New energy automobile brake pressure control method based on dynamic environment perception

Abstract

According to the new energy automobile brake pressure control method based on dynamic environment perception, a multi-window environment perception dynamic time warping algorithm is constructed, so that the similarity of different driver brake sequences is analyzed more meticulously; according to the method, a pavement adhesion coefficient observer for optimizing variable structure Kalman filtering based on a deep reinforcement learning algorithm is provided to calculate a pavement adhesion coefficient; secondly, a slip rate dynamic adjustment factor is obtained through calculation based on a wheel dynamics model, a vehicle density correction coefficient is obtained through calculation based on vehicle density, a road adhesion correction coefficient is obtained based on a road adhesion correction parameter constraint expression, and influences of different external environments on the road adhesion coefficient can be fully considered; the brake pressure calculation is more reasonable, and the driving safety is improved. According to the method, classification results of different drivers are combined, a comprehensive self-adaptive brake pressure calculation model is constructed to cope with different types of drivers, and corresponding final brake pressure is obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of smart cars, and in particular to a brake pressure control method for new energy vehicles based on dynamic environment perception. Background Art

[0002] With the rapid development of the automotive industry, drivers are increasingly concerned about safety issues during driving, and braking operation is a crucial component of driving safety. Determining the driver's desired braking pressure intensity—that is, their braking intention—is key to improving braking efficiency and, consequently, driving safety.

[0003] The invention patent with publication number CN109204283A discloses an electronic hydraulic braking system based on motor-based braking intention acquisition, and proposes an innovative electronic hydraulic braking system that uses a dual-motor configuration to achieve braking intention acquisition and power supply, while introducing a dual main hydraulic pipeline design to enhance the safety and reliability of the system. The system omits traditional sensors, reducing costs, and uses intelligent control of the electromagnetic clutch to ensure that the braking function can still be maintained when any motor fails. However, this solution does not take into account the impact of the environment. In rainy and snowy environments, the road adhesion coefficient and traffic conditions will change. At this time, the braking intention acquisition under normal conditions may not meet the driver's expected braking strength.

[0004] When driving in different external environments, drivers' desired brake pressure levels vary. For example, these external environments impact drivers in several key ways: First, rainy and snowy conditions can reduce the road's adhesion coefficient, leading drivers to expect stronger braking force. Furthermore, rainy and snowy conditions can easily lead to traffic jams and higher vehicle density. For safety reasons, drivers tend to drive cautiously, potentially resulting in a stronger braking response. Therefore, developing a brake pressure control method for new energy vehicles based on dynamic environmental perception to improve driving safety is a technical challenge facing those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a new energy vehicle brake pressure control method based on dynamic environment perception to improve driving safety.

[0006] A new energy vehicle brake pressure control method based on dynamic environment perception, comprising: Step S1: Collecting data from vehicle sensors and preprocessing the data to obtain preprocessed data, which includes at least braking distance. Based on the dynamic time warping algorithm, a multi-modal weighted distance and environmental perception constraints are introduced to construct a multi-window environmental perception dynamic time warping algorithm. A distance matrix is ​​calculated, and the driver is classified by analyzing the similarity between the distance matrix and the braking distance. Step S2: performing an overall calculation of the braking force control model parameters based on the preprocessed data, wherein a road adhesion coefficient observer based on a variable structure Kalman filter is optimized based on a deep reinforcement learning algorithm to calculate the road adhesion coefficient, a slip rate dynamic adjustment factor is calculated based on a wheel dynamics model, a vehicle density correction factor is calculated based on vehicle density, and a road adhesion correction factor is calculated based on a road adhesion correction parameter constraint expression; Step S3: Based on the road adhesion coefficient, the dynamic adjustment factor for the slip rate, the vehicle density correction factor, and the road adhesion correction factor, the input gate, output gate, forget gate, and cell update rule of the LSTM prediction model are improved. The brake pressure is predicted using the improved LSTM prediction model. The predicted brake pressure value is used as a feedforward signal to calculate the dynamic weight coefficient for adjusting the basic braking force. Step S4: Based on the driver classification results in step S1, as well as the dynamic adjustment factor of the slip rate, the vehicle density correction coefficient, the road adhesion correction coefficient, and the dynamic weight coefficient of the basic braking force, a comprehensive adaptive brake pressure calculation model is constructed, and the final brake pressure for different types of drivers is calculated through the comprehensive adaptive brake pressure calculation model.

