Driving behavior classification and evaluation method and system based on multi-source operation data of new energy commercial vehicle

Abstract

The invention provides a driving behavior classification and evaluation method and system based on new energy commercial vehicle multi-source operation data, and belongs to the technical field of new energy commercial vehicle operation data processing.The method comprises the steps that vehicle networking time sequence, environment monitoring and business record data generated in new energy commercial vehicle operation are collected; performing time synchronization and space alignment to generate a fusion data set of a unified space-time reference; performing rule cleaning and deletion repair on the fused data set in sequence to generate cleaned structured data; based on the structured data set, high-order scene feature parameters are calculated, and an enhanced feature set is generated; inputting the enhanced feature set into a classification model, and identifying and outputting a driving behavior category; and based on the driving behavior categories and the corresponding enhanced feature set parameters, combined with safety and economy evaluation rules, calculating a driving behavior comprehensive score, and generating an evaluation report. The data quality is improved, the driving behavior is accurately identified, the energy consumption and safety are optimized, and intelligent management is realized.

Description

Technical Field

[0001] This application belongs to the field of new energy commercial vehicle operation data processing technology, specifically involving a driving behavior classification and evaluation method and system based on multi-source operation data of new energy commercial vehicles. Background Technology

[0002] With the rapid development of new energy vehicles, the operational scale of new energy commercial vehicles continues to expand, generating a wealth of vehicle network data, environmental data, and business data, providing a data foundation for the accurate identification and assessment of driving behavior. Driving behavior classification and assessment are of great significance for improving vehicle operational safety, optimizing energy consumption management, and achieving intelligent scheduling.

[0003] Currently, most existing driving behavior analysis methods are based on traditional gasoline-powered vehicle data or general passenger car driving models, often employing rule-based thresholding or simple statistical models for discrimination, such as rapid acceleration and deceleration. However, new energy commercial vehicles have unique characteristics in terms of power systems (such as electric motor drive and regenerative braking), load variations, and operating conditions (such as heavy-load hill climbing and frequent start-stop in urban areas), making it difficult for general methods to accurately capture their behavioral characteristics. Furthermore, existing methods often directly use raw data or only perform simple cleaning, ignoring issues such as multi-source heterogeneity, spatiotemporal inconsistencies, noise, and missing data, leading to insufficient feature extraction, low accuracy in behavior classification, and insufficient reliability of evaluation results.

[0004] Therefore, there is an urgent need for a driving behavior classification and evaluation method specifically for multi-source operational data of new energy commercial vehicles, in order to reflect a high-quality, strongly representative feature system of commercial vehicle operation scenarios, and to establish a high-precision behavior recognition and evaluation model to support refined operation management. Summary of the Invention

[0005] In a first aspect, embodiments of this application provide a method for classifying and evaluating driving behavior based on multi-source operational data of new energy commercial vehicles, comprising the following steps: S1. Collect vehicle network time-series data, environmental monitoring data and business record data generated during the operation of new energy commercial vehicles, synchronize and spatially align the collected data, and perform fusion processing based on Kalman filter to generate a fusion dataset with a unified spatiotemporal reference. S2. Perform rule cleaning and missing data repair on the fused dataset in sequence to generate cleaned structured data; S3. Based on the structured dataset, extract and calculate high-order scene feature parameters to generate an enhanced feature set; the high-order scene feature parameters include at least the energy consumption fluctuation rate per unit mileage, the average load rate-speed coupling coefficient, and the typical braking energy recovery efficiency. S4. Input the enhanced feature set into the trained classification model to identify and output driving behavior categories; the behavior categories include at least smooth driving, aggressive driving, economical driving, and heavy-load hill climbing mode; S5. Based on the driving behavior category and the corresponding enhanced feature set parameters, combined with the preset safety and economy assessment rules, calculate the comprehensive driving behavior score and generate an assessment report.

[0006] Furthermore, in step S1, the vehicle-to-everything (V2X) timing data includes vehicle GPS positioning data, vehicle speed, battery SOC, and motor torque; environmental monitoring data includes weather, road conditions, and traffic congestion index; and business record data includes load and trip start and end times.

[0007] Furthermore, the specific steps of step S1 are as follows: S11. Perform time synchronization, using UTC time as the reference, synchronize the timestamps of each data source at the millisecond level, and establish a unified time series; S12. Perform spatial alignment. For the collected GPS positioning data, use linear interpolation to perform spatial alignment to match the sampling frequency of the remaining data. The interpolation formula is:

[0008] Where t represents the specific time at which location interpolation needs to be performed; t1 and t2 represent the timestamps of two adjacent valid GPS positioning points, and satisfy the following conditions: ; This represents the spatial position coordinates calculated by interpolation at time t; , These represent the valid GPS positioning coordinates collected at times t1 and t2, respectively. S13. Using a Kalman filter, perform fusion and noise reduction processing on the various data sources after time synchronization and spatial alignment, and output a fused dataset with unified spatiotemporal reference.

[0009] Furthermore, the specific steps of step S2 are as follows: S21. For the vehicle network time series data in the fused dataset, perform hard cleaning of anomaly detection and filtering based on threshold rules set according to the vehicle physical model and technical parameters to obtain an intermediate dataset; S22. For the intermediate dataset, dynamic rules are set based on the statistical distribution characteristics of historical data to perform flexible cleaning, thereby obtaining the dataset to be repaired; S23. Identify the locations of missing values ​​in the dataset to be repaired, and for each missing data point to be repaired, construct an input feature vector for prediction to generate a missing data repair feature set; S24. Input the missing data repair feature set into the pre-trained XGBoost regression model, predict the imputation values ​​of the missing data, and fill them back into the dataset to be repaired, generating cleaned structured data.

[0010] Furthermore, the specific steps of step S21 are as follows: S211. Perform vehicle speed anomaly detection: If the vehicle speed value v satisfies If so, it is marked as an exception; in, The maximum permissible speed is set according to the vehicle model; S212. Perform battery voltage anomaly detection: If the battery voltage U satisfies or If so, it is marked as an exception; in, , The minimum and maximum voltage values ​​are determined by the calibration range of the battery management system, respectively. S213. Perform motor torque anomaly detection: Based on the motor's external characteristic curve, determine whether the torque value at the current speed is within the feasible range; if it exceeds the range, mark it as abnormal. S214. Remove or empty the data marked as anomalous to obtain the intermediate dataset after hard cleaning; The specific steps for S22 are as follows: S221. An extreme value filtering algorithm based on historical percentiles is used to smooth or remove extreme values ​​in vehicle speed and acceleration, which are continuous variables. S222. A sliding window anomaly detection algorithm is adopted to calculate the degree of deviation of data points from statistical features within a set time window, and to detect the isolation of data points based on the isolated forest algorithm. Data that simultaneously meet the criteria of deviation exceeding the threshold and are identified as isolated points are marked as anomalies and then removed or smoothed to complete the flexible cleaning process. S223. Output the dataset to be repaired after flexible cleaning; The input feature vector in step S23 includes temporal context features and external environment features; The temporal context features are the vehicle network data of the previous moment of the current data point, including at least SOC value, vehicle speed, and acceleration; The external environmental features are the environmental monitoring data corresponding to the current data point at that time, including at least weather conditions and road slope. The objective function of the pre-trained XGBoost regression model in step S24 is:

[0011] in, The squared loss function, The number of leaf nodes. and For regularization parameters, These are the model parameters.

