A method for identifying driver behavior habit of adaptive commercial vehicle load

Abstract

The application relates to a self-adaptive commercial vehicle load driving behavior habit identification method, mainly comprising the following steps: offline data collection; data preprocessing and feature definition; clustering of the features; storage of the clustering identification results of the online driving vehicle into a result matrix; conversion of the result matrix into a probability matrix; calculation of the information entropy and entropy weight value of the driving behavior habit; and finally, the driving behavior habit classification of the driver is embodied in different intervals in the time domain; the application can identify the behaviors of commercial vehicle drivers under different working conditions and different loads, has high adaptability and generalization ability, and further enhances the accuracy of the identification method by combining time domain information, rolling calculation and result correction of online data.

Description

Technical Field

[0001] This invention relates to the field of driver behavior habit identification for commercial vehicles, and in particular to a method for identifying driver behavior habits based on the load of a commercial vehicle. Background Technology

[0002] For commercial vehicle fleets or individual users, fuel costs are typically the highest priority for cost management. Given cost pressures, energy conservation and emission reduction in commercial vehicles, a key tool in the logistics industry, are imperative. In the field of commercial vehicle energy conservation, besides optimizing hardware such as spoilers and tires to achieve lower driving resistance, fuel savings can also be achieved by improving driver control. Studying driver behavior can effectively identify driver actions such as pressing the accelerator and braking, thereby improving overall vehicle fuel consumption and emissions through optimized control signals.

[0003] In existing research, driver behavior identification models for commercial vehicles often simply copy the features and learning algorithms used in passenger vehicles. However, in reality, the speed differences between individual passenger vehicles under similar operating conditions are not significant, and the vehicle parameters are relatively similar, which does not match the actual situation of commercial vehicles. This leads to problems such as inappropriate feature selection and low algorithm matching when the driver behavior identification research for passenger vehicles is transferred to commercial vehicles. Summary of the Invention

[0004] This invention proposes a method for identifying the behavioral characteristics of drivers of commercial vehicles applicable to all types and loads. This method can avoid the deviation of traditional driver behavioral characteristics caused by differences in commercial vehicle models and loads, which would affect the accuracy of driving behavior identification.

[0005] The objective of this invention can be achieved by adopting the following technical solution:

[0006] An adaptive method for judging driver behavior habits based on commercial vehicle load consists of a first module, a second module, and a third module. The first module trains an offline classification model using commercial vehicle driving data stored on a cloud platform. The classification model training includes eigenvalue acquisition, data preprocessing, eigenvalue calculation, and clustering. After clustering, cluster centers are obtained, which include aggressive drivers, general drivers, and conservative drivers.

[0007] The second module records the online vehicle data to be classified through the cloud platform, and performs feature calculation and driver behavior habit classification based on the cluster center in the first module at a frequency of 1 time / minute to obtain a single judgment result and record the driver behavior habit category label corresponding to the result.

[0008] The third module performs a rolling calculation based on the category labels of the trip records in the second module.

[0009] The rolling calculation specifically includes the following steps:

[0010] Step S1: Store the driver behavior classification obtained from each data transmission into a category result matrix, as shown in the following expression:

[0011]

[0012] Where, x 11 ,x 12 ,x 13 The result vector obtained from the first data transmission has three elements representing the aggressive, general, and conservative modes, respectively. Based on the calculation results, the corresponding element is assigned a value of 1, and the other two are assigned a value of 0. n1 ,x n2 ,x n3 Let n be the result vector, representing the latest returned data;

[0013] Step S2: Standardize the elements in the resulting matrix according to the following formula to obtain the standardized matrix Z;

[0014]

[0015] Where, x ij For a single element in the resulting matrix;

[0016] Step S3: Calculate the probability matrix P based on the standardized result matrix Z;

[0017]

[0018] Among them, z ij For a single element in the standardized result matrix;

[0019] Step S4: with p ij Substituting this probability into the information entropy formula:

[0020]

[0021] Among them, e j The information entropy of each label is calculated. The higher the information entropy, the less information it contains and the lower its credibility.

[0022] Step S5: Calculate the entropy weights for each driver behavior habit classification result:

[0023]

[0024] Where w1, w2, and w3 represent the entropy weights of the classification results for aggressive, general, and conservative drivers, respectively. The classification with the largest entropy weight is selected as the result of driver behavior habit discrimination and output.

[0025] Preferably, the features acquired include: vehicle speed sensor signal, throttle position sensor signal, steering wheel angle sensor signal, and brake pedal position sensor signal.