[0007] The brake pressure control method for new energy vehicles based on dynamic environment perception provided by the present invention has the following beneficial effects: 1. This paper introduces multimodal weighted distance and environmental perception constraints based on the traditional dynamic time warping (DTW) algorithm, constructing a new multi-window environmental perception dynamic time warping algorithm. This algorithm calculates the distance matrix more accurately than the traditional DTW algorithm, making the analysis of the similarity of different drivers' braking sequences more detailed, thereby enabling accurate driver classification.

[0008] 2. The present invention proposes a road adhesion coefficient observer based on a variable structure Kalman filter optimized by a deep reinforcement learning algorithm to calculate the road adhesion coefficient; secondly, a dynamic adjustment factor of the slip rate is calculated based on the wheel dynamics model, a vehicle density correction coefficient is calculated based on the vehicle density, and a road adhesion correction coefficient is obtained based on the road adhesion correction parameter constraint expression, thereby improving the input gate, output gate, forget gate and cell update rules of the traditional LSTM prediction model, and then predicting the brake pressure. It can fully consider the impact of different external environments on the road adhesion coefficient, making the brake pressure calculation more reasonable, solving the problem of insufficient braking efficiency of the traditional braking system due to the reduction of the adhesion coefficient in different environments, and improving driving safety.

[0009] 3. The present invention combines the classification results of different drivers to construct a comprehensive adaptive brake pressure calculation model to cope with different types of drivers and obtain the corresponding final brake pressure, realizing dynamic adaptation to the driver's personalized braking style, making braking operation safer and more reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A schematic flow chart of a new energy vehicle brake pressure control method based on dynamic environment perception provided by an embodiment of the present invention; Figure 2 The figure shows a comparison of the brake pressure curves of three different braking styles and the ideal brake pressure curve. DETAILED DESCRIPTION

[0011] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the embodiments of the present invention, and should not be construed as limiting the present invention.

[0012] See also Figure 1 An embodiment of the present invention provides a new energy vehicle brake pressure control method based on dynamic environment perception, including steps S1-S4.

[0013] Step S1: collect data from vehicle sensors and preprocess the data to obtain preprocessed data. The preprocessed data at least includes braking distance. Based on the dynamic time warping algorithm, multimodal weighted distance and environmental perception constraints are introduced to construct a multi-window environmental perception dynamic time warping algorithm, calculate the distance matrix, and classify the drivers by analyzing the similarity between the distance matrix and the braking distance.

[0014] Wherein, step S1 specifically includes: Step S101, collecting data during driving, including speed and acceleration, through vehicle sensors; Step S102, processing the noise points and abnormal data in the data collected in step S101, the preprocessing process includes: threshold filtering, sliding average filtering, Z-Score abnormality elimination, and multi-sensor calibration; Step S103, identifying braking-related data from the data pre-processed in step S102, including deceleration, braking distance, and braking time; In step S104, a multi-modal weighted distance and environmental perception constraints are introduced on the basis of the dynamic time warping algorithm to construct a multi-window environmental perception dynamic time warping algorithm. Compared with the single-dimensional Euclidean distance, the present invention introduces a multi-modal weighted distance to address the single defect of the traditional solution. The expression of the multi-modal weighted distance is:

[0015]

[0016] in, represents the weighted Euclidean distance, is the time-varying feature weight, The reference sequence Moment dimensional eigenvalues, For the test sequence Moment dimensional eigenvalues, is the total number of dimensions, is the characteristic historical variance, is the sum of all feature variances, is the time attenuation coefficient, is the time difference term.

