test data to train and verify the elevation-curvature coupled fuel consumption prediction sub-model. S3: Construct a multi-objective optimization function with the core objectives of minimizing fuel consumption and carbon emissions, integrate multi-dimensional constraints, construct a path planning sub-model based on graph neural network, traverse the entire road network from the starting point to the end point to generate alternative paths, introduce a dynamic weight allocation mechanism, and complete the weighted calculation of the comprehensive cost of each alternative path. S4: Based on the multi-objective optimization function and the comprehensive cost weighted calculation results, sort the candidate paths and select the optimal path; monitor the dynamic changes in the road network status in real time, trigger the large model iterative update mechanism, dynamically adjust the optimal path, and synchronously quantify and output energy-saving and emission-reduction benefit data.

2. The method according to claim 1, characterized in that, In S1, the specific process of the standardization process includes: Acquire multi-source heterogeneous data on vehicle routes, store it in a preset database, classify and summarize it according to data type, divide the data into categories and establish data association indexes; Initiate a standardized processing flow, perform preprocessing operations on various types of data in sequence, unify the coordinate system of spatial data in multi-source heterogeneous data, and convert spatial coordinate data under different geographic coordinate systems into a preset unified geographic coordinate system; perform format normalization processing on various types of data, and convert unstructured data into structured data. The data is traversed step by step to identify and remove abnormal data caused by equipment failure or transmission interference during the data acquisition process, thereby generating the multidimensional standardized dataset.

3. The method according to claim 1, characterized in that, In S2, the process of constructing the training sample set includes: The input features for elevation-curvature coupled fuel consumption prediction are extracted from the multidimensional standardized dataset, normalized, and the feature values ​​are mapped to a preset range. By integrating real vehicle heavy-load operating condition test data and engine bench test data, the two types of data are format matched and consistency verified, and the training sample set is constructed by integrating them according to a preset ratio.

4. The method according to claim 1, characterized in that, In S2, the training and validation process of the elevation-curvature coupled fuel consumption prediction sub-model includes: The training sample set is randomly divided into a training set and a validation set according to a preset ratio; an initial elevation-curvature coupled fuel consumption prediction sub-model is constructed, the model input layer, hidden layer and output layer structure are defined, and the model iteration parameters are set. The preprocessed input features are used as model input, and the vehicle's fuel consumption per unit mileage and carbon emissions are used as model output. The model training process is started, and iterative training is performed using the training set. The weight parameters are adjusted in real time, and the training error is calculated. After each round of training iterations, a validation set is input, and the prediction error is calculated. Training continues iteratively until the prediction error falls below a preset error threshold, at which point model training stops.

5. The method according to claim 1, characterized in that, In S3, the process of constructing the multi-objective optimization function includes: The core optimization objectives are minimizing total fuel consumption and total carbon emissions, with travel time and toll costs as auxiliary optimization objectives; weighting coefficients are set according to the importance of the optimization objectives. The multi-objective optimization problem is transformed into a single-objective optimization problem by using the weighted summation method, clarifying the mathematical expression of the optimization function, and defining the optimization variables, objective function, and boundary conditions.

6. The method according to claim 1, characterized in that, In S3, the process of constructing the path planning sub-model based on the graph neural network includes: The entire road network is defined as a graph structure, road intersections in the road network are defined as nodes in the graph structure, and road segments between adjacent intersections are defined as edges in the graph structure. The fuel consumption, travel time, travel cost, and quantified constraints of each road segment are extracted as attribute information of the graph structure edges and associated with the corresponding road segments. Initialize the graph neural network model parameters, input the global road network graph structure, start the graph neural network algorithm, perform a depth-first traversal of the global road network from the starting point to the end point, and search for all feasible paths that meet the constraints; perform deduplication and filtering on the searched feasible paths, remove duplicate paths and paths that do not meet the core constraints, and generate alternative paths.

7. The method according to claim 1, characterized in that, In S3, the specific process of the dynamic weight allocation mechanism includes: The elevation-curvature coupled fuel consumption prediction sub-model is invoked to obtain the fuel consumption prediction value of the corresponding road segment of the alternative route, which is used as the core weight factor for comprehensive cost calculation. The initial weight of the core weight factor is set. Based on the travel time, toll, and driving safety factor of each road segment, the corresponding auxiliary weight factor is set, and the initial weight of each auxiliary weight factor is clarified. Real-time road network status data is acquired, and the values ​​of core weight factors and auxiliary weight factors are dynamically adjusted according to the degree of road network congestion and weather conditions. The comprehensive cost value of each alternative route is generated by weighted summation.

8. The method according to claim 1, characterized in that, In S4, the process of sorting the candidate paths and selecting the optimal path includes: The parameters of each candidate path are input into the multi-objective optimization function to calculate the optimization objective value of each candidate path. The optimization objective value is then integrated with the comprehensive cost value to comprehensively rank all candidate paths. Based on the ranking results, the path with the lowest fuel consumption, optimal carbon emissions, and lowest overall cost is selected, and the feasibility of the selected path is verified.

9. The method according to claim 1, characterized in that, In S4, the monitoring process for the dynamic changes in the road network status includes: The acquired road network status data is analyzed in real time, compared with the preset normal range, and if a change in the road network status is detected, the large model iterative update mechanism is triggered, and the process of adjusting the multi-objective optimization function, traversing the road network of the path planning sub-model, generating alternative paths, and calculating the comprehensive cost weighting is re-executed.

10. The method according to claim 1, characterized in that, In S4, the process of quantifying and outputting the energy conservation and emission reduction benefit data includes: Retrieve fuel consumption and carbon emission prediction data for the selected optimal route, and also retrieve historical fuel consumption and carbon emission reference data for the shortest distance and shortest time routes. Based on the output of the elevation-curvature coupled fuel consumption prediction sub-model, the fuel savings and pollutant emission reductions of the optimal path compared to the traditional shortest distance path and shortest time path are quantitatively calculated, and the data on energy saving and emission reduction benefits are generated according to a preset format.

Images