End-to-end automobile chassis integrated control method

Abstract

The invention discloses an end-to-end automobile chassis integrated control method, which comprises the following steps of: 1, collecting and analyzing vehicle sensor signals in a real driving scene so as to construct a chassis state training database with labels; 2, establishing a chassis control training instruction library based on a chassis state training database; 3, constructing an operation evaluation network, and training to obtain an operation evaluation model; 4, constructing a chassis integrated control network, and training to obtain a chassis integrated control model; 5, constructing a chassis domain controller based on the joint learning network; and 6, collecting sensor data through the chassis domain controller to supplement the chassis state training database. According to the invention, the control instruction can be directly output based on the information of the vehicle chassis sensor, so that the stability control of the vehicle in a complex and changeable driving environment can be realized.

Description

Technical Field

[0001] The present invention belongs to the field of vehicle dynamics and stability control under the background of automobile intelligence, and specifically is an end-to-end automobile chassis integrated control method. Background Art

[0002] With the development of automobile intelligence, the types of chassis subsystems are increasing, and chassis control functions are becoming increasingly complex. Faced with complex and uncertain working conditions, the existing chassis integrated control methods are difficult to effectively coordinate these subsystems, and have limitations such as low integration and insufficient coordinated control. For example, although the integrated suspension system and steering system can improve the lateral stability of the vehicle, and the integrated steering system with the braking system or power system can help enhance the yaw stability, these methods still have difficulty in fully exerting the synergistic effect when faced with complex working conditions. Existing control algorithms face the problem of poor adaptability to vehicle state changes and environmental changes, especially under extreme road conditions and variable driving conditions, the control effect may be significantly affected, and it is difficult to effectively achieve multi-objective comprehensive optimization. In addition, with the continuous expansion of the functions of vehicle chassis control systems, control algorithms are becoming increasingly complex, especially after the introduction of advanced methods such as model predictive control (MPC) and adaptive control, the system's computing requirements have increased dramatically. This complex real-time calculation may lead to slow system response, especially under high-speed driving or complex driving conditions, the vehicle must respond quickly to factors such as road changes and driving behavior. Otherwise, problems such as control lag and instability may occur, affecting the safety and stability of the vehicle.

[0003] Therefore, the system integration capability and real-time response capability of existing control methods need to be improved urgently to adapt to more complex and changeable driving environments. Summary of the invention

[0004] The present invention aims to address the deficiencies of the above-mentioned prior art and propose an end-to-end automobile chassis integrated control method, so as to directly output control instructions based on the vehicle chassis sensor information, greatly improve the chassis integration coordination efficiency, and thus achieve vehicle stability control in complex and changeable driving environments.

[0005] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme:

[0006] The end-to-end automobile chassis integrated control method of the present invention is characterized in that it is carried out according to the following steps:

[0007] Step 1: Collect and analyze vehicle sensor signals in real driving scenarios to build a labeled chassis status training database ,in, Indicates the transition of the vehicle from an unstable state to a stable state in a real driving scenario. chassis status dataset, and , express Middle Chassis status time series data set within a time period, and , express Middle The chassis state parameters at each time step are: express The total number of time steps in , and the interval between time steps is ,and ,in, express Middle The front wheel turning angle of the vehicle at each time step; express Middle The wheel speed at each time step is express Middle Suspension parameters at time steps; express Middle The braking force of the wheel in each time step

[0008] make The label is recorded as ;and ,in, Indicates efficient operation. Indicates inefficient operation;

[0009] Step 2: Training database based on chassis status , establish chassis control command training library , Indicates Vehicle chassis status dataset The corresponding chassis control instruction set, and , express Middle The chassis control instructions for a period of time, that is, the chassis state time series data set is The chassis control command at the time, and ,in, express Middle Front wheel angular acceleration in a time period; express Middle The rate of change of wheel speed in a time period, and , express Middle The rate of change of the left front wheel speed in a time period is represents the rate of change of the right front wheel speed, express Middle The rate of change of the left rear wheel speed in a time period is express Middle The rate of change of the right rear wheel speed within a time period; express Middle The change rate of the suspension parameters within a time period, and ,in, express Middle The change rate of the stiffness coefficient of the left front suspension in a time period, express Middle The change rate of the suspension stiffness coefficient within a time period, express Middle The change rate of the suspension stiffness coefficient within a time period, express Middle The change rate of the suspension stiffness coefficient within a time period, express Middle The change rate of the damping coefficient of the left front suspension in a time period, express Middle The change rate of the damping coefficient of the right front suspension within a time period, express Middle The change rate of the damping coefficient of the left rear suspension in a time period, express Middle The change rate of the damping coefficient of the right rear suspension within a time period; express Middle The rate of change of the wheel braking force in a time period, and , express Middle The rate of change of the left front wheel braking force in a time period is express Middle The rate of change of the right front wheel braking force in a time period is express Middle The rate of change of the left rear wheel braking force in a time period is express Middle The rate of change of the right rear wheel braking force within a time period;