[0017] The traditional method has no constraints on path search. The present invention introduces environment perception constraints. The expression of environment perception constraints is:

[0018] in, is the environmental constraint, is the weight coefficient of road surface difference (used to adjust the contribution ratio of road surface adhesion coefficient difference to the total constraint term), The reference sequence is The road adhesion coefficient at the moment, For the test sequence The road adhesion coefficient at the moment, is the reference value of the standard road adhesion coefficient, is the traffic flow-vehicle speed coupling weight coefficient (used to adjust the contribution ratio of traffic flow density and speed difference to the total constraint term), The reference sequence is The speed of the car at the moment, For the test sequence The speed of the car at the moment, is the maximum effective speed, is the vehicle density; The constructed multi-window environment-aware dynamic time warping algorithm satisfies the following formula:

[0019] in, For the improved dynamic time warping distance, is the normalization factor, is a non-empty space-time set, is the environmental compensation weight, is a multimodal feature sequence, , is the brake pressure signal, is the steering wheel angle signal, is the vehicle acceleration signal, is the pedal displacement signal, It is a four-dimensional driving behavior feature; Step S105: Calculate the distance matrix based on the multi-window environment perception dynamic time warping algorithm ; Step S106 , by analyzing the similarity between the distance matrix and the braking distance, the drivers are divided into three categories: emergency braking drivers, smooth braking drivers, and early braking drivers.

[0020] Table 1 lists the dynamic time warping distance ranges of the three types of drivers in major scenarios. It should be noted that the dynamic time warping distance is dimensionless.

[0021] Table 1

[0022] Step S2: performing an overall calculation of the braking force control model parameters based on the preprocessed data, wherein a road adhesion coefficient observer based on a variable structure Kalman filter is optimized using a deep reinforcement learning algorithm to calculate the road adhesion coefficient, a slip rate dynamic adjustment factor is calculated based on a wheel dynamics model, a vehicle density correction coefficient is calculated based on vehicle density, and a road adhesion correction coefficient is obtained based on a road adhesion correction parameter constraint expression.

[0023] The steps for constructing a road adhesion coefficient observer using a variable structure Kalman filter are as follows: Based on the vehicle dynamics model, the forward Euler method is used to discretize the variable structure Kalman state space equation, which is expressed as:

[0024] in, is the state vector The first derivative of , For vertical direction, is the lateral velocity, is the yaw angular velocity, is the current road adhesion coefficient, represents transpose, is the state transition function, is the control input, is the process noise; The observation equation is:

[0025] in, is the observation vector, is the observation function, is the observation noise; Design a variable structure Kalman filter, where the time update is:

[0026]

[0027] in, for The prior state estimate at time , for The posterior state estimate at time , for The control input at the moment, for The prior error covariance matrix at time , for The Jacobian matrix of the state transfer matrix at the moment, for The posterior error covariance matrix at time , for The process noise covariance matrix at time t; The measurement is updated as follows:

[0028] in, for The Kalman gain matrix at time , for The Jacobian matrix of the moment observation matrix, for The time-adaptive observation noise covariance matrix, Update via:

[0029] in, For the forgetting factor, for The time-adaptive observation noise covariance matrix, for The sequence of new information at each moment, , for The actual observed value at time, is the predicted observation value; Then based on and get The posterior estimate of the moment , the expression is:

[0030] Based on and get The posterior error covariance matrix at time , the expression is:

[0031] in, is the identity matrix; Then, deep reinforcement learning is used to optimize the road adhesion coefficient observer of the variable structure Kalman filter, where the state space of the deep reinforcement learning algorithm is for:

[0032] in, for The estimated road adhesion coefficient at time , The change in the estimated value of the road adhesion coefficient, , for The estimated road adhesion coefficient at time , is the norm of the innovation sequence, is the trace of the estimated error covariance matrix; Action Space of Deep Reinforcement Learning Algorithms for:

[0033] in, is the adjustment amount of the process noise covariance matrix, is the adjustment of the observation noise covariance matrix, Adjust the parameters for the adaptive factor; Reward Functions for Deep Reinforcement Learning Algorithms for:

[0034] in, and is the weight coefficient, is the estimated value of the current road adhesion coefficient; The TD3 algorithm is used to train the actor-critic network of the deep reinforcement learning algorithm. The network update strategy is:

[0035] in, is the objective function gradient, is the expectation operator, is the gradient of the action-value function with respect to the action, is the environmental state, For action changes, To fit the action-value function; is the gradient of the update policy with respect to the parameters, are the parameters of the actor network; The parameters are dynamically adjusted to:

[0036]

[0037] in, for The process noise covariance matrix at time , is the initial process noise covariance matrix, is the initial adaptive observation noise covariance matrix, is a symbolic function, is the change in road adhesion coefficient; In specific implementation, the TD3 algorithm is optimized and The feedback is then passed back to the variable structure Kalman filter.

[0038] Finally, the road adhesion coefficient is calculated using the following formula:

[0039] in, for The posterior estimate of the road adhesion coefficient at time , for The predicted value of the road adhesion coefficient at time , yes The 4th row element in .

[0040] In step S2, the slip rate dynamic adjustment factor is calculated based on the wheel dynamics model, and the expression is:

[0041]

[0042] in, is the slip rate dynamic adjustment factor, is the attenuation coefficient, Indicates taking the maximum value, is the current slip rate, is the safety slip rate threshold, is the wheel angular velocity, is the tire radius.

[0043] In step S2, the vehicle density correction coefficient is calculated based on the vehicle density, and the expression is:

[0044] in, is the vehicle density correction factor, In order to obtain the number of surrounding vehicles through body cameras and radars, is the vehicle baseline density.

[0045] In step S2, the road adhesion correction coefficient is obtained based on the road adhesion correction parameter constraint expression, and the expression is:

[0046]

[0047] in, is the road adhesion correction coefficient, Indicates taking the minimum value, The original calculated value of the ground attachment correction parameter.

[0048] In step S3, based on the road adhesion coefficient, the dynamic adjustment factor of the slip rate, the vehicle density correction coefficient, and the road adhesion correction coefficient, the input gate, output gate, forget gate, and cell update rule of the LSTM prediction model are improved. The brake pressure is predicted using the improved LSTM prediction model, and the predicted brake pressure value is used as a feedforward signal to calculate the dynamic weight coefficient for adjusting the basic braking force.

[0049] Among them, the input gate, output gate, forget gate and cell update rules of the LSTM prediction model are improved. The improved input gate function is:

[0050]

[0051]

[0052]

[0053] in, for The input gate activation value at time , for The cell state at any moment, is the sigmoid activation function, is the normalization function, and are the weight matrix and bias matrix of the input gate, yes The hidden state of the moment; yes The features after time-space fusion are comprehensive features generated through attention mechanism or multimodal fusion, which have noise reduction and cross-modal correlation; and are the weight matrix and bias matrix of the memory unit, is the hyperbolic tangent function, For the The attention weights of the input channels, For the The sensors in Input at any moment (such as pedal displacement, wheel cylinder pressure), is the total number of sensors with input features, is the weight matrix of learnable attention weights; The improved output gate function is:

[0054]

[0055] in, for The output gate activation value at time , is the normalization operation, and are the weight matrix and bias matrix of the output gate respectively; for The original input features at each moment come directly from the instantaneous signal of the sensor (such as pedal displacement and wheel cylinder pressure), retaining the original high-frequency characteristics of the signal; for The final hidden state output at time t; The improved forget gate function is:

[0056]

