A method for evaluating the behavior of a heavy truck driver and for guiding economical driving

Abstract

The application discloses a heavy truck driver behavior evaluation and economic driving guidance method, utilizes big data extraction technology to extract vehicle operation data required by a model from a multi-source data set, and further completes data abnormal frame identification, data cleaning and vehicle feature expansion, and constructs a driving behavior data set based on operation interruption time division; a K-means algorithm is adopted to construct a clustering analysis model of the driving behavior data set, four types of driving behavior clusters, i.e., high-speed driving, relaxed driving, moderate driving and aggressive driving, are generated; for the aggressive driving behavior segment, a Transformer instantaneous fuel consumption rate prediction model is adopted to evaluate the energy consumption level, and then a dynamic programming algorithm model with the energy consumption optimization as a target is established, and the energy consumption level comparison and the best fuel economy strategy planning of the driving behavior segment are completed. The application realizes driving behavior type identification based on vehicle networking data, provides an economic operation scheme for high energy consumption driving segments, is favorable for standardizing in-route driving operation of drivers, and provides a reliable strategy for fuel consumption optimization.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of vehicle economic driving, and particularly relates to a heavy truck driver behavior evaluation and economic driving guidance method. BACKGROUND

[0002] The social and economic development drives the improvement of consumption level, and the improvement of transportation facilities corresponds to the progress of transportation tools. The motor vehicle ownership in China is increasing. The fuel-powered vehicles developed in the industrial era still dominate the motor vehicles and become the transportation tools of more and more people. As of 2023, the ownership of vehicles has reached 336 million, and the fuel consumption and environmental pollution caused by vehicle driving present a more severe situation. The demand for improving vehicle fuel economy and energy saving and emission reduction is increasing. The current transportation industry mainly includes shipping, air transportation, rail transportation, public transportation and pipeline transportation. The public transportation occupies the first place in the total amount of comprehensive transportation freight with 73.8%. The heavy-duty trucks dominate the highway freight transportation due to their superior transportation capacity, transportation efficiency and wide transportation range. The high operating cost of heavy-duty trucks restricts the development of the heavy truck market, and the fuel consumption cost is an important reason for the high operating cost of heavy trucks. Therefore, the present application is developed for the purpose of reducing the fuel consumption of heavy-duty trucks.

[0003] There are many factors affecting the fuel consumption of heavy-duty trucks, and the relationship between each factor is complex and mutually influenced. The vehicle structure has been perfected, and the external environmental factors are difficult to improve effectively in a short time. As a relatively controllable factor, the driving behavior is concerned by many scholars and engineers in the current hot field of automation research. Scientific driving behavior can effectively reduce unnecessary fuel consumption, vehicle wear and tear and exhaust emission, improve transportation benefit and vehicle service life, and due to the development of technology, the on-board terminal OBD (On Board Diagnostics) device can collect real-time vehicle operating state data, which contains parameters related to driver behavior, such as gear information, engine start and brake time, throttle opening, etc., so that complete driver behavior analysis is possible.

[0004] The vehicle fuel consumption prediction model can map the driving behavior through fuel consumption and driving characteristics, thereby predicting fuel consumption through simulating driving behavior, and comparing to obtain fuel saving strategies. The existing driving data is summarized to correspond to the fuel consumption of driving behavior, to provide more accurate basis for fuel-saving driving suggestions or driving behavior correction. The dynamic programming is used to prompt the driver before the driving behavior occurs, to realize the effective application of fuel-saving strategies. Therefore, the research on driving behavior can focus on the vehicle fuel consumption prediction model, and the dynamic programming is used to optimize the fuel-saving strategies on this basis. SUMMARY

[0005] The application provides a heavy truck driver behavior evaluation and economic driving guidance method, which can realize identification of the behavior type of a heavy truck driver and economic speed planning of a high-energy-consumption driving segment.

[0006] The application provides a heavy truck driver behavior evaluation and economic driving guidance method, which can realize identification of the behavior type of a heavy truck driver and economic speed planning of a high-energy-consumption driving segment.