[0010] Step 3: Construct an operation evaluation network based on spatiotemporal feature sequences and evaluate any Process and get the corresponding prediction label , thus constructing the loss function , used to train the operation evaluation network to obtain the optimal operation evaluation model;

[0011] Step 4: Build a chassis integrated control network and use it to control the randomly selected Process and get the corresponding control instructions , thus constructing the loss function , used to train the chassis integrated control network and obtain the optimal chassis integrated control model;

[0012] Step 5: Build a chassis domain controller based on the joint learning network, including: the trained optimal operation evaluation model, the trained optimal chassis integrated control model and the chassis state scoring module, and perform chassis state time series data at the current time t. Processing, the current operation evaluation result , control instructions and cumulative score ;

[0013] Step 6, when and When the chassis domain controller executes the chassis control command, ; Otherwise, the chassis domain controller does not intervene in chassis control and is in silent state;

[0014] when When the chassis domain controller uploads the sensor data of the vehicle during the transition from an unstable state to a stable state to the cloud to supplement the chassis state training database, it returns to step 1 and executes the better operation evaluation model and the better chassis integrated control model further trained and updated in the chassis domain controller;

[0015] Step 7, after t+1 is assigned to t, return to step 5.1 and execute sequentially.

[0016] The end-to-end automobile chassis integrated control method of the present invention is also characterized in that step 3 is performed according to the following steps:

[0017] Step 3.1, the operation evaluation network includes: a gated recurrent unit layer, a fully connected layer, and an activation function layer;

[0018] The gated recurrent unit layer is Process and output Middle Hidden state in time period ;

[0019] The fully connected layer is based on , output Middle The raw score set for the time period ;in, represents the weight matrix of the fully connected layer, represents the bias term of the fully connected layer, and , express Middle The raw score of efficient operations in a time period; express Middle The raw score of inefficient operations in a time period, express Middle The raw score of transition operations in the time period;

[0020] The activation function layer Perform nonlinear transformation to obtain Middle Predicted labels for time periods ,in, express The original score of the i-th type The predicted probability after nonlinear transformation, and , They represent efficient operation, inefficient operation and transition operation respectively;

[0021] Step 3.2: Use formula (1) to construct the loss function of the operation evaluation network ;

[0022] (1)

[0023] In formula (1), represents the weighting coefficient;

[0024] Step 3.3: Use formula (2) to construct the loss function The gradient update formula is:

[0025] (2)

[0026] Step 3.4: Use equations (3) and (4) to get a new weight matrix and the new bias term :

[0027] (3)

[0028] (4)

[0029] In formula (3) and formula (4), represents the learning rate;

[0030] Step 3.5, use formula (5) to construct the parameter update formula of the gated recurrent unit layer:

[0031] (5)

[0032] In formula (5), represents the updated parameters of the gated recurrent unit layer, represents the parameters of the gated recurrent unit layer before updating, represents the learning rate;

[0033] Step 3.6: Based on equations (3), (4) and (5), the operation evaluation network is trained by gradient descent method to obtain the optimal operation evaluation model after training.

[0034] Further, the step 4 is performed according to the following steps:

[0035] Step 4.1, the chassis integrated control network is composed of a gated recurrent unit layer and a deep reinforcement learning layer;

[0036] The gated recurrent unit layer is Chassis status time series data set for the nth time period and Chassis status time series data set for the n+1th time period Process and output The corresponding low-dimensional chassis state vector and The corresponding low-dimensional chassis state vector ;