[0057] in, for The output value of the forget gate at time , is the weight matrix of the forget gate, is the fixed bias of the forget gate, To offset the environmental impact, is the weight coefficient of environmental impact bias; The improved cell renewal rule satisfies the following formula:

[0058]

[0059] in, for The cell state at any moment, for The cell state at any moment, is the output of the input gate, To dynamically adjust the forgetting strength parameter, is the candidate cell state, Dynamically adjust the forgetting strength weight parameter; The improved LSTM prediction model is used to predict the displacement of the brake master cylinder and wheel cylinder pressure As input, the brake pressure is predicted to obtain the brake pressure prediction value ; The brake pressure prediction value As a feedforward signal, calculate the dynamic weight coefficient for adjusting the basic braking force , the expression is:

[0060] in, and is a learnable parameter, To predict the pressure change rate.

[0061] The mediation logic is: when the predicted pressure is significantly higher than the current actual brake pressure, it indicates that the To match the urgent needs; otherwise, reduce To respond smoothly.

[0062] Step S4: Based on the driver classification results in step S1, as well as the dynamic adjustment factor of the slip rate, the vehicle density correction coefficient, the road adhesion correction coefficient, and the dynamic weight coefficient of the basic braking force, a comprehensive adaptive brake pressure calculation model is constructed, and the final brake pressure for different types of drivers is calculated through the comprehensive adaptive brake pressure calculation model.

[0063] Among them, the expression of the comprehensive adaptive brake pressure calculation model is:

[0064] in, is the final brake pressure, is the basic brake pressure; is the driver type gain coefficient, According to the dynamic value of the driving style classification result in step S1, the emergency braking type, as the main target of dynamic correction, should be given a higher weight, such as 1.2. The smooth braking type has a lower weight, such as 0.9, to maintain linearity and reduce interference in dynamic correction. The early braking type is between the emergency braking type and the smooth braking type, such as 1.05, to reflect moderate compensation. is the tire longitudinal force, is the density correction gain coefficient, when hour, ,when hour, .

[0065] Specifically, based on the driver classification result obtained in step S1, the adaptive mechanism is: Emergency braking drivers have an emergency braking style, which is dominated by dynamic correction. The core variables are and , dynamically increases according to the urgency of the dynamic time warping algorithm based on multi-window environment perception , strengthening the deceleration requirements. Follow The reduction of the pressure increases the risk of tire locking.

[0066] The smooth braking style of the driver is dominated by the basic braking force. Increase, environmental compensation decreases, the weight of the traditional braking model increases, the braking force is kept stable, in the high road adhesion coefficient area, the road adhesion correction coefficient Increase, Approaching 0, suppressing the interference of dynamic correction amount.

[0067] Early braking style, environmental compensation and prediction fusion, congested road conditions Improve, compensate for the uncertainty when following the car, in the low road adhesion coefficient area, the road adhesion correction coefficient Reduce, Increased, enhanced dynamic compensation of tire forces.

[0068] The present invention is simulated and tested below. The test results are as follows: Figure 2 shown.

[0069] Figure 2 The dynamic changes of brake pressure over time under different braking types in dynamic environments are shown, and two typical scenarios, rain and snow low adhesion area and high density traffic area, are marked. Figure 2 As can be clearly seen, the proposed method closely tracks the ideal braking curve during vehicle braking in rainy, snowy conditions and fluctuating traffic density, enabling precise adaptation of optimal braking pressure. Furthermore, different driving styles influence the braking force performance: in both the single rainy, snowy, and high-density traffic scenarios, the braking force of drivers who brake early is smaller than the ideal braking force curve, reflecting their relatively conservative driving style; the braking force of drivers who brake gently is largely consistent with the ideal braking force curve, reflecting a more common and stable driving style; and the braking force of drivers who brake hard is greater than the ideal braking force curve, indicating a more aggressive driving style. In areas where rainy, snowy, and high-density traffic overlap, it is particularly clear that all three types of drivers maintain their unique driving styles. Furthermore, because the proposed method comprehensively considers factors such as road adhesion coefficient and traffic density across multiple dimensions, the braking curves under different conditions provide overall reasonable braking pressure control compared to the ideal braking force curve, further enhancing braking safety and stability.