[0026] Preferably, the data processing removes the vehicle speed feature commonly used in traditional studies, and the data processing uses the variance of acceleration rate of change, maximum acceleration, variance of accelerator pedal rate of change, variance of brake pedal rate of change, and variance of steering wheel angle rate of change as feature values ​​to participate in the clustering algorithm.

[0027] Preferably, the method for the cloud platform to record online vehicle data to be classified includes: acquiring sensor signals in real time through the vehicle terminal T-BOX based on the vehicle's CAN bus, performing online calculations through the vehicle network cloud platform, calculating the Euclidean distance with the cluster centers of the offline clustering algorithm, and finally transmitting the calculated driver behavior classification results to the vehicle at time intervals.

[0028] Beneficial technical effects of the present invention:

[0029] 1. Feature selection: Commercial vehicles with different loads and driving conditions often have large differences in speed. Since this method excludes the speed feature that traditional algorithms generally rely on, it can be used to identify driver behavior habits in commercial vehicles under various environments.

[0030] 2. Vehicle-to-Everything (V2X) Cloud Platform: Real-time data collection based on the cloud platform makes it possible to continuously optimize the model's judgment results;

[0031] 3. Rolling Optimization of Results: The results of each judgment are constructed into a result matrix, and the entropy weight method is used to calculate each category, thereby reflecting the driver behavior characteristics at different stages in the time domain. At the same time, the results are also corrected based on the original clustering-based judgment, which significantly optimizes the timeliness and accuracy of the judgment. Attached Figure Description

[0032] Figure 1 For the overall technology roadmap

[0033] Figure 2 T-BOX Data Acquisition and Communication Architecture Diagram

[0034] Figure 3 This is a flowchart of the rolling optimization process based on the entropy weight method. Detailed Implementation

[0035] To enable those skilled in the art to understand the technical solution of the present invention more clearly, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0036] In this embodiment, as Figure 1 As shown, the present invention is composed of a first module, a second module and a third module. The first module performs offline classification model training using commercial vehicle driving data stored on a cloud platform. The classification model training includes symptom acquisition, data preprocessing, feature value calculation and clustering. After clustering, cluster centers are obtained. The cluster centers include aggressive drivers, general drivers and conservative drivers.

[0037] The second module records the online vehicle data to be classified through the cloud platform, and performs feature calculation and driver behavior habit classification based on the cluster center in the first module at a frequency of 1 time / minute to obtain a single judgment result and record the driver behavior habit category label corresponding to the result.

[0038] The third module performs rolling calculations based on the category labels of the trip records in the second module. Specifically, it includes the following steps: constructing a category result matrix based on the category labels; standardizing the elements in the matrix; converting it into a probability matrix; calculating the information entropy; obtaining the weight of each category; and finally outputting the category with the highest weight as the classification result of the driver behavior habit model.

[0039] In this embodiment, the features acquired include: vehicle speed sensor signal, throttle position sensor signal, steering wheel angle sensor signal, and brake pedal position sensor signal.

[0040] In this embodiment, the vehicle speed feature commonly used in traditional studies is removed from the data processing, and the data processing uses the variance of the rate of change of acceleration, the maximum acceleration, the variance of the rate of change of the accelerator pedal, the variance of the rate of change of the brake pedal, and the variance of the rate of change of the steering wheel angle as feature values ​​to participate in the clustering algorithm.

[0041] In this embodiment, the specific classification model training process is as follows:

[0042] 1. Remove outliers and convert them into missing values. After interpolating the missing values, calculate the five feature values ​​for each driver sample in the dataset:

[0043] a i =(var_a',a_max,var_acc',var_bra',var_ste')

[0044] 2. Select 3 initial samples as initial cluster centers:

[0045] a = a1, a2, a3

[0046] 3. For each sample b in the dataset i Calculate its distance to the three cluster centers and assign it to the category corresponding to the cluster center with the smallest distance.

[0047] 4. For each category a j Recalculate its cluster centers:

[0048]

[0049] 5. Repeat steps 3 and 4 for a total of 300 iterations;

[0050] The clustering results are shown in the table below:

[0051]

[0052] In this embodiment, as Figure 2 As shown, the above sensor signals are collected in real time through the vehicle terminal T-BOX based on the vehicle's CAN bus, and the vehicle network cloud platform is used for online calculation. The Euclidean distance is calculated with the cluster centers of the offline clustering algorithm, and finally the calculated driver behavior classification results are transmitted to the vehicle at time intervals.