[0007] (1) Multi-source data acquisition: data including a timestamp, an elevation and a vehicle speed are acquired through a GPS module, and data including an engine speed, an engine required torque and gear information are acquired through a CAN bus;

[0008] (2) Data cleaning and feature expansion: corresponding cleaning strategies are implemented for error frame data such as missing and noise in running data, and interpolation, filtering and other preprocessing measures are taken for running data that does not meet the continuity condition; then, road slope features, vehicle acceleration and jerk features at all times in a whole road section are acquired through the preprocessed data;

[0009] (3) Driver behavior data set construction: a trip segment is divided according to the parking and engine-off time length, extreme value, mean value, standard deviation and other characteristic indexes of vehicle running data in each trip segment are acquired, strong correlation features are extracted based on principal component analysis algorithm dimension reduction, and a driver behavior feature data set is constructed;

[0010] (4) Driver behavior feature clustering evaluation: K-means algorithm is used for clustering analysis of the segment driving behavior data set, four clustering centers of high-speed driving, relaxed driving, moderate driving and aggressive driving are constructed, and then the clustering results of each driving behavior segment are output;

[0011] (5) Driving segment fuel consumption prediction: a transient fuel consumption rate prediction model based on a Transformer is constructed, vehicle running features and extended road structure features after preprocessing are selected as model inputs, and the transient fuel consumption rate level of each driving behavior segment is output;

[0012] (6) Driving segment economic speed planning: a dynamic programming solving model with the minimum energy consumption as the target is constructed, the transient fuel consumption rate model constructed in the foregoing is selected as the cost function, and the economic driving speed scheme with the optimal energy consumption is solved and output.

[0013] Further, the implementation process of the step (2) is as follows:

[0014] Considering the difference of the running features of non-homologous data, error frames in the GPS source data and the CAN bus source data are artificially extracted and labeled respectively, error frame data samples are separated based on the continuity criterion, and the multi-source data error frame features and cleaning strategies are as follows.

[0015] GPS source data: The main source of error frames is data interruption. Combined with data feedback timestamp, it can be divided into short-time data loss and long-time signal interruption. Set the time threshold δ = 30 s. For short-time data loss less than the threshold, linear interpolation is used for the nearest non-error data value before and after the error frame. For long-time signal interruption exceeding the threshold, the data during the interruption is discarded.

[0016] CAN bus source data: The main error frame data is short-time data loss, high-frequency measurement noise and non-continuous mutation.

[0017] Short-time data loss mainly faces vehicle gear data detection. Based on the principle of continuous change of vehicle gear, the last non-anomalous data is used to fill the short-time single-point missing data.

[0018] High-frequency measurement noise mainly faces engine speed and engine torque data detection. Savitzky-Golay filter is designed to process noise data.

[0019] Non-continuous mutation mainly faces engine torque data detection. The feature detection standard is set as:

[0020]

[0021] Where R is the engine torque data at a certain time, T is the corresponding time stamp, t is the calculation time, z is the time step, and δ is the detection threshold. For the data points obtained according to the above detection standard, linear interpolation method is used to cover the corresponding time data value.

[0022] After cleaning the vehicle running data, based on the vehicle parameter attributes and running characteristics, the vehicle running speed characteristics are obtained:

[0023]

[0024] Then the vehicle running acceleration and jerk characteristics are obtained:

[0025]

[0026] Where n is the engine speed, r is the wheel rolling radius, i g is the transmission ratio, i0 is the rear axle ratio, Δt is the time step, V t is the vehicle speed, a t is the acceleration value, and j is the jerk.

[0027] Using the preprocessed vehicle speed data and elevation data, the road slope is obtained:

[0028]

[0029] wherein, l is a time step calculated backward; h t is an elevation value at time t, h t+n is an elevation value at time t+n, unit: m; v i is a vehicle speed value, unit:

[0030] According to the distance span 100m, the point slope is calculated, and the elevation data in the single calculation step is ensured to be monotonic, and the time step k satisfies:

[0031]

[0032] (h i -h i-1 )·(h i+1 -h i )>0,(i=t+1,t+2,…,t+n-1)

[0033] Based on the above calculation strategy, the road slope characteristics, vehicle acceleration and jerk characteristics at all times of the whole road section are obtained.

[0034] Further, the implementation process of the step (3) is as follows:

[0035] The parking shutdown time threshold Δt is set, the time points at which the parking shutdown time exceeds the threshold are marked, and the vehicle operation data between the two parking time points is recorded as a driving trip segment, that is, the driving behavior segment is divided.

[0036] Based on the preprocessed vehicle operation data, the extreme value, mean value, standard deviation and other characteristic indexes of speed, acceleration, throttle and other data (throttle data obtained by CAN bus) are expanded, and the complete 20 driving behavior characteristic parameter indexes are as follows.