[0070] In summary, the new energy vehicle brake pressure control method based on dynamic environment perception according to the above embodiment has the following beneficial effects: 1. This paper introduces multimodal weighted distance and environmental perception constraints based on the traditional dynamic time warping (DTW) algorithm, constructing a new multi-window environmental perception dynamic time warping algorithm. This algorithm calculates the distance matrix more accurately than the traditional DTW algorithm, making the analysis of the similarity of different drivers' braking sequences more detailed, thereby enabling accurate driver classification.

[0071] 2. The present invention proposes a road adhesion coefficient observer based on a variable structure Kalman filter optimized by a deep reinforcement learning algorithm to calculate the road adhesion coefficient; secondly, a dynamic adjustment factor of the slip rate is calculated based on the wheel dynamics model, a vehicle density correction coefficient is calculated based on the vehicle density, and a road adhesion correction coefficient is obtained based on the road adhesion correction parameter constraint expression, thereby improving the input gate, output gate, forget gate and cell update rules of the traditional LSTM prediction model, and then predicting the brake pressure. It can fully consider the impact of different external environments on the road adhesion coefficient, making the brake pressure calculation more reasonable, solving the problem of insufficient braking efficiency of the traditional braking system due to the reduction of the adhesion coefficient in different environments, and improving driving safety.

[0072] 3. The present invention combines the classification results of different drivers to construct a comprehensive adaptive brake pressure calculation model to cope with different types of drivers and obtain the corresponding final brake pressure, realizing dynamic adaptation to the driver's personalized braking style, making braking operation safer and more reliable.

[0073] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be determined by the appended claims.

Claims

1. A new energy vehicle brake pressure control method based on dynamic environment perception, characterized in that: include: Step S1: Collecting data from vehicle sensors and preprocessing the data to obtain preprocessed data, which includes at least braking distance. Based on the dynamic time warping algorithm, a multi-modal weighted distance and environmental perception constraints are introduced to construct a multi-window environmental perception dynamic time warping algorithm. A distance matrix is ​​calculated, and the driver is classified by analyzing the similarity between the distance matrix and the braking distance. Step S2: performing an overall calculation of the braking force control model parameters based on the preprocessed data, wherein a road adhesion coefficient observer based on a variable structure Kalman filter is optimized based on a deep reinforcement learning algorithm to calculate the road adhesion coefficient, a slip rate dynamic adjustment factor is calculated based on a wheel dynamics model, a vehicle density correction factor is calculated based on vehicle density, and a road adhesion correction factor is calculated based on a road adhesion correction parameter constraint expression; Step S3: Based on the road adhesion coefficient, the dynamic adjustment factor for the slip rate, the vehicle density correction factor, and the road adhesion correction factor, the input gate, output gate, forget gate, and cell update rule of the LSTM prediction model are improved. The brake pressure is predicted using the improved LSTM prediction model. The predicted brake pressure value is used as a feedforward signal to calculate the dynamic weight coefficient for adjusting the basic braking force. Step S4: Based on the driver classification results in step S1, as well as the dynamic adjustment factor of the slip rate, the vehicle density correction coefficient, the road adhesion correction coefficient, and the dynamic weight coefficient of the basic braking force, a comprehensive adaptive brake pressure calculation model is constructed, and the final brake pressure for different types of drivers is calculated through the comprehensive adaptive brake pressure calculation model.