[0037] The deep reinforcement learning layer receives and , and use the DDPG algorithm to Process and generate Corresponding predicted chassis control commands ,Right now Indicates chassis integrated control network based on Output the predicted change rate of chassis parameters in the next time period, express The predicted rate of change of the front wheel angle in the n+1th time period, that is, S, express The predicted rate of change of wheel speed in the n+1th time period, express The predicted change rate of the suspension parameters in the n+1th time period, express The predicted rate of change of the wheel braking force in the n+1th time period, and ; express The predicted rate of change of the left front wheel speed in the n+1th time period, express The predicted rate of change of the right front wheel speed in the n+1th time period, express The predicted rate of change of the left rear wheel speed in the n+1th time period, express The predicted rate of change of the right rear wheel speed in the n+1th time period; express Middle The predicted change rate of the suspension parameters within a time period, , express The predicted change rate of the left front suspension stiffness coefficient in the n+1th time period, express The predicted change rate of the right front suspension stiffness coefficient in the n+1th time period, express The predicted change rate of the left rear suspension stiffness coefficient in the n+1th time period, express The predicted change rate of the right rear suspension stiffness coefficient in the n+1th time period, express The predicted change rate of the left front suspension damping coefficient in the n+1th time period, express The predicted change rate of the right front suspension damping coefficient in the n+1th time period, express The predicted change rate of the left rear suspension damping coefficient in the n+1th time period, express The predicted change rate of the right rear suspension damping coefficient in the n+1th time period; express Middle The predicted rate of change of the braking force of the wheel within a time period, and , express The predicted change rate of the left front wheel braking force in the n+1th time period, express The predicted change rate of the right front wheel braking force in the n+1th time period, express The predicted change rate of the left rear wheel braking force in the n+1th time period, express The predicted rate of change of the right rear wheel braking force in the n+1th time period;

[0038] Step 4.2: Calculate the vehicle chassis control prediction command according to formula (6): The chassis status prediction dataset for the n+1th time period after :

[0039] (6)

[0040] In formula (5), express The chassis state prediction parameters at the t-th time step in ;

[0041] Step 4.3, use formula (7) to establish The reward function :

[0042] (7)

[0043] In formula (7), and Represents 2 weight coefficients; To make predictive control instructions The reward function of the predicted chassis state is obtained, and:

[0044] (8)

[0045] In formula (8), and Respectively The raw scores of efficient and inefficient operations in the operation evaluation network;

[0046] In formula (7), Produce predictive control instructions for the chassis integrated control network The reward function is:

[0047] (9)

[0048] In formula (9), Predictive control instructions The steering error reward is:

[0049] (10)

[0050] In formula (9), Predictive control instructions The wheel speed error reward is:

[0051] (11)

[0052] In formula (9), Predictive control instructions The braking force error reward is:

[0053] (12)

[0054] In formula (9), Predictive control instructions The suspension parameter error reward is:

[0055] (13)

[0056] Step 4.4, use formula (14) to construct Chassis status dataset The corresponding loss function of the gated recurrent unit layer :

[0057] (14)

[0058] In formula (14), and are 2 weight coefficients, express The corresponding value network loss function of the deep reinforcement learning layer, express The corresponding policy network loss function of the deep reinforcement learning layer;

[0059] Step 4.5, use formula (15) to construct the parameter update formula of the gated recurrent unit layer:

[0060] (15)

[0061] In formula (15), represents the updated parameters of the gated recurrent unit layer, represents the parameters of the gated recurrent unit layer before updating, represents the learning rate;

[0062] Step 4.6: Based on formula (15), the parameters of the gated recurrent unit layer in the chassis integrated control network are adjusted by the gradient descent method to obtain the optimal chassis integrated control model after training.

[0063] Further, step 5 is performed according to the following steps:

[0064] Step 5.1: The chassis domain controller receives the sensor data in real time and analyzes it to form the chassis status time series data at the current time t. ;

[0065] Step 5.2: The optimal operation evaluation model and the optimal chassis integrated control model respectively calculate the chassis state time series data at the current time t. Processing is performed to obtain the operation evaluation result at the current moment. and control instructions ;

[0066] Step 5.3, evaluate the results based on the operation , the chassis status score module accumulates the scores ;

[0067] like For efficient operation, the cumulative score at time t ;

[0068] like is the transition operation, the cumulative score at time t ;

[0069] like is an inefficient operation, and the cumulative score at time t is ,in, represents the cumulative score at time t-1. When t=1, let =0, Represents a standard score.

[0070] The electronic device of the present invention includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the automobile chassis integrated control method described in the claims, and the processor is configured to execute the program stored in the memory.

[0071] The present invention provides a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and the computer program executes the steps of the automobile chassis integrated control method when the computer program is run by a processor.