[0012] Furthermore, the specific steps of step S3 are as follows: S31. Obtain the cleaned historical structured dataset from the historical database, extract the speed sequence and acceleration sequence from the vehicle network data of all historical trip segments, and construct a two-dimensional joint distribution of speed and acceleration; S32. The K-Means clustering algorithm is used to cluster the two-dimensional joint distribution of velocity and acceleration to obtain K typical working condition clusters. And for each operating condition cluster Calculate the average energy consumption per 100 kilometers for all included travel segments, as the operating condition cluster. Baseline energy consumption And establish a working condition-baseline energy consumption mapping table; S33. For the current trip to be evaluated, extract the actual energy consumption per 100 kilometers from its cleaned structured data. Average temperature in environmental monitoring data With average road slope And the air conditioning status indicator in the vehicle network data ; Based on the average speed and acceleration of the current journey, the operating condition cluster to which the current journey belongs is matched according to the operating condition-baseline energy consumption mapping table, and the corresponding baseline energy consumption is obtained. ; Based on the actual energy consumption per 100 kilometers of the current journey and the baseline energy consumption of the current travel cycle's operating condition cluster. The energy consumption fluctuation rate per unit mileage is calculated using the following formula. :

[0013] in, This is a correction factor function for temperature, slope, and air conditioning power consumption, which was determined in advance through calibration experiments. S34. Simultaneously extract the load rate time series from the cleaned structured data of the current trip's business record data. Speed ​​time series in vehicle network data ,in N is the total number of time points; The mean load factor-velocity coupling coefficient was calculated using the Pearson correlation coefficient formula. :

[0014] in, and These are the mean values ​​of the load rate sequence and the velocity sequence, respectively. S35. Extract the electrical energy recovered during braking events from the cleaned structured data of the current trip's journey, based on the vehicle-to-everything (V2X) data. And the total mechanical energy consumed by braking, estimated by the vehicle dynamics model during the same period. ; Based on the electrical energy recovered during the braking event and the total mechanical energy consumed by braking during the same period. Calculate typical regenerative braking efficiency : ; S36. Calculate the energy consumption fluctuation rate per unit mileage for the current trip. Mean load factor-velocity coupling coefficient Typical braking energy recovery efficiency The three high-order scene feature parameters are combined to form the enhanced feature set for the current journey.

[0015] Furthermore, the specific steps of step S4 are as follows: S41. Obtain a preset number of enhanced feature sets of historical trips and corresponding expert or rule-labeled driving behavior category tags from the historical database to generate a training sample set; the behavior category tags include at least smooth driving, aggressive driving, economical driving and heavy-load hill climbing mode; S42. Using the aforementioned training sample set, train a multi-class classification model based on the LightGBM algorithm, specifically including: The parameters in the enhanced feature set are standardized. Set the hyperparameters of the LightGBM model, including the learning rate, the number of leaf nodes, the maximum depth of the tree, and the feature sampling ratio; Using standardized enhanced feature set parameters as input features and driving behavior category labels as output targets, the LightGBM model is trained using a multi-class log loss function, and the optimal hyperparameter combination is selected through cross-validation. After training is complete, save and deploy the final model parameters to complete the multi-classification model deployment; S43. Input the enhanced feature set of the current trip into the deployed multi-classification model, output the probability of each category of driving behavior, and determine the driving behavior category with the highest probability as the dominant driving behavior mode of the current trip. If the probability of a non-dominant driving behavior category exceeds the preset auxiliary threshold, then mark this non-dominant driving behavior category as a secondary driving behavior mode. S44. The final output includes driving behavior classification results that include the dominant driving behavior pattern and the secondary driving behavior pattern.

[0016] Furthermore, the specific steps of step S5 are as follows: S51. Based on the driving behavior classification results, extract the cumulative duration or event frequency of aggressive driving mode as the aggressive driving percentage; Combined with typical regenerative braking efficiency of enhanced feature set The security score is calculated using the following formula. : ; in, and These are preset weighting coefficients; S52. Based on the driving behavior classification results, identify the proportion of economical driving modes; Combined with the enhanced feature set of energy consumption volatility per unit mile Coupling coefficient with mean load factor-velocity The economic score is calculated using the following formula. :

[0017] in, This is a preset positive coefficient used to balance the contributions of various indicators; S53. Based on security scoring Economic score Calculate the weighted comprehensive score :

[0018] in, Safety weights set according to fleet management strategy; S54. Generate an assessment report that includes an overview of driving behavior classification, safety and economy sub-scores and a comprehensive score, key feature parameter analysis, and specific improvement suggestions based on score shortcomings.

[0019] Furthermore, it also includes the following steps: S61. Automatically store the original trip data, its enhanced feature set, classification results and final score corresponding to the evaluation report with a comprehensive weighted score lower than the preset threshold into the special case sample library; S62. Periodically use newly added samples from the special case sample library as incremental training data to fine-tune and optimize the multi-classification model trained in step S42, so that the multi-classification model can adapt to new and difficult-to-classify driving modes or changes in vehicle status. S63. Based on the rating distribution of a set number of trips, statistically analyze the weighting coefficients in steps S51 and S52. The rationality of the fleet operation is assessed and calibrated based on changes in fleet operation objectives.

[0020] Secondly, embodiments of this application also provide a driving behavior classification and evaluation method system based on multi-source operational data of new energy commercial vehicles, including: The data acquisition and fusion module is used to collect vehicle network time-series data, environmental monitoring data and business record data generated by new energy commercial vehicles during operation. It performs time synchronization and spatial alignment on the collected data, and performs fusion processing based on Kalman filter to generate a fusion dataset with a unified spatiotemporal reference. The data cleaning and repair module is used to perform rule cleaning and missing data repair on the fused dataset in sequence to generate cleaned structured data. The feature enhancement module is used to extract and calculate high-order scene feature parameters based on a structured dataset to generate an enhanced feature set; the high-order scene feature parameters include at least the energy consumption fluctuation rate per unit mile, the average load rate-speed coupling coefficient, and the typical braking energy recovery efficiency. The behavior classification module is used to input the enhanced feature set into the trained classification model, identify and output driving behavior categories; the behavior categories include at least smooth driving, aggressive driving, economical driving and heavy-load hill climbing mode; The comprehensive assessment and report generation module is used to calculate a comprehensive driving behavior score based on the driving behavior category and the corresponding enhanced feature set parameters, combined with preset safety and economy assessment rules, and generate an assessment report.