[0037] The Pearson correlation coefficient is used to evaluate the correlation strength of the 20 characteristic parameters with the fuel consumption rate, and then the principal component analysis algorithm is used to evaluate the cumulative variance contribution rate of the complete characteristic parameters. According to the driving trip segment divided in the foregoing, the dimension reduction feature group of the vehicle operation data in each segment is obtained, and a driver behavior feature data set is constructed.

[0038] Further, the implementation process of the step (4) is as follows:

[0039] Based on K-means, a clustering model is constructed, data objects with high similarity are divided into the same cluster, and data with high dissimilarity are divided into different clusters, so as to effectively distinguish different operating conditions of the vehicle, and the specific implementation steps are as follows:

[0040] (1) Set the number of clusters K = 4, corresponding to four types (driver behavior characteristic types) of high-speed driving, relaxed, moderate, and aggressive, and use m feature parameters.

[0041] (2) Randomly select K initial cluster center points μ.

[0042] (3) For each driving behavior segment in the driver behavior characteristic data set, calculate the distance from each cluster center using the Euclidean distance, and the calculation formula is:

[0043]

[0044] In the formula, x i is the sample, μ j is the center cluster, and x ik is the kth parameter of the sample coordinate.

[0045] (4) Assign each driving behavior segment to the cluster to which the nearest cluster center point belongs, i.e., the corresponding cluster with the smallest Euclidean distance, and update the cluster center with the mean of all samples in the cluster to minimize the loss function of the function convergence:

[0046]

[0047] (5) Repeat step 4 until the cluster assignment changes by less than a predetermined threshold.

[0048] The above steps complete the K-means clustering analysis based on the driver behavior characteristic data set, and generate four types of cluster centers of high-speed driving, relaxed, moderate, and aggressive, where the aggressive driving behavior cluster represents high-energy consumption driving behavior of the vehicle. In actual vehicle operating condition applications, input the driving behavior segment data in the above pre-trained clustering model according to a unified format, and obtain the driver behavior type of the corresponding driving behavior segment; extract the driving behavior data set under the aggressive cluster for subsequent economic speed planning as the optimized driving behavior segment.

[0049] Further, the implementation process of step (5) is as follows:

[0050] A transient fuel consumption rate prediction model is constructed based on a Transformer encoder, which is used to evaluate the energy consumption level of the vehicle under different driving speeds and different operating conditions.

[0051] The input parameter of the Transformer encoder is the vehicle operation data after preprocessing, the input sequence length is consistent with the length of the clustering model using the same driving behavior segment segmentation criterion as the K-means clustering model; the decoder with the multi-head attention mechanism is used to obtain the sequence prediction result, and the output result is the prediction value of the instantaneous fuel consumption rate corresponding to each time of the driving behavior segment.

[0052] Further, the implementation process of the step (6) is as follows:

[0053] A dynamic programming solving model with the objective of minimizing energy consumption is constructed, and the optimal vehicle speed at each step is derived based on the Bellman optimality principle at each stage, thereby realizing global problem solving. Iterative dynamic programming is a multi-stage decision problem, and the dynamic programming algorithm steps proposed in the patent are as follows:

[0054] (1) Stage design and parameter selection: the global energy consumption minimization problem is decomposed into sub-problems, and the speed planning scheme that minimizes energy consumption is solved in each time step, considering the slope as an influencing factor of the driver's behavior, using the road slope extension feature as the system input variable, using the vehicle running speed as the system state variable, and using the engine output torque as the system decision variable.

[0055] (2) State transition equation establishment: the state transition equation is constructed based on the vehicle longitudinal dynamics, and the vehicle longitudinal dynamics equation used is:

[0056]

[0057] (3) Cost function design: the aforementioned Transformer-based instantaneous fuel consumption rate prediction model is used, the input parameter is adjusted to the speed variable and engine demand torque variable of each stage of the dynamic programming model, the prediction model output is the instantaneous fuel consumption rate corresponding to each stage of the dynamic programming, and the fuel consumption per 100 kilometers prediction value is calculated based on the instantaneous fuel consumption rate prediction value as the decision cost function;

[0058] (4) Recursive solution: based on the state transition equation, the optimal solution of the vehicle speed at each stage is sequentially derived, and the energy-optimal economic driving speed scheme is obtained.