[0072] Compared with the prior art, the present invention has the following beneficial effects:

[0073] 1. The present invention combines the strong perception ability of deep learning algorithms and the strong decision-making ability of reinforcement learning algorithms, designs a reward function suitable for collaborative control, constructs and trains an operation evaluation model and a chassis integrated control model, and can efficiently and collaboratively control the chassis subsystem;

[0074] 2. The present invention designs a chassis integrated control model that integrates gated recurrent units (GRU) and deep deterministic policy gradient algorithms (DDPG), which can make optimal control outputs based on the coupling between various systems. Different from the existing complex control algorithms, the chassis integrated control model has a small amount of calculation. At the same time, the gated recurrent unit has the characteristics of time series processing, which can play a predictive role and improve the real-time performance of the model.

[0075] 3. The chassis domain controller designed by the present invention can re-optimize the operation evaluation model and chassis integrated control model inside the controller, overcoming the problem of poor adaptability of the controller to changes in vehicle status and environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 This is a diagram of the training process of the operation evaluation model of the present invention;

[0077] Figure 2 This is a diagram of the chassis integrated control network training process of the present invention;

[0078] Figure 3 This is a working architecture diagram of the chassis domain controller of the present invention. DETAILED DESCRIPTION

[0079] In this embodiment, an end-to-end vehicle chassis integrated control method is performed according to the following steps:

[0080] Step 1: Collect and analyze vehicle sensor signals in real driving scenarios to build a labeled chassis status training database , the collected real vehicle data has high reliability and accuracy;

[0081] in, Indicates the transition of the vehicle from an unstable state to a stable state in a real driving scenario. Chassis status datasets. Collecting chassis status training data according to different working conditions can improve the model's adaptability to changes in vehicle status and environment; and , express Middle The chassis status time series data set in a time period has: , express Middle The chassis state parameters at each time step are: express The total number of time steps in , and the interval between time steps is ,and ,in, express Middle The front wheel turning angle of the vehicle at each time step; express Middle The wheel speed at each time step is , express The left front wheel speed at the t-th time step, express The right front wheel speed at the t-th time step, express The left rear wheel speed at the t-th time step, express The right rear wheel speed at the tth time step; express Middle The suspension parameters at time steps are ,in, express The left front suspension stiffness coefficient at the t-th time step is, express The right front suspension stiffness coefficient at the t-th time step is, express The left rear suspension stiffness coefficient at the t-th time step is, express The stiffness coefficient of the right rear suspension at the tth time step in is, express The damping coefficient of the left front suspension at the tth time step in , express The damping coefficient of the right front suspension at the t-th time step in , express The damping coefficient of the left rear suspension at the t-th time step in , express The damping coefficient of the right rear suspension at the tth time step in ; express The braking force of the wheel at the t-th time step, and , express The braking force of the left front wheel at the t-th time step is, express The braking force of the right front wheel at the t-th time step is, express The braking force of the left rear wheel at the t-th time step is, express The braking force of the right rear wheel at the t-th time step;

[0082] make The label is recorded as ;and ,in, Indicates efficient operation. Indicates an inefficient operation.

[0083] Step 2: Training database based on chassis status , establish chassis control training instruction library , Indicates the first Vehicle chassis status dataset The obtained action control instruction set, and , express Middle The chassis control instructions for a period of time, that is, the chassis state time series data set is The chassis control command at the time, and ,in, express Middle Front wheel angular acceleration in a time period; express Middle The rate of change of wheel speed in a time period, and , express Middle The rate of change of the left front wheel speed in a time period is express Middle The rate of change of the right front wheel speed in a time period is express Middle The rate of change of the left rear wheel speed in a time period is express Middle The rate of change of the right rear wheel speed within a time period; express Middle The change rate of the suspension parameters within a time period, and ,in, express Middle The change rate of the stiffness coefficient of the left front suspension in a time period, express Middle The change rate of the suspension stiffness coefficient within a time period, express Middle The change rate of the suspension stiffness coefficient within a time period, express Middle The change rate of the suspension stiffness coefficient within a time period, express Middle The change rate of the damping coefficient of the left front suspension in a time period, express Middle The change rate of the damping coefficient of the right front suspension in a time period, express Middle The change rate of the damping coefficient of the left rear suspension in a time period, express Middle The change rate of the damping coefficient of the right rear suspension within a time period; express Middle The rate of change of the wheel braking force in a time period, and , express Middle The rate of change of the left front wheel braking force in a time period is express Middle The rate of change of the right front wheel braking force in a time period is express Middle The rate of change of the left rear wheel braking force in a time period is express Middle The rate of change of the right rear wheel braking force within a time period.