[0021] As can be seen from the above technical solutions, this application has the following advantages: The driving behavior classification and evaluation method and system based on multi-source operational data of new energy commercial vehicles provided in this application improves the quality of multi-source operational data of new energy commercial vehicles through layered cleaning and repair, combined with spatiotemporal alignment and fusion processing, providing a data foundation for analysis and application. Based on high-order scene feature parameters and classification models, it can accurately identify various driving behavior modes, such as smooth driving, aggressive driving, and economical driving, providing a reasonable basis for fleet management and driver training. Combined with safety and economic evaluation rules, it generates driving behavior evaluation reports and provides targeted improvement suggestions, which can optimize fleet operation efficiency, reduce energy consumption, and improve safety. Through special case sample library and model fine-tuning, it can adapt to new driving modes and vehicle status changes, with good scalability and adaptability, meeting the operational needs of different fleets, realizing intelligent operation management of new energy commercial vehicles, and achieving refined scheduling, energy consumption optimization, and safety early warning, thereby improving overall operational efficiency. Attached Figure Description

[0022] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating the driving behavior classification and evaluation method based on multi-source operational data of new energy commercial vehicles according to the present invention.

[0024] Figure 2 This is a schematic diagram of the driving behavior classification and evaluation system based on multi-source operation data of new energy commercial vehicles according to the present invention. Detailed Implementation

[0025] The various embodiments of this disclosure will be described more fully in the following detailed description of the specific steps of the driving behavior classification and evaluation method based on multi-source operational data of new energy commercial vehicles. This disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of this disclosure to the specific embodiments disclosed herein, but rather this disclosure should be understood to cover all adjustments, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of this disclosure.

[0026] This embodiment provides a method for classifying and evaluating driving behavior based on multi-source operational data of new energy commercial vehicles, which improves the data quality and processing efficiency of new energy commercial vehicles, accurately identifies driving behavior patterns, optimizes energy consumption and safety, and enhances intelligent operation management.

[0027] 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.

[0028] Please see Figure 1 The diagram shows a flowchart of a driving behavior classification and evaluation method based on multi-source operational data of new energy commercial vehicles in a specific embodiment. The method includes the following steps: S1. Collect vehicle network time-series data, environmental monitoring data and business record data generated during the operation of new energy commercial vehicles, synchronize and spatially align the collected data, and perform fusion processing based on Kalman filter to generate a fusion dataset with a unified spatiotemporal reference. It should be noted that collecting vehicle network time-series data, environmental monitoring data, and business record data ensures the comprehensiveness of data sources and provides basic data for analysis. By using time synchronization and spatial alignment, combined with Kalman filter fusion processing, the problem of spatiotemporal inconsistency of multi-source data is solved, generating a fusion dataset with a unified spatiotemporal benchmark and improving data quality. S2. Perform rule cleaning and missing data repair on the fused dataset in sequence to generate cleaned structured data; It should be noted that by combining rule cleaning and model repair, outliers and missing values ​​in the data are comprehensively processed, improving the integrity and accuracy of the data and reducing the impact of noisy data on data analysis. S3. Based on the structured dataset, extract and calculate high-order scene feature parameters to generate an enhanced feature set; the high-order scene feature parameters include at least the energy consumption fluctuation rate per unit mileage, the average load rate-speed coupling coefficient, and the typical braking energy recovery efficiency. It should be noted that by extracting high-order scene feature parameters such as energy consumption fluctuation rate per unit mileage, average load rate-speed coupling coefficient, and typical braking energy recovery efficiency, an enhanced feature set is constructed, which provides feature support for driving behavior classification and improves classification accuracy. S4. Input the enhanced feature set into the trained classification model to identify and output driving behavior categories; the behavior categories include at least smooth driving, aggressive driving, economical driving, and heavy-load hill climbing mode; It should be noted that by inputting the enhanced feature set into the trained classification model, it is possible to identify multiple driving behavior patterns, providing accurate classification results for assessment and management, and enabling refined operation management. S5. Based on the driving behavior category and the corresponding enhanced feature set parameters, combined with the preset safety and economy assessment rules, calculate the comprehensive driving behavior score and generate an assessment report; It should be noted that by combining driving behavior categories and enhanced feature set parameters, safety and economy scores are calculated, and an evaluation report containing classification results, sub-scores and improvement suggestions is generated, providing a reasonable basis for fleet management and driving optimization, and improving operational management efficiency and safety.

[0029] This embodiment addresses the heterogeneity, noise, and missing data issues of multi-source data for new energy commercial vehicles through layered cleaning, spatiotemporal alignment, and feature enhancement, thereby improving data quality. By combining high-level scene features and classification models, it accurately identifies driving behavior, optimizes energy consumption and safety, and provides a basis for intelligent scheduling and operation management.

[0030] Furthermore, as a refinement and extension of the specific implementation of the above embodiments, in order to fully illustrate the specific implementation process in this embodiment, another method for classifying and evaluating driving behavior based on multi-source operational data of new energy commercial vehicles is provided. Taking 10 4.2-meter pure electric vans of a new energy logistics fleet in a certain city as the application object, the entire implementation process of the method for classifying and evaluating driving behavior based on multi-source operational data of new energy commercial vehicles is described in detail. This fleet mainly undertakes the urban supermarket delivery business, with an average daily mileage of 120-180km and a load fluctuation range of 0.3-1.2 tons. The operating scenarios cover typical working conditions such as congested urban roads, suburban expressways, and park slopes. The method includes the following steps: S1. Collect vehicle network time-series data, environmental monitoring data and business record data generated during the operation of new energy commercial vehicles, synchronize and spatially align the collected data, and perform fusion processing based on Kalman filter to generate a fusion dataset with a unified spatiotemporal reference. In step S1, the vehicle-to-everything (V2X) timing data includes vehicle GPS positioning data, vehicle speed, battery SOC, and motor torque; environmental monitoring data includes weather, road conditions, and traffic congestion index; and business record data includes load and trip start and end times. The specific steps of step S1 are as follows: S11. Perform time synchronization, using UTC time as the reference, synchronize the timestamps of each data source at the millisecond level, and establish a unified time series; S12. Perform spatial alignment. For the collected GPS positioning data, use linear interpolation to perform spatial alignment to match the sampling frequency of the remaining data. The interpolation formula is:

[0031] Where t represents the specific time when location interpolation needs to be performed (its timestamp lies between two valid GPS points); t1 and t2 represent the timestamps of two adjacent valid GPS positioning points, and satisfy the following conditions: ; This represents the spatial position coordinates calculated by interpolation at time t; , These represent the valid GPS positioning coordinates collected at times t1 and t2, respectively. S13. Using a Kalman filter, perform fusion and noise reduction processing on the various data sources after time synchronization and spatial alignment, and output a fused dataset with unified spatiotemporal reference; For example, this implementation collected operational data from 10 vehicles over 30 consecutive days, covering three types of data sources: This implementation collected operational data from 10 vehicles over 30 consecutive days, covering three types of data sources: vehicle-to-everything (V2X) time-series data: GPS positioning data (sampling frequency 1Hz), vehicle speed (10Hz), battery SOC (5Hz), motor torque (10Hz), and battery voltage (5Hz) were collected via the in-vehicle T-BOX, accumulating approximately 2.16 × 10⁻⁶ times the data. 7 Items; Environmental monitoring data: Weather data (sunny / cloudy / rainy / snowy) is obtained from the city's meteorological platform, and road conditions (smooth / potholes / construction) and traffic congestion index (0-10, 0 for smooth traffic and 10 for severe congestion) are obtained from the road monitoring system, accumulating approximately 2.59 × 10-10. 4 Items; Business record data: Load data, start and end times, and delivery station information for each trip are obtained through the fleet management system, accumulating to approximately 1.8 × 10 items. 3 strip; Using UTC time as the benchmark, all data sources are calibrated with millisecond-level timestamps. For example, vehicle speed data at 08:30:25.120 on 2025-06-10 is synchronously matched with battery SOC, environmental congestion index, and load data at the same millisecond-level timestamp to eliminate time discrepancies between multi-source data. Spatial alignment: To address the issue of GPS positioning data sampling frequency being lower than other data, linear interpolation is used to supplement position coordinates. For example, given GPS coordinates at t1=08:30:25 (116.3521, 39.9215) and t2=08:30:26 (116.3523, 39.9217), the interpolated coordinates at t=08:30:25.6 are: =(116.35222,39.92162) After spatial alignment, the GPS data sampling frequency is unified to 10Hz, which is consistent with the frequency of vehicle speed and motor torque data. Kalman filter fusion performs noise reduction processing on the synchronized and aligned multi-source data. Taking vehicle speed data as an example, the original vehicle speed data fluctuates by ±2 km / h due to sensor noise. After Kalman filtering, the noise deviation is reduced to within ±0.5 km / h. Simultaneously, environmental congestion index, road gradient, and vehicle speed data are fused to output a fused dataset with a unified spatiotemporal reference, accumulating to approximately 2.3 × 10⁻⁶ fused data points. 7 strip; S2. Perform rule cleaning and missing data repair on the fused dataset sequentially to generate cleaned structured data; the specific steps of step S2 are as follows: S21. For the vehicle-to-everything (V2X) time-series data in the fused dataset, perform hard cleaning of anomalies and filtering based on threshold rules set according to the vehicle physical model and technical parameters to obtain an intermediate dataset; the specific steps of step S21 are as follows: S211. Perform vehicle speed anomaly detection: If the vehicle speed value v satisfies If so, it is marked as an exception; in, This is the maximum permissible speed set according to the vehicle model, with a range of 80-120 km / h; S212. Perform battery voltage anomaly detection: If the battery voltage U satisfies or If so, it is marked as an exception; in, , The minimum and maximum voltage values ​​are determined by the calibration range of the battery management system, respectively. S213. Perform motor torque anomaly detection: Based on the motor's external characteristic curve, determine whether the torque value at the current speed is within the feasible range; if it exceeds the range, mark it as abnormal. S214. Remove or empty the data marked as anomalous to obtain the intermediate dataset after hard cleaning; S22. For the intermediate dataset, dynamic rules are set based on the statistical distribution characteristics of historical data to perform flexible cleaning and generate a dataset to be repaired; For example, threshold rules are set based on the vehicle physics model to filter anomalies in the fused dataset: Speed ​​anomaly detection: Maximum permissible speed for this model of truck =90km / h, a vehicle speed of 110km / h was detected during a certain period, which was judged as abnormal and removed. A total of approximately 1.2×10 abnormal vehicle speed data were filtered. 4 strip; Battery voltage anomaly detection: The battery management system of this vehicle is calibrated to a voltage range of 280V-350V. At a certain time, battery voltages of 265V (below the minimum) and 362V (above the maximum) were detected, marked as anomalies and set to empty. Approximately 8.5 × 10⁻⁶ voltage anomaly data points were processed. 3 strip; Motor torque anomaly detection: According to the motor external characteristic curve, when the motor speed is 1500 r / min, the maximum feasible torque is 280 N. m, the torque at a certain data point is detected to be 320N. m was identified as abnormal and removed, resulting in the filtering of approximately 6.3 × 10⁻⁶ abnormal torque data. 3 strip; After hard cleaning, the intermediate dataset is approximately 2.297 × 10⁻⁶. 7 strip; The specific steps for S22 are as follows: S221. An extreme value filtering algorithm based on historical percentiles is used to smooth or remove extreme values ​​in vehicle speed and acceleration, which are continuous variables. S222. A sliding window anomaly detection algorithm is adopted to calculate the degree of deviation of data points from statistical features (such as mean and standard deviation) within a set time window, and to detect the isolation of data points based on the isolated forest algorithm. Data that simultaneously meet the criteria of deviation exceeding the threshold and are identified as isolated points are marked as anomalies and then removed or smoothed to complete the flexible cleaning process. S223. Output the dataset to be repaired after flexible cleaning; For example, historical percentile extreme value filtering: for vehicle speed and acceleration data, the historical 95th percentile is used as the threshold (vehicle speed threshold is 75km / h, acceleration threshold is 2.5m / s²). 2 ), for the vehicle speed data of 82km / h and 3.2m / s during a certain trip, 2 The acceleration data was smoothed, and approximately 3.1 × 10⁻⁶ extreme values ​​were processed. 3 strip; Sliding window + Isolated Forest anomaly detection: A time window of 10 seconds is set. The mean and standard deviation of vehicle speeds within the window are calculated. Data points deviating from the mean by more than three times the standard deviation and identified as isolated points by the Isolated Forest algorithm are removed. For example, if a vehicle's speed suddenly spikes to 60 km / h in a congested area (window mean 12 km / h), and this is an isolated point, it is identified as an anomaly and removed. Approximately 1.8 × 10⁻⁶ such data points are processed. 3 strip; After gentle cleaning, the dataset to be repaired is approximately 2.295 × 10⁻⁶. 7 strip; S23. Identify the locations of missing values ​​in the dataset to be repaired, and for each missing data point to be repaired, construct an input feature vector for prediction to generate a missing data repair feature set; The input feature vector in step S23 includes temporal context features and external environment features; The temporal context features are the vehicle network data of the previous moment of the current data point, including at least SOC value, vehicle speed, and acceleration; The external environmental features are the environmental monitoring data corresponding to the current data point at that time, including at least weather conditions and road slope. S24. Input the missing data repair feature set into the pre-trained XGBoost regression model, predict the imputation values ​​of the missing data, and fill them back into the dataset to be repaired, generating cleaned structured data. The objective function of the pre-trained XGBoost regression model in step S24 is:

[0032] in, The squared loss function, The number of leaf nodes. and For regularization parameters, These are model parameters; For example, construct a repair feature set: for missing values ​​in the dataset to be repaired (such as missing battery SOC data for a certain period), extract the SOC (58%), vehicle speed (32km / h), and acceleration (1.2m / s²) from the previous moment. 2 The repair input vector is constructed using temporal context features such as the current weather (light rain) and road slope (2°). XGBoost model infilling: The repair feature set is input into a pre-trained XGBoost regression model (objective function is...). (where γ=0.1 and λ=0.5), the predicted missing SOC value was 56.8%, and imputed; a total of approximately 4.2 × 10⁻⁶ missing data were repaired in this implementation. 3 The final cleaned structured data is approximately 2.295 × 10⁻⁶. 7 strip; S3. Based on the structured dataset, extract and calculate high-order scene feature parameters to generate an enhanced feature set; the high-order scene feature parameters include at least the energy consumption fluctuation rate per unit mile, the average load rate-speed coupling coefficient, and the typical braking energy recovery efficiency; the specific steps of step S3 are as follows: S31. Obtain the cleaned historical structured dataset from the historical database, extract the speed sequence and acceleration sequence from the vehicle network data of all historical trip segments, and construct a two-dimensional joint distribution of speed and acceleration; S32. The K-Means clustering algorithm is used to cluster the two-dimensional joint distribution of velocity and acceleration to obtain K typical working condition clusters. And for each operating condition cluster Calculate the average energy consumption per 100 kilometers for all included travel segments, as the operating condition cluster. Baseline energy consumption And establish a working condition-baseline energy consumption mapping table; For example, six months of structured data for the fleet were extracted from the historical database, and a two-dimensional joint distribution of velocity and acceleration was constructed. The K-Means clustering algorithm (K=5) was used to obtain five typical operating condition clusters, as shown in Table 1. Table 1

[0033] S33. For the current trip to be evaluated, extract the actual energy consumption per 100 kilometers from its cleaned structured data. Average temperature in environmental monitoring data With average road slope And the air conditioning status indicator in the vehicle network data ; Based on the average speed and acceleration of the current journey, the operating condition cluster to which the current journey belongs is matched according to the operating condition-baseline energy consumption mapping table, and the corresponding baseline energy consumption is obtained. ; Based on the actual energy consumption per 100 kilometers of the current journey and the baseline energy consumption of the current travel cycle's operating condition cluster. The energy consumption fluctuation rate per unit mileage is calculated using the following formula. :

[0034] in, This is a correction factor function for temperature, slope, and air conditioning power consumption, which was determined in advance through calibration experiments. S34. Simultaneously extract the load rate time series from the cleaned structured data of the current trip's business record data. Speed ​​time series in vehicle network data ,in N is the total number of time points; The mean load factor-velocity coupling coefficient was calculated using the Pearson correlation coefficient formula. :

[0035] in, and These are the mean values ​​of the load rate sequence and the velocity sequence, respectively. S35. Extract the electrical energy recovered during braking events from the cleaned structured data of the current trip's journey, based on the vehicle-to-everything (V2X) data. And the total mechanical energy consumed by braking, estimated by the vehicle dynamics model during the same period. ; Based on the electrical energy recovered during the braking event and the total mechanical energy consumed by braking during the same period. Calculate typical regenerative braking efficiency : ; S36. Calculate the energy consumption fluctuation rate per unit mileage for the current trip. Mean load factor-velocity coupling coefficient Typical braking energy recovery efficiency The three high-order scene feature parameters are combined to form the enhanced feature set for the current journey; For example, consider a delivery trip by a vehicle from 09:00 to 10:30 on June 15, 2025 (trip distance 15km, load 0.8 tons, light rain, average slope 1.5°, air conditioning on): Energy consumption fluctuation per unit distance: The average speed for this trip is 32 km / h, and the operating condition is C2, with a baseline energy consumption of [missing information]. =32.5kWh / 100km; Actual energy consumption per 100km =36.8kWh / 100km; temperature, gradient, air conditioning correction factor If the value is 1.12, then the energy consumption volatility is... ; Average load rate-velocity coupling coefficient: Extract the load rate sequence for this stroke. (mean 0.67) and velocity sequence (Mean speed 32 km / h), calculated using Pearson correlation coefficient. =-0.72, which shows an ideal coupling relationship of heavy load low speed and light load high speed; Typical regenerative braking efficiency: During this stroke, a total of 12 braking events occurred, resulting in the cumulative recovery of electrical energy. =1.2kWh, total mechanical energy consumed during braking =5.8kWh, then the recovery efficiency ≈0.207; The three features mentioned above are combined into an enhanced feature set for this trip, generating a total of 1.8 × 10⁻⁶ features for the entire 30-day trip of the convoy. 3 Group-enhanced feature set; S4. Input the enhanced feature set into the trained classification model to identify and output driving behavior categories; the behavior categories include at least smooth driving, aggressive driving, economical driving, and heavy-load hill climbing mode; the specific steps of step S4 are as follows: S41. Obtain a preset number of enhanced feature sets of historical trips and corresponding expert or rule-labeled driving behavior category tags from the historical database to generate a training sample set; the behavior category tags include at least smooth driving, aggressive driving, economical driving and heavy-load hill climbing mode; S42. Using the aforementioned training sample set, train a multi-class classification model based on the LightGBM algorithm, specifically including: The parameters in the enhanced feature set are standardized. Set the hyperparameters of the LightGBM model, including the learning rate, the number of leaf nodes, the maximum depth of the tree, and the feature sampling ratio; Using standardized enhanced feature set parameters as input features and driving behavior category labels as output targets, the LightGBM model is trained using a multi-class log loss function, and the optimal hyperparameter combination is selected through cross-validation. After training is complete, save and deploy the final model parameters to complete the multi-classification model deployment; S43. Input the enhanced feature set of the current trip into the deployed multi-classification model, output the probability of each category of driving behavior, and determine the driving behavior category with the highest probability as the dominant driving behavior mode of the current trip. If the probability of a non-dominant driving behavior category exceeds the preset auxiliary threshold, then mark this non-dominant driving behavior category as a secondary driving behavior mode. S44. The final output includes driving behavior classification results containing the dominant driving behavior pattern and the secondary driving behavior pattern; For example, the training sample set is constructed as follows: Select 1.5×10 from the historical enhanced feature set 3 The sample groups were labeled with driving behavior categories by fleet safety management experts. The sample distribution was as follows: smooth driving (600 groups), aggressive driving (225 groups), economical driving (450 groups), and heavy-load climbing mode (225 groups). LightGBM model training: The enhanced feature set is standardized (mean 0, standard deviation 1); Model hyperparameters are set as follows: learning rate 0.05, number of leaf nodes 32, maximum tree depth 6, feature sampling ratio 0.8; The model was trained using a multi-class log loss function, and the hyperparameters were optimized through 5-fold cross-validation. The final model achieved a classification accuracy of 92.3% and an F1 score of 0.918, meeting the needs of practical applications. Behavior classification results: Inputting a feature set of a specific trip into the model outputs the probabilities of each category: smooth driving (0.72), economical driving (0.25), aggressive driving (0.02), and heavy-load hill climbing (0.01). Setting an auxiliary threshold of 0.2 determines the dominant driving behavior for the trip as smooth driving and the secondary driving behavior as economical driving. The 30-day trip classification results for the entire fleet show: smooth driving accounted for 68%, economical driving for 21%, aggressive driving for 7%, and heavy-load hill climbing for 4%. S5. Based on the driving behavior category and the corresponding enhanced feature set parameters, combined with preset safety and economy assessment rules, calculate the comprehensive driving behavior score and generate an assessment report; the specific steps of step S5 are as follows: S51. Based on the driving behavior classification results, extract the cumulative duration or event frequency of aggressive driving mode as the aggressive driving percentage; Combined with typical regenerative braking efficiency of enhanced feature set (Insufficient efficiency may be accompanied by emergency braking). Calculate the safety score using the following formula. : ; in, and The preset weighting coefficients, and The scoring results are normalized to 0-100 points; S52. Based on the driving behavior classification results, identify the proportion of economical driving modes; Combined with the enhanced feature set of energy consumption volatility per unit mile Coupling coefficient with mean load factor-velocity (Ideally, there should be a negative correlation, i.e., heavy load at low speed and light load at high speed.) The economic score is calculated using the following formula. :