2. The new energy vehicle brake pressure control method based on dynamic environment perception according to claim 1 is characterized in that: Step S1 specifically includes: Step S101, collecting data during driving, including speed and acceleration, through vehicle sensors; Step S102, processing the noise points and abnormal data in the data collected in step S101, the preprocessing process includes: threshold filtering, sliding average filtering, Z-Score abnormality elimination, and multi-sensor calibration; Step S103, identifying braking-related data from the data pre-processed in step S102, including deceleration, braking distance, and braking time; In step S104, a multi-modal weighted distance and an environment-aware constraint are introduced on the basis of the dynamic time warping algorithm to construct a multi-window environment-aware dynamic time warping algorithm. The expression of the multi-modal weighted distance is: in, represents the weighted Euclidean distance, is the time-varying feature weight, The reference sequence Moment dimensional eigenvalues, For the test sequence Moment dimensional eigenvalues, is the total number of dimensions, is the characteristic historical variance, is the sum of all feature variances, is the time attenuation coefficient, is the time difference term; The expression of the environment perception constraint is: in, is the environmental constraint, is the weight coefficient of road surface difference, The reference sequence is The road adhesion coefficient at the moment, For the test sequence The road adhesion coefficient at the moment, is the reference value of the standard road adhesion coefficient, is the traffic flow-vehicle speed coupling weight coefficient, The reference sequence is The speed of the car at the moment, For the test sequence The speed of the car at the moment, is the maximum effective speed, is the vehicle density; The constructed multi-window environment-aware dynamic time warping algorithm satisfies the following formula: in, For the improved dynamic time warping distance, is the normalization factor, is a non-empty space-time set, is the environmental compensation weight; Step S105: Calculate the distance matrix based on the multi-window environment perception dynamic time warping algorithm. The expression is: in, is the distance matrix The value in row 1 and column 1, is the distance matrix The first row, The value of the column, is the distance matrix Middle The value of row, column 1, is the distance matrix Middle Row, No. The value of the column; Step S106 , by analyzing the similarity between the distance matrix and the braking distance, the drivers are divided into three categories: emergency braking drivers, smooth braking drivers, and early braking drivers.

3. The new energy vehicle brake pressure control method based on dynamic environment perception according to claim 2 is characterized in that: In step S2, the steps of constructing a road adhesion coefficient observer of a variable structure Kalman filter are as follows: Based on the vehicle dynamics model, the forward Euler method is used to discretize the variable structure Kalman state space equation, which is expressed as: in, is the state vector The first derivative of , For vertical direction, is the lateral velocity, is the yaw angular velocity, is the current road adhesion coefficient, represents transpose, is the state transition function, is the control input, is the process noise; The observation equation is: in, is the observation vector, is the observation function, is the observation noise; Design a variable structure Kalman filter, where the time update is: in, for The prior state estimate at time , for The posterior state estimate at time , for The control input at the moment, for The prior error covariance matrix at time , for The Jacobian matrix of the state transfer matrix at the moment, for The posterior error covariance matrix at time , for The process noise covariance matrix at time t; The measurement is updated as follows: in, for The Kalman gain matrix at time , for The Jacobian matrix of the moment observation matrix, for The time-adaptive observation noise covariance matrix, Update via: in, For the forgetting factor, for The time-adaptive observation noise covariance matrix, for The sequence of new information at each moment, , for The actual observed value at time, is the predicted observation value; Then based on and get The posterior estimate of the moment , the expression is: Based on and get The posterior error covariance matrix at time , the expression is: in, is the identity matrix; Then, deep reinforcement learning is used to optimize the road adhesion coefficient observer of the variable structure Kalman filter, where the state space of the deep reinforcement learning algorithm is for: in, for The estimated road adhesion coefficient at time , The change in the estimated value of the road adhesion coefficient, , for The estimated road adhesion coefficient at time , is the norm of the innovation sequence, is the trace of the estimated error covariance matrix; Action Space of Deep Reinforcement Learning Algorithms for: in, is the adjustment amount of the process noise covariance matrix, is the adjustment of the observation noise covariance matrix, Adjust the parameters for the adaptive factor; Reward Functions for Deep Reinforcement Learning Algorithms for: in, and is the weight coefficient, is the estimated value of the current road adhesion coefficient; The TD3 algorithm is used to train the actor-critic network of the deep reinforcement learning algorithm. The network update strategy is: in, is the objective function gradient, is the expectation operator, is the gradient of the action-value function with respect to the action, is the environmental state, For action changes, To fit the action-value function; is the gradient of the update policy with respect to the parameters, are the parameters of the actor network; The parameters are dynamically adjusted to: in, for The process noise covariance matrix at time , is the initial process noise covariance matrix, is the initial adaptive observation noise covariance matrix, is a symbolic function, is the change in road adhesion coefficient; Finally, the road adhesion coefficient is calculated using the following formula: in, for The posterior estimate of the road adhesion coefficient at time , for The predicted value of the road adhesion coefficient at time , yes The 4th row element in .

