A braking energy recovery method for pure electric vehicles based on driver style recognition

Abstract

The present invention discloses a pure electric vehicle braking energy recovery method based on driver style recognition, which belongs to the field of driverless vehicles. The method comprises the following steps: S101: classifying the driver style by preprocessing the NGSIM data set and extracting and optimizing the driver style feature parameters; S102: training the sample data using a convolutional neural network to establish a driver style recognition model; S103: analyzing the driver's braking intention by the driver style type, distributing the braking force according to the proposed energy recovery strategy, and controlling the motor to recover energy. The present invention can identify different driving styles of the driver, and match the best braking energy recovery mode according to the driving style, thereby increasing the mileage of the pure electric vehicle, improving the driver's experience, and achieving the energy consumption economy of the pure electric vehicle.

Description

Technical Field

[0001] The invention relates to the technical field of pure electric vehicle control, and in particular to a pure electric vehicle braking energy recovery method based on driver style recognition. Background Art

[0002] With the start of automobile intelligence, people's demand for a good automobile experience makes people hope that cars can customize corresponding service content and assisted driving according to their own status and needs, accurately adapt to the driver's style, and provide drivers with more humane services and safer and more comfortable assisted driving. Braking energy recovery is an important means to improve the energy utilization efficiency of electric vehicles. When the motor is running in the power generation state, it can generate braking torque to achieve vehicle deceleration. At the same time, part of the braking energy of the vehicle is converted into electrical energy to charge the power battery, thereby increasing the vehicle's driving range.

[0003] The Chinese patent with the authorization announcement number CN 108081961 B discloses "a brake energy recovery control method, device and electric vehicle". The method includes: determining the total required braking force of the driver according to the brake pedal displacement, and determining the required braking force of the motor in combination with the maximum braking force that the motor can provide and the judgment threshold of the braking force distribution, and then controlling the motor to perform brake energy recovery. The method performs electric brake recovery to the greatest extent without changing the total required braking force of the driver, and can achieve higher energy recovery efficiency.

[0004] The Chinese patent with authorization announcement number CN 109278566 B discloses a "rear-wheel drive pure electric vehicle brake energy recovery control method and device". The method includes: during the vehicle braking process, controlling the distribution of mechanical braking force and regenerative braking force, maximizing the regenerative braking force distribution coefficient on the basis of ensuring braking safety and braking efficiency, making full use of the motor's power feeding capacity to recover the energy lost during braking to charge the power battery, improving the braking energy recovery efficiency, reducing braking energy loss, and optimizing the vehicle's driving range.

[0005] Although the above-mentioned prior art also starts from the perspective of braking energy recovery to achieve the effect of improving the driving range of pure electric vehicles, the regenerative braking energy recovery method of the prior art does not take into account the factors of different driver styles and cannot meet the driver's needs for comfort and driving experience. Summary of the invention

[0006] In order to make up for the shortcomings of the prior art, the purpose of the present invention is to propose a pure electric vehicle braking energy recovery method and system based on driver style recognition, which can identify the different driving styles of drivers and match the best braking energy recovery mode according to their driving styles, thereby increasing the mileage of pure electric vehicles, improving the driver's experience, and achieving energy economy of pure electric vehicles.

[0007] The purpose of the present invention is achieved through the following technical solutions:

[0008] A pure electric vehicle braking energy recovery method and system based on driver style recognition, comprising the following steps:

[0009] S101: classifying the driver style by preprocessing the NGSIM data set and selecting and optimizing the driver style feature parameters;

[0010] S102: Using a convolutional neural network to train sample data and establish a driver style recognition model;

[0011] S103: Analyze the driver's braking intention according to the driver's style type, distribute braking force according to the proposed energy recovery strategy, and control the motor to perform energy recovery.

[0012] Furthermore, the step S101 specifically includes the following steps:

[0013] S1011: Remove abnormal data from the NGSIM data set, use the Savitzky-Golay filter algorithm to smooth and remove noise from the data stream, and use the max-min method to standardize. The function of the Savitzky-Golay filter algorithm is:

[0014]

[0015] in is the smoothing coefficient obtained by least squares fitting.