[0036] in, This is a preset positive coefficient used to balance the contributions of various indicators; S53. Based on security scoring Economic score Calculate the weighted comprehensive score :

[0037] in, Safety weights (0 < λ < 1) are set according to the fleet management strategy. S54. Generate an assessment report that includes an overview of driving behavior classification (dominant and secondary modes), safety and economy sub-scores and a comprehensive score, key characteristic parameter analysis (such as high energy consumption fluctuation range, inefficient braking event points), and specific improvement suggestions based on the score shortcomings (e.g., "The proportion of aggressive driving is too high, it is recommended to strengthen the prediction of smooth start" or "The braking energy recovery efficiency is lower than the average level of the fleet, it is recommended to check the braking system and optimize the braking strategy"). For example, the safety and economy score: Safety score: extracts the proportion of aggressive driving during the trip (0.02), braking energy recovery efficiency (0.207), and sets weights. =0.3、 =0.2 ( The security score is then calculated. =100-(0.3×20+0.2×79.3)=80.14; Economic efficiency score: The trip is identified as having an economical driving ratio of 0.25%, an energy consumption fluctuation rate of 0.021, a load-speed coupling coefficient of -0.72, and set coefficients α=0.4, β=0.3, and γ=0.3. Therefore, the economic efficiency score is calculated. =0.4×25+0.3×97.9+0.3×72=71.97; Overall Score: With a safety weight λ=0.6, the overall score is... =0.6×80.14+0.4×71.97=77.07; Assessment report generation: The assessment report generated for this trip includes the following: Driving behavior classification overview: dominant smooth driving, secondary economical driving, no risk of aggressive or heavy-load uphill driving; Rating details: Safety 80.14 points, Economy 71.97 points, Overall 77.07 points; Key feature analysis: Energy consumption fluctuation rate 0.021 (better than the fleet average of 0.035), Braking recovery efficiency 0.207 (slightly lower than the fleet average of 0.23). Improvement suggestion: Optimize braking timing and improve energy recovery efficiency, which can further reduce energy consumption by 1-2 kWh per 100 kilometers.

[0038] Unlike the examples described above, another embodiment of the present invention further includes the following steps: S61. Automatically store the original trip data, its enhanced feature set, classification results and final score corresponding to the evaluation report with a comprehensive weighted score lower than the preset threshold into the special case sample library; S62. Periodically (e.g., monthly), new samples from the special case sample library are used as incremental training data to fine-tune and optimize the multi-classification model trained in step S42, so that the multi-classification model can adapt to new and difficult-to-classify driving modes or changes in vehicle status. S63. Based on the rating distribution of a set number of trips, statistically analyze the weighting coefficients in steps S51 and S52. The rationality of the fleet operation objectives should be verified and calibrated accordingly. For example, a special case sample library is constructed: The original data, enhanced feature sets, and scoring results of 42 evaluation reports with a comprehensive score below 70 were stored in a special case sample library, which included cases with an excessively high proportion of aggressive driving (18 cases), extremely low braking regeneration efficiency (15 cases), and excessive energy consumption fluctuation rate (9 cases). Model fine-tuning: Each month, newly added cases in the sample library are used as incremental data to fine-tune the LightGBM model; after two incremental training sessions, the model's classification accuracy for low-scoring cases has increased to 95.7%, effectively adapting to the new driving mode. Weighting coefficient calibration: The analysis of the 30-day trip rating distribution revealed that the overall safety rating was too high and the economy rating was too low. In line with the fleet's "cost reduction first" operational goal, the safety weight λ was adjusted from 0.6 to 0.5. After recalculation, the overall rating is more in line with the fleet management needs.

[0039] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0040] like Figure 2 As shown, the following are embodiments of the driving behavior classification and evaluation method system based on multi-source operation data of new energy commercial vehicles provided in this disclosure. This system and the driving behavior classification and evaluation method based on multi-source operation data of new energy commercial vehicles in the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the driving behavior classification and evaluation method system based on multi-source operation data of new energy commercial vehicles, please refer to the embodiments of the driving behavior classification and evaluation method based on multi-source operation data of new energy commercial vehicles.