[0059] Based on the above dynamic programming solving steps, the global economic speed planning of the driving behavior segment is completed, the economic driving scheme, the global instantaneous fuel consumption rate prediction value and the fuel consumption per 100 kilometers prediction value are output for the driver, and the original driving behavior speed spectrum and fuel consumption rate are compared to provide a reference for the driver to drive the vehicle.

[0060] Beneficial effects: compared with the prior art, the beneficial effects of the present application are: 1, the present application considers the technical advantages and technical level of the existing heavy goods vehicle, combines the data content and characteristics of the Internet of vehicles, adopts mature data preprocessing methods for more typical Internet of vehicles data types, and completes the economic speed planning of the high energy consumption driving segment based on the known vehicle and road state information on the basis of accurately identifying the driving behavior characteristics; 2, the present application adopts K-means clustering model, and can accurately and efficiently realize high-precision identification of the driving behavior type by preprocessing the Internet of vehicles data, forming four types of driving behavior types of high-speed driving, moderate, moderate and aggressive, which can basically cover the classification needs of various traffic scenes; 3, based on the identification of high energy consumption driving behavior segment, the present application adopts a dynamic programming method to construct a minimum energy consumption solving problem model, and combines the Transformer instantaneous fuel consumption rate prediction model, so as to ensure the accuracy and optimality of the driving speed scheme at each time step in the driving behavior segment. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 is a flowchart of the present application. DETAILED DESCRIPTION

[0062] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all.

[0063] The present application provides a heavy truck driver behavior evaluation and economic driving guidance method, which can realize the identification of the heavy truck driver behavior type and the economic speed planning of the high energy consumption driving segment.

[0064] Technical scheme: the heavy truck driver behavior evaluation and economic driving guidance method provided by the present application comprises the following steps:

[0065] (1) Multi-source data acquisition: the data including timestamp, elevation and vehicle speed are acquired through the GPS module, and the data including engine speed, engine demand torque and gear information are acquired through the CAN bus;

[0066] (2) Data cleaning and feature expansion: corresponding cleaning strategies are implemented for the missing, noise and other error frame data in the running data, and interpolation, filtering and other preprocessing measures are taken for the running data that does not meet the continuity condition; then the road slope feature, vehicle acceleration and jerk feature at all times of all road sections are obtained through the preprocessing data;

[0067] (3) Driver behavior dataset construction: According to the parking shutdown time length, the trip segments are divided, the extreme value, mean value, standard deviation and other characteristic indexes of the vehicle running data in each trip segment are obtained, the strong correlation features are extracted based on the principal component analysis algorithm, and the driver behavior feature dataset is constructed;

[0068] (4) Driver behavior feature clustering evaluation: based on the K-means algorithm, the clustering analysis is performed on the segment driving behavior dataset, four clustering centers of high-speed driving, moderate, moderate, and aggressive are constructed, and then the clustering results of each driving behavior segment are output;

[0069] (5) Driving segment fuel consumption prediction: a transient fuel consumption rate prediction model based on Transformer is constructed, the preprocessed vehicle running features and extended road structure features are selected as the model input, and the transient fuel consumption rate level of each driving behavior segment is output;

[0070] (6) Driving segment economy speed planning: a dynamic programming solving model is constructed to minimize energy consumption, the transient fuel consumption rate model constructed in the foregoing is selected as the cost function, and the energy-optimal economy driving speed scheme is solved and output.

[0071] Further, the implementation process of step (2) is as follows:

[0072] Considering the difference of non-homologous data running characteristics, the error frames in GPS source data and CAN bus source data are respectively artificially extracted and labeled, the error frame data samples are separated based on the continuity criterion, and the multi-source data error frame characteristics and cleaning strategy are as follows.

[0073] GPS source data: error frames mainly come from data interruption, combined with data feedback timestamp, can be divided into short-time data loss and long-time signal interruption, set time threshold δ = 30s. For short-time data loss less than the threshold, linear interpolation is used for the nearest non-error data value before and after the error frame. For long-time signal interruption exceeding the threshold, the data during the interruption is discarded.

[0074] CAN bus source data: error frame data mainly includes short-time data loss, high-frequency measurement noise and non-continuity mutation.

[0075] Short-time data loss mainly faces vehicle gear data detection, based on the principle of continuous change of vehicle gear, the last non-abnormal data is used to fill the short-time single-point missing data.