[0016] S1012: Select average vehicle speed v mean , maximum vehicle speed v max , minimum vehicle speed v min , speed standard deviation σ v , average acceleration a mean , maximum acceleration a max , minimum acceleration a min , acceleration standard deviation σ a and the distance l to the front vehicle are the driver style characteristics and their specific parameters are calculated and optimized. The factor analysis method is used to reduce the dimension of the characteristic parameters. The factor analysis model is expressed as:

[0017] xi =a1F1+a2F2+...+a p F p +η i

[0018] where x i is the i-th observable variable (i=1, 2, 3, ..., k), F j is a common factor (j=1, 2, 3, ..., p), and p<k; a p is the factor loading.

[0019] S1013: Select the average vehicle speed v according to the index that the sum of squares of the extracted load is greater than 75%. mean , maximum vehicle speed v max , minimum vehicle speed v min , speed standard deviation σ v The common factors are clustered by K-means++ clustering algorithm to minimize the loss function, and the results are analyzed. The loss function is defined as:

[0020]

[0021] where x i represents the i-th sample, c i is x i The cluster to which it belongs, μ ci Represents the center point corresponding to each cluster, and M is the total number of samples.

[0022] Furthermore, the step S102 specifically includes the following steps:

[0023] S1021: Given an input driving data set, set a penalty function L(y, y^) to determine the error between the output prediction value and the actual value, and use the Adam optimizer to adjust its learning rate and correct the gradient estimate;

[0024] S1022: Select the convolution kernel size as 3 and the number as (16, 64, 128), select Relu as the activation function for acceleration, set the number of neurons in the fully connected layer to 5, and build a driver style recognition model based on the CNN convolutional neural network;

[0025] S1023: Use the categorical cross entropy penalty function and accuracy to evaluate and verify the effectiveness. The cross entropy penalty function formula is:

[0026]

[0027] Among them, N is the total amount of data, K is the number of target tasks, and y i，kis the value of a sample data in the Kth category, p i，k Represents the possible probability of predicting it as the Kth class.

[0028] The accuracy formula is:

[0029]

[0030]

[0031] Furthermore, the step S103 specifically includes the following steps:

[0032] S1031: Inputting the driver's driving data into a driver style recognition model to obtain the current driver's style;

[0033] S1032: analyzing the driver's braking intention according to the identified driver style type, and obtaining a target braking force;

[0034] S1033: Under the condition of meeting the I curve and ECE regulations, the maximum energy recovery control strategy is adopted, and the front and rear axle braking forces and mechanical and friction braking forces are distributed. The I curve expression is:

[0035]

[0036] where F br is the rear axle braking force, F bf is the front axle braking force, m is the vehicle mass, g is the gravity coefficient, h g is the height of the center of mass, b is the distance from the center of mass to the rear axle, and L is the distance between the front and rear axles;

[0037] The ECE regulation expression is:

[0038]

[0039] Where z is the braking intensity.

[0040] S1034: Control the motor to perform braking energy recovery.

[0041] The beneficial effects of the present invention are: compared with the prior art:

[0042] (1) The present invention uses NGSIM data set information to select feature information, preprocesses driver data, selects strongly correlated data, and uses strongly correlated feature data to classify driver styles, thereby reducing the amount of model calculation and improving the recognition speed of the method.

[0043] (2) The present invention utilizes a clustering algorithm to classify driver feature information, and employs a convolutional neural network to train sample data to establish a driver style recognition model, thereby improving the recognition accuracy of the driver style.

[0044] (3) The present invention can identify different driving styles of drivers and analyze the driver's braking intention according to their driving styles, greatly improving system integration and driving experience.

[0045] The present invention can, under the premise of ensuring the comfort of the driver, adopt a maximum braking energy recovery control strategy to perform energy recovery, increase the mileage of the pure electric vehicle, and achieve energy consumption economy of the pure electric vehicle. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The above and / or additional aspects and advantages of the embodiments of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0047] Figure 1 This is a flow chart of a pure electric vehicle braking energy recovery method based on driver style recognition according to an embodiment of the present invention;

[0048] Figure 2 is a specific flow chart of step S101;

[0049] Figure 3 This is the driving style characteristic parameter distribution result diagram;

[0050] Figure 4 This is the accuracy change diagram of the training set and the validation set;

[0051] Figure 5 The relationship curve between the braking force required by vehicles with different driving styles and the brake pedal displacement;

[0052] Figure 6 FIG. 4 is a vehicle braking force distribution curve diagram in one embodiment. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] The embodiment of the present invention proposes a pure electric vehicle braking energy recovery method based on driver style recognition. Figure 1 The present invention is a flowchart of the method described in the embodiment of the present invention.