4. The new energy vehicle brake pressure control method based on dynamic environment perception according to claim 3 is characterized in that: In step S2, the slip rate dynamic adjustment factor is calculated based on the wheel dynamics model, and the expression is: in, is the slip rate dynamic adjustment factor, is the attenuation coefficient, Indicates taking the maximum value, is the current slip rate, is the safety slip rate threshold, is the wheel angular velocity, is the tire radius.

5. The new energy vehicle brake pressure control method based on dynamic environment perception according to claim 4 is characterized in that: In step S2, the vehicle density correction coefficient is calculated based on the vehicle density, and the expression is: in, is the vehicle density correction factor, In order to obtain the number of surrounding vehicles through body cameras and radars, is the vehicle baseline density.

6. The new energy vehicle brake pressure control method based on dynamic environment perception according to claim 5 is characterized in that: In step S2, the road adhesion correction coefficient is obtained based on the road adhesion correction parameter constraint expression, and the expression is: in, is the road adhesion correction coefficient, Indicates taking the minimum value, The original calculated value of the ground attachment correction parameter.

7. The new energy vehicle brake pressure control method based on dynamic environment perception according to claim 6 is characterized in that: In step S3, the input gate, output gate, forget gate and cell update rule of the LSTM prediction model are improved. The improved input gate function is: in, for The input gate activation value at time , for The cell state at any moment, is the sigmoid activation function, is the normalization function, and are the weight matrix and bias matrix of the input gate, yes The hidden state of the moment, yes The characteristics after time and space fusion, and are the weight matrix and bias matrix of the memory unit, is the hyperbolic tangent function, For the The attention weights of the input channels, For the The sensors in Input at the moment, is the total number of sensors with input features, is the weight matrix of learnable attention weights; The improved output gate function is: in, for The output gate activation value at time , is the normalization operation, and are the weight matrix and bias matrix of the output gate respectively, for The original input features at time t, for The final hidden state output at time t; The improved forget gate function is: in, for The output value of the forget gate at time , is the weight matrix of the forget gate, is the fixed bias of the forget gate, To offset the environmental impact, is the weight coefficient of environmental impact bias; The improved cell renewal rule satisfies the following formula: in, for The cell state at any moment, for The cell state at any moment, is the output of the input gate, To dynamically adjust the forgetting strength parameter, is the candidate cell state, Dynamically adjust the forgetting strength weight parameter; The brake pressure is predicted by the improved LSTM prediction model to obtain the brake pressure prediction value ; The brake pressure prediction value As a feedforward signal, calculate the dynamic weight coefficient for adjusting the basic braking force , the expression is: in, and is a learnable parameter, To predict the pressure change rate.

8. The method for controlling brake pressure of a new energy vehicle based on dynamic environment perception according to claim 7 is characterized in that: In step S4, the expression of the comprehensive adaptive brake pressure calculation model is: in, is the final brake pressure, is the basic brake pressure, is the driver type gain coefficient, is the tire longitudinal force, is the density correction gain coefficient, when hour, ,when hour, .

Images