[0076] High-frequency measurement noise mainly faces engine speed and engine torque data detection, a Savitzky-Golay filter is designed to process noise data, the filter is constructed by a quadratic polynomial, and the filter window length is set to 19.

[0077] The discontinuous mutation mainly faces the engine torque data detection, and sets the feature detection standard as:

[0078]

[0079] Wherein, R is the engine torque data at a certain time, T is the time stamp at the corresponding time, t is the calculation time, z is the time step, and δ is the set detection threshold; for the data points obtained according to the above detection standard, the linear interpolation method is used to cover the data value at the corresponding time.

[0080] After the vehicle operation data cleaning is completed, the vehicle operation speed characteristics are obtained based on the vehicle parameter attributes and operation characteristics:

[0081]

[0082] Further, the vehicle running acceleration and jerk characteristics are obtained:

[0083]

[0084] Wherein, n is the engine speed, r is the wheel rolling radius, i g is the transmission ratio, i0 is the rear axle ratio, Δt is the time step, V t is the vehicle speed, a t is the acceleration value, and j is the jerk;

[0085] Using the pretreated vehicle speed data and elevation data, the road slope is obtained:

[0086]

[0087] In the formula, l is the time step calculated backward; h t is the elevation value at t time, h t+n is the elevation value at t+n time, unit: m; v i is the vehicle speed, unit:

[0088] According to the distance span of 100 m, the point slope is calculated, and the elevation data in the single calculation step is ensured to be monotonic, and the time step k satisfies:

[0089]

[0090] (h i -h i-1 )·(h i+1 -h i )>0,(i=t+1,t+2,…,t+n-1)

[0091] Based on the above calculation strategy to obtain the full section of the road slope characteristics, vehicle acceleration and jerk characteristics of all time.

[0092] Further, the implementation process of step (3) is as follows:

[0093] Set the parking shutdown duration threshold Δt, mark the time points when the parking shutdown duration exceeds the threshold, and record the vehicle operation data between the two parking time points as a driving trip segment, i.e., divide the driving behavior segment.

[0094] Based on the pre-processed vehicle operation data, the extreme value, mean value, and standard deviation of the speed, acceleration, and throttle data (CAN bus acquired throttle data) are expanded to obtain 20 driving behavior characteristic parameter indicators as shown in the following table.

[0095]

[0096] The Pearson correlation coefficient is used to evaluate the correlation between the 20 feature parameters and the fuel consumption rate, and then the principal component analysis algorithm is used to evaluate the cumulative variance contribution rate of the complete feature parameters. The top 10 summary features with a cumulative variance contribution rate of 95% form the dimensionality reduction feature group. According to the driving trip segments divided as described above, the dimensionality reduction feature group of the vehicle operation data in each segment is obtained to form the driver behavior feature data set.

[0097] Further, the implementation process of step (4) is as follows:

[0098] Based on K-means clustering model, data objects with high similarity are divided into the same cluster, and data with high dissimilarity are divided into different clusters to effectively distinguish different operating conditions of the vehicle. The specific implementation steps are as follows:

[0099] (1) Set the number of clusters K = 4, corresponding to high-speed driving, moderate, moderate, and aggressive four types (driver behavior feature types), and use m = 10 feature parameters.

[0100] (2) Randomly select K initial cluster center points μ.

[0101] (3) For each driving behavior segment in the driver behavior feature data set, use the Euclidean distance to calculate the distance from each cluster center, and the calculation formula is:

[0102]

[0103] In the formula, x i is the sample, μ j is the center cluster, and x ik is the kth parameter of the sample coordinate.

[0104] (4) Assign each driving behavior segment to the cluster to which the nearest cluster center point belongs, i.e., the corresponding cluster with the smallest Euclidean distance, and update the cluster center with the mean of all samples in the cluster to minimize the loss function that converges the function:

[0105]

[0106] (5) Repeat step 4 until the amplitude of cluster assignment changes is less than a preset threshold.

[0107] The K-means clustering analysis based on the driver behavior feature dataset is completed through the above steps, and the cluster centers of four types of high-speed driving, moderate, moderate, and aggressive are generated, where the aggressive driving behavior cluster represents high-energy consumption driving behavior of the vehicle. In actual vehicle operation application, the driving behavior segment data is input into the above pre-trained clustering model in a unified format, and the driver behavior type of the corresponding driving behavior segment is obtained; the driving behavior dataset under the aggressive clustering cluster is extracted as the optimized driving behavior segment for subsequent economic speed planning.