[0084] By fitting Next time period The parameter change rate in The control command training data set is based on the chassis state time series data set of the real driving environment. The control command obtained can fully consider the vehicle parameter coupling relationship. Improved accuracy of control instructions for initial state data;

[0085] The chassis control command is calculated as formula (d), and the chassis state loss function is constructed using formula (a) :

[0086] (a)

[0087] Use formula (b) to calculate the chassis state loss function about Find the derivative and set it to 0 using formula (c), so as to find the function that minimizes the loss function using formula (d). ;

[0088] (b)

[0089] (c)

[0090] (d)

[0091] Similarly, we can get:

[0092] .

[0093] Step 3: Construct an operation evaluation network based on spatiotemporal feature sequences and evaluate any Process and get the corresponding prediction label , thus constructing the loss function , used to train the operation evaluation network to obtain the optimal operation evaluation model;

[0094] Step 3.1, the operation evaluation network includes: a gated recurrent unit layer, a fully connected layer, and an activation function layer;

[0095] like Figure 1 As shown, the gated recurrent unit layer is Process and extract chassis status time series data set The chassis parameter change timing characteristics in the data are reduced, the data dimension is reduced, the subsequent calculation amount is reduced, and the output is Middle Hidden state in time period ;

[0096] The fully connected layer is based on , output Middle The raw score set for the time period ;in, represents the weight matrix of the fully connected layer, represents the bias term of the fully connected layer, and , express Middle The raw score of efficient operations in a time period; express Middle The raw score of inefficient operations in a time period, express Middle The raw score of transition operations in the time period;

[0097] Activation function layer pair By performing nonlinear transformation, the activation function layer can improve the expressiveness of the model and enhance the computational efficiency. Middle Predicted labels for time periods ,in, express The original score of the i-th type The predicted probability after nonlinear transformation, and , They represent efficient operation, inefficient operation and transition operation respectively.

[0098] Step 3.2: Use formula (1) to construct the loss function of the operation evaluation network ;

[0099] (1)

[0100] In formula (1), represents the weighting coefficient;

[0101] Step 3.3: Use formula (2) to construct the loss function The gradient update formula is:

[0102] (2)

[0103] Step 3.4: Use equations (3) and (4) to get a new weight matrix and the new bias term :

[0104] (3)

[0105] (4)

[0106] In formula (3) and formula (4), Represents the learning rate.

[0107] Step 3.5, use formula (5) to construct the parameter update formula of the gated recurrent unit layer:

[0108] (5)

[0109] In formula (5), represents the updated parameters of the gated recurrent unit layer, represents the parameters of the gated recurrent unit layer before updating, represents the learning rate;

[0110] Step 3.6: Based on equations (3), (4) and (5), the operation evaluation network is trained by gradient descent method to obtain a trained operation evaluation model;

[0111] Step 4: Build a chassis integrated control network and use it to analyze the chassis status data set Process and construct the loss function , used to train the chassis integrated control network to obtain the chassis integrated control model;

[0112] Step 4.1, the chassis integrated control network is composed of a gated recurrent unit layer and a deep reinforcement learning layer;

[0113] like Figure 2 As shown, the gated recurrent unit layer Chassis status time series data set for the nth time period and Chassis status time series data set for the n+1th time period Process and output The corresponding low-dimensional chassis state vector and The corresponding low-dimensional chassis state vector ;

[0114] The deep reinforcement learning layer receives and , and use the DDPG algorithm to Process and generate Corresponding predicted chassis control commands ,Right now Indicates chassis integrated control network based on Output the predicted change rate of chassis parameters in the next time period, express The predicted rate of change of the front wheel angle in the n+1th time period, that is, S, express The predicted rate of change of wheel speed in the n+1th time period, express The predicted change rate of the suspension parameters in the n+1th time period, express The predicted rate of change of the wheel braking force in the n+1th time period, and ; express The predicted rate of change of the left front wheel speed in the n+1th time period, express The predicted rate of change of the right front wheel speed in the n+1th time period, express The predicted rate of change of the left rear wheel speed in the n+1th time period, express The predicted rate of change of the right rear wheel speed in the n+1th time period; express Middle The predicted change rate of the suspension parameters within a time period, , express The predicted change rate of the left front suspension stiffness coefficient in the n+1th time period, express The predicted change rate of the right front suspension stiffness coefficient in the n+1th time period, express The predicted change rate of the left rear suspension stiffness coefficient in the n+1th time period, express The predicted change rate of the right rear suspension stiffness coefficient in the n+1th time period, express The predicted change rate of the left front suspension damping coefficient in the n+1th time period, express The predicted change rate of the right front suspension damping coefficient in the n+1th time period, express The predicted change rate of the left rear suspension damping coefficient in the n+1th time period, express The predicted change rate of the right rear suspension damping coefficient in the n+1th time period; express Middle The predicted rate of change of the braking force of the wheel within a time period, and , express The predicted change rate of the left front wheel braking force in the n+1th time period, express The predicted change rate of the right front wheel braking force in the n+1th time period, express The predicted change rate of the left rear wheel braking force in the n+1th time period, express The predicted rate of change of the right rear wheel braking force in the n+1th time period;