[0041] The system includes: The data acquisition and fusion module is used to collect vehicle network time-series data, environmental monitoring data and business record data generated by new energy commercial vehicles during operation. It performs time synchronization and spatial alignment on the collected data, and performs fusion processing based on Kalman filter to generate a fusion dataset with a unified spatiotemporal reference. The data cleaning and repair module is used to perform rule cleaning and missing data repair on the fused dataset in sequence to generate cleaned structured data. The feature enhancement module is used to extract and calculate high-order scene feature parameters based on a structured dataset to generate an enhanced feature set; the high-order scene feature parameters include at least the energy consumption fluctuation rate per unit mile, the average load rate-speed coupling coefficient, and the typical braking energy recovery efficiency. The behavior classification module is used to input the enhanced feature set into the trained classification model, identify and output driving behavior categories; the behavior categories include at least smooth driving, aggressive driving, economical driving and heavy-load hill climbing mode; The comprehensive assessment and report generation module is used to calculate a comprehensive driving behavior score based on the driving behavior category and the corresponding enhanced feature set parameters, combined with preset safety and economy assessment rules, and generate an assessment report.

[0042] This embodiment achieves improved data quality and processing efficiency for new energy commercial vehicles, accurate identification of driving behavior patterns, optimization of energy consumption and safety, and enhanced intelligent operation management through the interactive collaboration of data acquisition and fusion module, data cleaning and repair module, feature enhancement module, behavior classification module, and comprehensive evaluation and report generation module.

[0043] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for classifying and evaluating driving behavior based on multi-source operational data of new energy commercial vehicles, characterized in that, Includes the following steps: S1. Collect vehicle network time-series data, environmental monitoring data and business record data generated during the operation of new energy commercial vehicles, synchronize and spatially align the collected data, and perform fusion processing based on Kalman filter to generate a fusion dataset with a unified spatiotemporal reference. S2. Perform rule cleaning and missing data repair on the fused dataset in sequence to generate cleaned structured data; S3. Based on the structured dataset, extract and calculate high-order scene feature parameters to generate an enhanced feature set; The high-order scenario characteristic parameters include at least the energy consumption fluctuation rate per unit mileage, the average load rate-speed coupling coefficient, and the typical braking energy recovery efficiency. S4. Input the enhanced feature set into the trained classification model to identify and output driving behavior categories; the behavior categories include at least smooth driving, aggressive driving, economical driving, and heavy-load hill climbing mode; S5. Based on the driving behavior category and the corresponding enhanced feature set parameters, combined with the preset safety and economy assessment rules, calculate the comprehensive driving behavior score and generate an assessment report.

2. The method for classifying and evaluating driving behavior based on multi-source operational data of new energy commercial vehicles according to claim 1, characterized in that, In step S1, the vehicle-to-everything (V2X) timing data includes vehicle GPS positioning data, vehicle speed, battery SOC, and motor torque; environmental monitoring data includes weather, road conditions, and traffic congestion index; and business record data includes load and trip start and end times.

3. The driving behavior classification and evaluation method based on multi-source operational data of new energy commercial vehicles according to claim 1, characterized in that, The specific steps of step S1 are as follows: S11. Perform time synchronization, using UTC time as the reference, synchronize the timestamps of each data source at the millisecond level, and establish a unified time series; S12. Perform spatial alignment. For the collected GPS positioning data, use linear interpolation to perform spatial alignment to match the sampling frequency of the remaining data. The interpolation formula is: Where t represents the specific time at which location interpolation needs to be performed; t1 and t2 represent the timestamps of two adjacent valid GPS positioning points, and satisfy the following conditions: ; This represents the spatial position coordinates calculated by interpolation at time t; , These represent the valid GPS positioning coordinates collected at times t1 and t2, respectively. S13. Using a Kalman filter, perform fusion and noise reduction processing on the various data sources after time synchronization and spatial alignment, and output a fused dataset with unified spatiotemporal reference.

4. The method for classifying and evaluating driving behavior based on multi-source operational data of new energy commercial vehicles according to claim 1, characterized in that, The specific steps of step S2 are as follows: S21. For the vehicle network time series data in the fused dataset, perform hard cleaning of anomaly detection and filtering based on threshold rules set according to the vehicle physical model and technical parameters to obtain an intermediate dataset; S22. For the intermediate dataset, dynamic rules are set based on the statistical distribution characteristics of historical data to perform flexible cleaning, thereby obtaining the dataset to be repaired; S23. Identify the locations of missing values ​​in the dataset to be repaired, and for each missing data point to be repaired, construct an input feature vector for prediction to generate a missing data repair feature set; S24. Input the missing data repair feature set into the pre-trained XGBoost regression model, predict the imputation values ​​of the missing data, and fill them back into the dataset to be repaired, generating cleaned structured data.

5. The driving behavior classification and evaluation method based on multi-source operational data of new energy commercial vehicles according to claim 4, characterized in that, The specific steps of step S21 are as follows: S211. Perform vehicle speed anomaly detection: If the vehicle speed value v satisfies If so, it is marked as an exception; in, The maximum permissible speed is set according to the vehicle model; S212. Perform battery voltage anomaly detection: If the battery voltage U satisfies or If so, it is marked as an exception; in, , The minimum and maximum voltage values ​​are determined by the calibration range of the battery management system, respectively. S213. Perform motor torque anomaly detection: Based on the motor's external characteristic curve, determine whether the torque value at the current speed is within the feasible range; if it exceeds the range, mark it as abnormal. S214. Remove or empty the data marked as anomalous to obtain the intermediate dataset after hard cleaning; The specific steps for S22 are as follows: S221. An extreme value filtering algorithm based on historical percentiles is used to smooth or remove extreme values ​​in vehicle speed and acceleration, which are continuous variables. S222. A sliding window anomaly detection algorithm is adopted to calculate the degree of deviation of data points from statistical features within a set time window, and to detect the isolation of data points based on the isolated forest algorithm. Data that simultaneously meet the criteria of deviation exceeding the threshold and are identified as isolated points are marked as anomalies and then removed or smoothed to complete the flexible cleaning process. S223. Output the dataset to be repaired after flexible cleaning; The input feature vector in step S23 includes temporal context features and external environment features; The temporal context features are the vehicle network data of the previous moment of the current data point, including at least SOC value, vehicle speed, and acceleration; The external environmental features are the environmental monitoring data corresponding to the current data point at that time, including at least weather conditions and road slope. The objective function of the pre-trained XGBoost regression model in step S24 is: in, The squared loss function, The number of leaf nodes. and For regularization parameters, These are the model parameters.