[0053] In this embodiment, as Figure 3 As shown, after traditional clustering algorithms classify drivers, this invention, through the aforementioned T-BOX data acquisition and communication architecture, employs a correction method based on the entropy weight method and rolling optimization concept of the driver behavior habit model, specifically including the following steps:

[0054] Step S1: Store the driver behavior classification obtained from each data transmission into the category result matrix:

[0055]

[0056] Where, x 11 ,x 12 ,x 13 This is the result vector obtained from the first data transmission. The three element positions represent the aggressive, general, and conservative types, respectively. Based on the calculation results, the corresponding element position is assigned a value of 1, and the other two are assigned a value of 0. Similarly, x... n1 ,x n2 ,x n3 This is the result vector for the nth iteration, representing the latest returned data.

[0057] Step S2: Standardize the elements in the resulting matrix according to the following formula to obtain the standardized matrix Z.

[0058] Where, x ij This refers to a single element in the resulting matrix.

[0059] Step S3: Calculate the probability matrix P based on the standardized result matrix Z.

[0060]

[0061] Among them, z ij This represents a single element in the standardized result matrix.

[0062] Step S4: with p ij Substituting this probability into the information entropy formula:

[0063]

[0064] Among them, e j The information entropy of each label is calculated. The higher the information entropy, the less information it contains and the lower its credibility.

[0065] Step S5: Calculate the entropy weights for each driver behavior habit classification result:

[0066]

[0067] Where w1, w2, and w3 represent the entropy weights of the classification results for aggressive, general, and conservative drivers, respectively. The classification with the largest entropy weight is selected as the result of driver behavior habit discrimination and output.

[0068] The above are merely further embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope disclosed in the present invention, based on the technical solution and concept of the present invention, shall fall within the scope of protection of the present invention.

Claims

1. A method for adaptively determining driver behavior habits based on commercial vehicle load, characterized in that: It consists of a first module, a second module and a third module. The first module performs offline classification model training using commercial vehicle driving data stored on the cloud platform. The classification model training includes symptom acquisition, data preprocessing, feature value calculation and clustering. After clustering, cluster centers are obtained. The cluster centers include aggressive drivers, general drivers and conservative drivers. The second module records the online vehicle data to be classified through the cloud platform, and performs feature calculation and driver behavior habit classification based on the cluster center in the first module at a frequency of 1 time / minute to obtain a single judgment result and record the driver behavior habit category label corresponding to the result. The third module performs rolling calculations based on the category labels of the trip records in the second module. The rolling calculation specifically includes the following steps: Step S1: Store the driver behavior classification obtained from each data transmission into a category result matrix, as shown in the following expression: Where, x 11 ,x 12 ,x 13 The result vector obtained from the first data transmission has three elements representing the aggressive, general, and conservative modes, respectively. Based on the calculation results, the corresponding element is assigned a value of 1, and the other two are assigned a value of 0. n1 ,x n2 ,x n3 Let n be the result vector, representing the latest returned data; Step S2: Standardize the elements in the resulting matrix according to the following formula to obtain the standardized matrix Z; Where, x ij For a single element in the resulting matrix; Step S3: Calculate the probability matrix P based on the standardized result matrix Z; Among them, z ij For a single element in the standardized result matrix; Step S4: with p ij Substituting this probability into the information entropy formula: Among them, e j The information entropy of each label is calculated. The higher the information entropy, the less information it contains and the lower its credibility. Step S5: Calculate the entropy weights for each driver behavior habit classification result: Where w1, w2, and w3 represent the entropy weights of the classification results for aggressive, general, and conservative drivers, respectively. The classification with the largest entropy weight is selected as the result of driver behavior habit discrimination and output.

2. The method for determining driver behavior habits based on adaptive commercial vehicle load according to claim 1, characterized in that: The features collected include: vehicle speed sensor signal, throttle position sensor signal, steering wheel angle sensor signal, and brake pedal position sensor signal.

3. The method for determining driver behavior habits based on adaptive commercial vehicle load according to claim 2, characterized in that: The data processing removed the vehicle speed feature commonly used in traditional studies, and used the variance of acceleration rate of change, maximum acceleration, variance of accelerator pedal rate of change, variance of brake pedal rate of change, and variance of steering wheel angle rate of change as feature values ​​in the clustering algorithm.

4. The method for determining driver behavior habits based on adaptive commercial vehicle load according to claim 3, characterized in that: The method for the cloud platform to record online vehicle data to be classified includes: real-time acquisition of sensor signals based on the vehicle's CAN bus via the vehicle-mounted terminal T-BOX, online calculation via the vehicle network cloud platform, Euclidean distance between the data and the cluster centers of the offline clustering algorithm, and finally, transmission of the calculated driver behavior classification results to the vehicle at time intervals.

Images