[0055] A pure electric vehicle braking energy recovery method based on driver style recognition comprises the following steps:

[0056] S101: classifying the driver style by preprocessing the NGSIM data set and selecting and optimizing the driver style feature parameters;

[0057] S102: Using a convolutional neural network to train sample data and establish a driver style recognition model;

[0058] S103: Analyze the driver's braking intention according to the driver's style type, distribute braking force according to the proposed energy recovery strategy, and control the motor to perform energy recovery.

[0059] Further, as a preferred implementation, Figure 2 As shown, the step S101 specifically includes the following steps:

[0060] S1011: Remove abnormal data from the NGSIM data set, use the Savitzky-Golay filter algorithm to smooth and remove noise from the local x and local y values ​​in the data set, recalculate the velocity and acceleration, and use matrix expressions for optimization. The Savitzky-Golay filter algorithm function is:

[0061]

[0062] in is the smoothing coefficient obtained by least squares fitting.

[0063] Furthermore, the mix-max normalization method is used to match the dimensions and units of different feature parameters in the data set. The function formula is:

[0064]

[0065] Where max(x) is the maximum value of vector x, and min(x) is the minimum value of vector x.

[0066] S1012: Select average vehicle speed v mean , maximum vehicle speed v max , minimum vehicle speed v min , speed standard deviation σ v , average acceleration a mean , maximum acceleration a max , minimum acceleration a min , acceleration standard deviation σ aand the distance l to the front vehicle are the driver style characteristics and their specific parameters are calculated and optimized. The factor analysis method is used to reduce the dimension of the characteristic parameters. The factor analysis model is expressed as:

[0067] x i =a1F1+a2F2+...+a p F p +η i

[0068] where x i is the i-th observable variable (i=1, 2, 3, ..., k), F j is a common factor (j=1, 2, 3, ..., p), and p<k; a p is the factor loading.

[0069] Furthermore, assuming that the influence of each common factor on each characteristic variable is a linear relationship, the Z-score standardization method is used to standardize the sample data:

[0070] x * =(x-μ) / σ

[0071] where x * is the value of the input and output driving data x after standardization, μ is the average value of the driving data to be standardized, and σ represents the standard deviation of the driving data to be standardized.

[0072] Furthermore, it is deduced that the common factor can be expressed as:

[0073] F j =b j1 x1+b j2 x2+...+b jp x p , j = 1, 2, ..., m

[0074] The calculated common factors are standardized data with a mean of 0 and a standard deviation of 1.

[0075] S1013: Select the average vehicle speed v according to the index that the sum of squares of the extracted load is greater than 75%. mean , maximum vehicle speed v max , minimum vehicle speed v min , speed standard deviation σ v The common factors are clustered by K-means++ clustering algorithm to minimize the loss function, and the results are analyzed. The loss function is defined as:

[0076]

[0077] where x irepresents the i-th sample, c i is x i The cluster to which it belongs, μ ci Represents the center point corresponding to each cluster, and M is the total number of samples.

[0078] Furthermore, the sample data set and the initial cluster number k are taken as input, and the initial cluster center point (c1 (1) , c2 (1) , ..., c k (1) ) ; Calculate the Euclidean distance between each data and each cluster center Divide each data into the cluster center closest to it. Each time the iterative calculation is performed, the cluster center of the cluster will be updated once. When the calculation process meets a certain condition, the cost function is calculated To judge the quality of the driver style clustering model, the smaller the objective function, the better the clustering model.

[0079] It should be noted that, according to the distribution results of driving style characteristic parameters, Figure 3 As shown, category 1 driving style is a cautious personality, category 2 driving style is a normal personality, and category 3 driving style is an aggressive personality.

[0080] S1014: Analyze the clustering results using SPSS software to obtain driver style classification results.