[0108] Further, the implementation process of step (5) is as follows:

[0109] A transient fuel consumption rate prediction model is constructed based on a Transformer encoder, which is used to evaluate the energy consumption level of the vehicle under different driving speeds and different operating conditions.

[0110] The input parameters of the Transformer encoder are the above-mentioned pre-processed vehicle operating data, the driving behavior segment segmentation criteria consistent with the K-means clustering model, and the input sequence length consistent with the clustering model length; a decoder with a multi-head attention mechanism is used to obtain sequence prediction results, and the output result is the transient fuel consumption rate prediction value corresponding to each time of the driving behavior segment.

[0111] Further, the implementation process of step (6) is as follows:

[0112] A dynamic programming solution model is constructed to minimize energy consumption, and based on the Bellman optimality principle, the optimal speed at each step is derived at each stage to solve the global problem. Iterative dynamic programming is a multi-stage decision problem, and the dynamic programming algorithm steps proposed in this patent are as follows:

[0113] (1) Stage design and parameter selection: The global energy consumption minimization problem is decomposed into sub-problems, and the speed planning scheme that minimizes energy consumption is solved at each time step, taking into account the slope as an influencing factor of the driver's behavior, using the road slope extension feature as the system input variable, using the vehicle operating speed as the system state variable, and using the engine output torque as the system decision variable.

[0114] (2) State transition equation establishment: based on the vehicle longitudinal dynamics to construct the state transition equation, the vehicle longitudinal dynamics equation used is:

[0115]

[0116] Use vehicle acceleration as an intermediate variable:

[0117]

[0118] Based on the kinematic relationship between speed and distance to construct the state transition equation:

[0119] u k+1 = u k + a k t k

[0120]

[0121] In the formula, a k is the vehicle acceleration obtained by calculating the fuel economy optimization of the kth stage, u k is the vehicle running speed of the kth stage, and d k is the running distance of the vehicle from the kth stage to the k+1th stage.

[0122] (3) Cost function design: using the aforementioned constructed Transformer-based instantaneous fuel consumption rate prediction model, adjusting the input parameters to the speed variables and engine demand torque variables of each stage of the dynamic programming model, the prediction model output is the instantaneous fuel consumption rate corresponding to each stage of the dynamic programming, based on the instantaneous fuel consumption rate prediction value to calculate the fuel consumption per 100 kilometers prediction value as the decision cost function, the fuel consumption per 100 kilometers calculation formula is:

[0123]

[0124] (4) Recursion solution: based on the state transition equation, sequentially derive the optimal solution of the vehicle speed of each stage to obtain the energy-optimal economic driving speed scheme.

[0125] Based on the above dynamic programming solution steps, the global economic speed planning of the driving behavior segment is completed, the economic driving scheme, the global instantaneous fuel consumption rate prediction value and the fuel consumption per 100 kilometers prediction value are output to the driver, and the original driving behavior speed spectrum and fuel consumption rate are provided. Comparison, provide driving behavior and vehicle operation scheme reference for the driver.

[0126] The above merely describes preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art, according to the technical solution and inventive concept of the present application, makes equivalent replacement or change within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for heavy truck driver behavior assessment and economic driving guidance, characterized in that, The method comprises the following steps: (1) Multi-source data acquisition: data including time stamp, elevation, and vehicle speed are acquired through a GPS module, and data including engine speed, engine required torque, and gear information are acquired through a CAN bus; (2) Data cleaning and feature expansion: corresponding cleaning strategies are implemented for missing, noise and error frame data in the running data, and interpolation and filtering preprocessing measures are taken for the running data that does not meet the continuity condition; then, the road slope feature, vehicle acceleration and jerk feature at all times of the whole road section are acquired through the preprocessed data; (3) Driver behavior data set construction: the journey segments are divided according to the parking and engine-off time length, the extreme value, mean value and standard deviation feature indexes of the vehicle running data in each journey segment are acquired, the strong correlation features are extracted based on the principal component analysis algorithm, and the driver behavior feature data set is constructed; (4) Driver behavior feature clustering evaluation: the K-means algorithm is used for clustering analysis on the segment driving behavior data set, four clustering centers of high-speed driving, relaxed, moderate and aggressive are constructed, and then the clustering results of each driving behavior segment are output; (5) Driving segment fuel consumption prediction: a transient fuel consumption rate prediction model based on a Transformer is constructed, the preprocessed vehicle running features and the expanded road structure features are selected as the model inputs, and the transient fuel consumption rate level of each driving behavior segment is output; (6) Economic driving speed planning of driving segment: a dynamic programming solving model with the minimum energy consumption as the target is constructed, the transient fuel consumption rate prediction model constructed in the foregoing is selected as the cost function, and the energy consumption optimal economic driving speed scheme is solved and output. The step (2) is implemented as follows:

2. A method of evaluating the behavior of a heavy goods vehicle driver and guiding economical driving according to claim 1, characterized in that, Considering the difference of the running features of non-homologous data, the error frames in the GPS source data and the CAN bus source data are manually extracted and labeled respectively, the error frame data samples are separated based on the continuity criterion, and the multi-source data error frame features and cleaning strategies are as follows: The GPS source data: the error frames are mainly caused by data interruption, and can be divided into short-time data loss and long-time signal interruption according to the data feedback time stamp, and a time threshold δ = 30 s is set; For short-time data loss less than the threshold, linear interpolation is performed on the nearest non-error data before and after the error frame; For long-time signal interruption exceeding the threshold, the data during the interruption is discarded; The CAN bus source data: the error frame data mainly includes short-time data loss, high-frequency measurement noise and non-continuity mutation; The short-time data loss is mainly for vehicle gear data detection, and the last non-abnormal data is used to fill the short-time single-point missing data based on the continuous change principle of vehicle gear; The high-frequency measurement noise is mainly for engine speed and engine torque data detection, and a Savitzky-Golay filter is designed to process the noise data; ​ The discontinuous mutation mainly faces the engine torque data detection, and sets the feature detection standard as: Wherein, R is the engine torque data at a certain time, T is the time stamp at the corresponding time, t is the calculation time, z is the time step, and δ is the set detection threshold; for the data points obtained according to the above detection standard, the linear interpolation method is used to cover the data values at the corresponding time; After the vehicle operation data cleaning is completed, the vehicle operation speed characteristics are obtained based on the vehicle structure and operation characteristics: Then the vehicle running acceleration and jerk characteristics are obtained: where n is the engine speed, r is the wheel rolling radius, i g is the transmission gear ratio, i0is the rear axle ratio, At is the time step, V t is the vehicle speed value, a t is the acceleration value, j is the jerk, V t-1 denotes the vehicle speed value at the previous time, a t-1 denotes the acceleration value at the previous time; The road slope is obtained by using the preprocessed speed data and elevation data: wherein, l is the time step calculated backward; h t is the elevation value at time t, h t+l is the elevation value at time t+l, unit: m; v i is the vehicle speed value, unit: v i+1 represents the vehicle speed value at next time The point slope is calculated according to the distance span of 100m, and the elevation data is ensured to be monotonic within a single calculation step, and the time step k satisfies: (h i -h i-1 )·(h i+1 -h i )>0, (i=t+1,t+2,…,t+n-1) Based on the above calculation strategy and formula, traverse all running segment data to obtain the road gradient characteristics, vehicle acceleration and jerk characteristics of the entire journey.

3. A method of evaluating the behavior of a heavy goods vehicle driver and guiding economical driving according to claim 2, characterized in that, The implementation process of the step (3) is as follows: A parking and engine-off time threshold Δt is set, time points with parking and engine-off time longer than the threshold are marked, and vehicle operation data between the two parking time points is recorded as a driving trip segment, that is, the driving behavior segment is divided; Based on the preprocessed vehicle operation data, the extreme value, mean value and standard deviation characteristics of the speed, acceleration and throttle data are extended, and 20 driving behavior characteristic parameter indexes are completely counted; The Pearson correlation coefficient is used to evaluate the correlation between the 20 characteristic parameters and the fuel consumption rate, and then the principal component analysis algorithm is used to evaluate the cumulative variance contribution rate of the complete characteristic parameters; according to the driving trip segment divided in the foregoing, the dimension reduction feature group of the vehicle operation data in each segment is obtained, and the driver behavior feature data set is constructed.