6. The method for classifying and evaluating driving behavior based on multi-source operational data of new energy commercial vehicles according to claim 1, characterized in that, The specific steps of step S3 are as follows: S31. Obtain the cleaned historical structured dataset from the historical database, extract the speed sequence and acceleration sequence from the vehicle network data of all historical trip segments, and construct a two-dimensional joint distribution of speed and acceleration; S32. The K-Means clustering algorithm is used to cluster the two-dimensional joint distribution of velocity and acceleration to obtain K typical working condition clusters. And for each operating condition cluster Calculate the average energy consumption per 100 kilometers for all included travel segments, as the operating condition cluster. Baseline energy consumption And establish a working condition-baseline energy consumption mapping table; S33. For the current trip to be evaluated, extract the actual energy consumption per 100 kilometers from its cleaned structured data. Average temperature in environmental monitoring data Compared with average road slope And the air conditioning status indicator in the vehicle network data ; Based on the average speed and acceleration of the current journey, the operating condition cluster to which the current journey belongs is matched according to the operating condition-baseline energy consumption mapping table, and the corresponding baseline energy consumption is obtained. ; Based on the actual energy consumption per 100 kilometers of the current journey and the baseline energy consumption of the current travel cycle's operating condition cluster. The energy consumption fluctuation rate per unit mileage is calculated using the following formula. : in, This is a correction factor function for temperature, slope, and air conditioning power consumption, which was determined in advance through calibration experiments. S34. Simultaneously extract the load rate time series from the cleaned structured data of the current trip's business record data. Speed ​​time series in vehicle network data ,in N is the total number of time points; The mean load factor-velocity coupling coefficient was calculated using the Pearson correlation coefficient formula. : in, and These are the mean values ​​of the load rate sequence and the velocity sequence, respectively. S35. Extract the electrical energy recovered during braking events from the cleaned structured data of the current trip's journey, based on the vehicle-to-everything (V2X) data. And the total mechanical energy consumed by braking, estimated by the vehicle dynamics model during the same period. ; Based on the electrical energy recovered during the braking event and the total mechanical energy consumed by braking during the same period. Calculate typical regenerative braking efficiency : ； S36. Calculate the energy consumption fluctuation rate per unit mileage for the current trip. Mean load factor-velocity coupling coefficient Typical braking energy recovery efficiency The three high-order scene feature parameters are combined to form the enhanced feature set for the current journey.

7. The method for classifying and evaluating driving behavior based on multi-source operational data of new energy commercial vehicles according to claim 1, characterized in that, The specific steps of step S4 are as follows: S41. Obtain a preset number of enhanced feature sets of historical trips and corresponding expert or rule-labeled driving behavior category tags from the historical database to generate a training sample set; the behavior category tags include at least smooth driving, aggressive driving, economical driving and heavy-load hill climbing mode; S42. Using the aforementioned training sample set, train a multi-class classification model based on the LightGBM algorithm, specifically including: The parameters in the enhanced feature set are standardized. Set the hyperparameters of the LightGBM model, including the learning rate, the number of leaf nodes, the maximum depth of the tree, and the feature sampling ratio; Using standardized enhanced feature set parameters as input features and driving behavior category labels as output targets, the LightGBM model is trained using a multi-class log loss function, and the optimal hyperparameter combination is selected through cross-validation. After training is complete, save and deploy the final model parameters to complete the multi-classification model deployment; S43. Input the enhanced feature set of the current trip into the deployed multi-classification model, output the probability of each category of driving behavior, and determine the driving behavior category with the highest probability as the dominant driving behavior mode of the current trip. If the probability of a non-dominant driving behavior category exceeds the preset auxiliary threshold, then mark this non-dominant driving behavior category as a secondary driving behavior mode. S44. The final output includes driving behavior classification results that include the dominant driving behavior pattern and the secondary driving behavior pattern.

8. The method for classifying and evaluating driving behavior based on multi-source operational data of new energy commercial vehicles according to claim 1, characterized in that, The specific steps of step S5 are as follows: S51. Based on the driving behavior classification results, extract the cumulative duration or event frequency of aggressive driving mode as the aggressive driving percentage; Combined with typical regenerative braking efficiency of enhanced feature set The security score is calculated using the following formula. : ； in, and These are preset weighting coefficients; S52. Based on the driving behavior classification results, identify the proportion of economical driving modes; Combined with the enhanced feature set of energy consumption volatility per unit mile Coupling coefficient with mean load factor-velocity The economic score is calculated using the following formula. : in, This is a preset positive coefficient used to balance the contributions of various indicators; S53. Based on security scoring Economic score Calculate the weighted comprehensive score : in, Safety weights set according to fleet management strategy; S54. Generate an assessment report that includes an overview of driving behavior classification, safety and economy sub-scores and a comprehensive score, key feature parameter analysis, and specific improvement suggestions based on score shortcomings.

9. The method for classifying and evaluating driving behavior based on multi-source operational data of new energy commercial vehicles according to claim 8, characterized in that, It also includes the following steps: S61. Automatically store the original trip data, its enhanced feature set, classification results and final score corresponding to the evaluation report with a comprehensive weighted score lower than the preset threshold into the special case sample library; S62. Periodically use newly added samples from the special case sample library as incremental training data to fine-tune and optimize the multi-classification model trained in step S42, so that the multi-classification model can adapt to new and difficult-to-classify driving modes or changes in vehicle status. S63. Based on the rating distribution of a set number of trips, statistically analyze the weighting coefficients in steps S51 and S52. The rationality of the fleet operation is assessed and calibrated based on changes in fleet operation objectives.

10. A system for classifying and evaluating driving behavior based on multi-source operational data of new energy commercial vehicles, characterized in that, include: The data acquisition and fusion module is used to collect vehicle network time-series data, environmental monitoring data and business record data generated by new energy commercial vehicles during operation. It performs time synchronization and spatial alignment on the collected data, and performs fusion processing based on Kalman filter to generate a fusion dataset with a unified spatiotemporal reference. The data cleaning and repair module is used to perform rule cleaning and missing data repair on the fused dataset in sequence to generate cleaned structured data. The feature enhancement module is used to extract and compute high-order scene feature parameters based on structured datasets to generate enhanced feature sets; The high-order scenario characteristic parameters include at least the energy consumption fluctuation rate per unit mileage, the average load rate-speed coupling coefficient, and the typical braking energy recovery efficiency. The behavior classification module is used to input the enhanced feature set into the trained classification model, identify and output driving behavior categories; the behavior categories include at least smooth driving, aggressive driving, economical driving and heavy-load hill climbing mode; The comprehensive assessment and report generation module is used to calculate a comprehensive driving behavior score based on the driving behavior category and the corresponding enhanced feature set parameters, combined with preset safety and economy assessment rules, and generate an assessment report.

Images