[0081] Furthermore, the step S102 specifically includes the following steps:

[0082] S1021: Given an input driving data set, set a penalty function L(y, y^) to determine the error between the output prediction value and the actual value, and calculate the net input value z for each layer l and activation value a l , output to the last layer, and then the error term δ of each layer l Back propagation, propagating layer by layer to the initial layer, updates all parameters by calculating the partial derivatives of the parameters. The Adam optimizer is used to adjust the learning rate and correct the gradient estimate. The Adam algorithm calculates the square of the gradient g n 2 and the gradient g n The exponentially weighted average of . Its parameter update difference is:

[0083]

[0084] in and To obtain the corrected weighted average, the learning rate α is set to 0.001.

[0085] S1022: Select the convolution kernel size as 3 and the number as (16, 64, 128), select Relu as the activation function for acceleration, set the number of neurons in the fully connected layer to 5, and build a driver style recognition model based on the CNN convolutional neural network;

[0086] S1023: Use the categorical cross entropy penalty function and accuracy to evaluate and verify the effectiveness. i，k =1,y i，k =1, indicating that the prediction is correct. i，k and logp i，k If the value is negative, the prediction result is false, the penalty function will become larger, and the penalty function is continuously adjusted. The Adam optimizer optimizes the gradient descent algorithm to ensure that the penalty function is minimized. The formula for the cross entropy penalty function is:

[0087]

[0088] Among them, N is the total amount of data, K is the number of target tasks, and y i，k is the value of a sample data in the Kth category, p i，k Represents the possible probability of predicting it as the Kth class.

[0089] The accuracy formula is:

[0090]

[0091]

[0092] Furthermore, the accuracy changes of the training data set and the validation data set are shown in the figure below. Figure 4 As shown, it can be found that the more iterations there are, the higher the training accuracy and verification accuracy are above 90%, which proves the effectiveness of the driver style recognition model.

[0093] The step S103 specifically includes the following steps:

[0094] S1031: Inputting the driver's driving data into a driver style recognition model to obtain the current driver's style;

[0095] It should be noted that the preset corresponding relationship between the brake pedal displacement and the required braking force corresponding to different driving style results is pre-established. When the car brakes, the brake pedal pushes the displacement change of the master cylinder push rod, and the master cylinder push rod pushes the change of the hydraulic chamber piston, resulting in changes in the master cylinder pressure and the front and rear wheel cylinder pressure (i.e., the "pV characteristics" of the wheel cylinder). The front and rear wheel cylinder pressure changes directly lead to changes in the braking force. Therefore, the embodiment of the present invention establishes a corresponding relationship between the brake pedal displacement and the total braking force required by the driver, such as Figure 5 As shown, the actual required pressure needs to be calibrated according to the actual vehicle.

[0096] S1032: Analyze the driver's braking intention according to the identified driver style type and obtain the target braking force. Meanwhile, the vehicle controller determines whether the vehicle meets the condition requirements for entering the braking energy recovery mode according to the received battery status signal, vehicle speed signal, and braking intensity requirement signal;

[0097] It should be noted that if the battery SOC signal is greater than the maximum state of charge SOC that allows energy recovery max , or the vehicle speed signal is less than the minimum speed allowed for energy recovery, or the braking intensity requirement signal is greater than the maximum braking intensity allowed for the vehicle, it is determined that the vehicle cannot enter the regenerative braking mode, and the regenerative braking force F m =0, the braking demand is completely provided by the front and rear axle friction braking force according to the ideal front and rear axle braking force distribution curve I curve.

[0098] S1033: When the vehicle meets the requirements for entering the braking energy recovery mode, the maximum energy recovery control strategy is adopted while meeting the I curve and ECE regulations, and the front and rear axle braking forces and mechanical and friction braking forces are distributed, such as Figure 6 As shown, the specific braking force distribution strategy is as follows:

[0099] When 0<z≤0.2 (the intersection of the lower boundary line of ECE and the horizontal axis), the total braking force is completely provided by the electric braking force of the front axle:

[0100]

[0101] When 0.2<z≤0.35 (the intersection of the ECE lower boundary line and the f line corresponding to the braking intensity z=0.5), the total braking force is distributed according to the ECE lower boundary line:

[0102]

[0103] When 0.35<z≤0.5, the total braking force is distributed along line f:

[0104]

[0105] When z>0.5, the total braking force is distributed along the I curve:

[0106]

[0107] where F br is the rear axle braking force, F bf is the front axle braking force, m is the vehicle mass, g is the gravity coefficient, h g is the height of the center of mass, b is the distance from the center of mass to the rear axle, L is the distance between the front and rear axles, l r is the distance from the front axle to the center of mass, z is the braking strength, and μ is the friction coefficient.