4. The method of evaluating driver behavior and guiding economical driving of a heavy duty vehicle according to claim 3, wherein The implementation process of the step (4) is as follows: Based on K-means, a clustering model is constructed, data objects with high similarity are divided into the same cluster, and data with high dissimilarity is divided into different clusters, so as to effectively distinguish different vehicle operating conditions, and the specific implementation steps are as follows: (1) Set the number of clusters K=4, corresponding to high-speed driving, moderate, moderate and aggressive four types; (2) Randomly select K initial cluster center points μ; (3) For each driving behavior segment in the driver behavior feature data set, the Euclidean distance between each cluster center is calculated, and the calculation formula is: where x i is a sample, μ j is a center cluster, x ik is the kth parameter of the sample coordinate, μ jk denotes the kth center cluster; (4) Each driving behavior segment is assigned to the cluster to which the nearest cluster center point belongs, that is, the corresponding cluster with the minimum Euclidean distance, and the cluster center is updated with the mean value of all samples in the cluster, so that the minimum loss function of the function convergence is: wherein C i represents a set of all samples contained in the i-th cluster; (5) Repeat step 4 until the cluster assignment changes less than the preset threshold; Through the above steps, K-means clustering analysis based on the driver behavior feature data set is completed, and the clustering centers of high-speed driving, moderate, moderate and aggressive four types are generated, wherein the aggressive driving behavior cluster represents high-energy driving behavior of the vehicle operation; In actual vehicle operating condition application, the driving behavior segment data is input into the above pre-trained clustering model according to the unified format, and the driver behavior type of the corresponding driving behavior segment is obtained; The driving behavior data set under the aggressive clustering cluster is extracted as the to-be-optimized driving behavior segment for subsequent economic speed planning.

5. A method of evaluating the behavior of a heavy goods vehicle driver and guiding economical driving according to claim 4, characterized in that, The implementation process of the step (5) is as follows: A prediction model of instantaneous fuel consumption rate is constructed based on a Transformer encoder, which is used to evaluate the energy consumption level of the vehicle under different driving speeds and different operating conditions. The input parameters of the Transformer encoder are the vehicle operating data after the preprocessing, the input sequence length is consistent with the length of the clustering model using the same driving behavior segment division criterion as the K-means clustering model; the decoder with a multi-head attention mechanism is used to obtain the sequence prediction result, and the output result is the prediction value of the instantaneous fuel consumption rate corresponding to each time of the driving behavior segment.

6. The method of evaluating driver behavior and guiding economical driving of a heavy duty vehicle according to claim 1, wherein The implementation process of the step (6) is as follows: The dynamic programming algorithm steps are as follows: (1) Stage design and parameter selection: the global energy consumption minimization problem is decomposed into sub-problems, and the speed planning scheme that minimizes the energy consumption is solved in each time step, considering the slope as an influencing factor of the driver's behavior, using the road slope extension feature as the system input variable, using the vehicle operating speed as the system state variable, and using the engine output torque as the system decision variable; (2) State transition equation establishment: the state transition equation is constructed based on the vehicle longitudinal dynamics, and the vehicle longitudinal dynamics equation used is: In the formula, T tq i is the engine torque; i0 is the engine rear axle speed ratio; i g The gear ratio; η T R is the transmission system efficiency; r is the wheel radius; C D ρ is the air drag coefficient; A is the vehicle's frontal area; ρ is the air density; u is the vehicle speed; M is the vehicle mass; g is the acceleration due to gravity; α is the road slope angle; f is the road surface friction coefficient; σ is the rotational mass conversion factor. To accelerate the vehicle; (3) Cost function design: the aforementioned prediction model of instantaneous fuel consumption rate based on Transformer is used, the input parameters are adjusted to the speed variables and engine demand torque variables of each stage of the dynamic programming model, the prediction model output is the instantaneous fuel consumption rate corresponding to each stage of the dynamic programming, the prediction value of the fuel consumption per 100 kilometers is calculated based on the prediction value of the instantaneous fuel consumption rate, which is used as the decision cost function; (4) Recursive solution: based on the state transition equation, the optimal solution of the vehicle speed of each stage is sequentially derived, and the economic driving speed scheme with optimal energy consumption is obtained; Based on the above dynamic programming solving steps, the global economic speed planning of the driving behavior segment is completed, the economic driving scheme, the global instantaneous fuel consumption rate prediction value and the fuel consumption per 100 kilometers prediction value are output for the driver, and the original driving behavior speed spectrum and the fuel consumption rate are compared, which provides a reference for the driver to adjust the driving behavior and vehicle operating scheme.

Images