[0108] It should be noted that when distributing the braking force, the present invention mainly considers the following constraints:

[0109] 1) To prevent dangerous skidding due to wheel locking, the actual front and rear braking force distribution lines of the vehicle should always be below the ideal braking force distribution curve (I curve);

[0110] 2) To maximize braking energy recovery, the embodiments of the present invention are designed and controlled to distribute as much braking force as possible to the front wheels while meeting the I curve and ECE regulations, that is, the total braking force distribution will be along the lower boundary curve of the ECE regulations.

[0111] S1034: Control the motor to perform braking energy recovery.

[0112] It should be noted that when the maximum motor force F m_max Greater than the braking force F required on the front axle bf , then the front axle braking force is entirely driven by the electric motor F m Provided, at this time the motor force F m =F bf If the motor power F m The maximum value is less than the required braking force F on the front axle bf , then the motor driving force F m The total braking force of the front axle cannot be met, and additional hydraulic braking force F is required hf Compensate for the remaining required braking force, at this time the motor braking force F m =F m_max , additional hydraulic braking force F hf =F bf -F m_max .

[0113] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0114] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0115] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A pure electric vehicle braking energy recovery method based on driver style recognition, characterized in that: include: S101: classifying the driver style by preprocessing the NGSIM data set and selecting and optimizing the driver style feature parameters; S102: Using a convolutional neural network to train sample data and establish a driver style recognition model; S103: analyzing the driver's braking intention according to the driver's style type, distributing the braking force according to the proposed energy recovery strategy, and controlling the motor to perform energy recovery; Step S101 specifically includes: S1011: Remove abnormal data from the NGSIM data set, use the Savitzky-Golay filter algorithm to smooth and denoise the data stream, and use the max-min method for normalization; S1012: Select average vehicle speed , Maximum speed , minimum speed , speed standard deviation , average acceleration , maximum acceleration , minimum acceleration , acceleration standard deviation And the distance involved To characterize the driver's style and calculate and optimize its specific parameters, the Factor Analysis method is used to reduce the dimension of the characteristic parameters; S1013: Select the average vehicle speed based on the index that the sum of squares of the extracted load is greater than 75%. , Maximum speed , minimum speed , speed standard deviation For common factors, K-means+ clustering algorithm is used to cluster the common factors to minimize the loss function, and the results are analyzed; Step S101 specifically includes: S1021: Given an input driving data set, set a penalty function L(y, y^) to determine the error between the output prediction value and the actual value, and use the Adam algorithm to adjust its learning rate and correct the gradient estimate; S1022: Select the convolution kernel size as 3 and the number as (16, 64, 128), select Relu as the activation function for acceleration, set the number of neurons in the fully connected layer to 5, and build a driver style recognition model based on the CNN convolutional neural network; S1023: Use cross entropy form penalty function and accuracy to evaluate and verify effectiveness; In step S1021, the Adam algorithm calculates the square of the gradient and gradient The exponentially weighted average of , the parameter update difference is: in and is the corrected weighted average, and the learning rate Set to 0.

001.

2. The pure electric vehicle braking energy recovery method based on driver style recognition according to claim 1 is characterized in that: Step S103 specifically includes: S1031: Inputting the driver's driving data into a driver style recognition model to obtain the current driver's style; S1032: analyzing the driver's braking intention according to the identified driver style type, and obtaining a target braking force; S1033: While meeting the I curve and ECE regulations, the maximum energy recovery control strategy is adopted, and the front and rear axle braking forces and the mechanical and friction braking forces are distributed.

3. The method for recovering braking energy of a pure electric vehicle based on driver style recognition according to claim 2, characterized in that: When the motor force is at its maximum value F m_max Greater than the braking force F required on the front axle bf , then the front axle braking force is entirely driven by the electric motor F m Provided, at this time the motor force F m =F bf If the motor power F m The maximum value is less than the required braking force F on the front axle bf , then the motor driving force F m The total braking force of the front axle cannot be met, and additional hydraulic braking force F is required hf Compensate for the remaining required braking force, at this time the motor braking force F m =F m_max , additional hydraulic braking force F hf =F bf -F m_max .

Images