[0115] Step 4.2: Calculate the vehicle chassis control prediction command according to formula (6): The chassis status prediction dataset for the n+1th time period after :

[0116] (6)

[0117] In formula (5), express The chassis state prediction parameters at the t-th time step in ;

[0118] and

[0119] , the specific calculation is as shown in formula (e);

[0120] (e)

[0121] Step 4.3, use formula (7) to establish The reward function :

[0122] (7)

[0123] In formula (7), and Represents 2 weight coefficients, is the reward function of the predicted chassis state obtained after making the predicted control command, and there are:

[0124] (8)

[0125] In formula (8), and Respectively The original scores of efficient and inefficient operations in the operation evaluation network are obtained by evaluating the predicted chassis state using the operation evaluation model, which can avoid the output of the chassis integrated control network simply fitting the control instructions. , improve the generalization ability of the model.

[0126] In formula (7), Produce predictive control instructions for the chassis integrated control network The reward function is:

[0127] (9)

[0128] In formula (9), Predictive control instructions The steering error reward is:

[0129] (10)

[0130] In formula (9), Predictive control instructions The wheel speed error reward is:

[0131] (11)

[0132] In formula (9), Predictive control instructions The braking force error reward is:

[0133] (12)

[0134] In formula (9), Predictive control instructions The suspension parameter error reward is:

[0135] (13)

[0136] Step 4.4, the chassis integrated control network aims to minimize the loss function, chassis state data set Perform small batch training, update the parameters of the deep reinforcement learning layer, and obtain the value network loss function and the policy network loss function ;

[0137] (f)

[0138] (g)

[0139] in, For the online Q network to the low-dimensional chassis state vector and control instructions The estimate, Represents the low-dimensional chassis state vector of the online policy network Output predictive control instructions, Represents the target policy network's response to the next low-dimensional chassis state vector The next predictive control instruction output, Represents the low-dimensional chassis state vector Taken by the target strategy network The Q value when making a predicted action.

[0140] Step 4.5: Use formula (14) to construct the loss function of the gated recurrent unit layer :

[0141] (14)

[0142] In formula (14), and are 2 weight coefficients, express The corresponding value network loss function of the deep reinforcement learning layer, express The corresponding policy network loss function of the deep reinforcement learning layer;

[0143] Step 4.6, use formula (14) to construct the parameter update formula of the gated recurrent unit layer:

[0144] (15)

[0145] In formula (15), represents the updated parameters of the gated recurrent unit layer, represents the parameters of the gated recurrent unit layer before updating, represents the learning rate;

[0146] Step 4.8: Based on formula (15), the parameters of the gated recurrent unit layer in the chassis integrated control network are adjusted by the gradient descent method to obtain the trained chassis integrated control model;

[0147] Step 5: Build a chassis domain controller based on the joint learning network, including: the trained operation evaluation model, the trained chassis integrated control model and the chassis status scoring module, such as Figure 3 As shown;

[0148] Step 5.1: The chassis domain controller receives the sensor data in real time and analyzes it to form the chassis status time series data at the current time t. ;

[0149] Step 5.2: The operation evaluation model and chassis integrated control model respectively calculate the chassis state time series data at the current time t. Processing is performed to obtain the operation evaluation result at the current moment. and control instructions ;

[0150] Step 5.3, evaluate the results based on the operation , the chassis status score module accumulates the scores ;

[0151] like For efficient operation, the cumulative score at time t ;

[0152] like is the transition operation, the cumulative score at time t ;

[0153] like is an inefficient operation, and the cumulative score at time t is ,in, represents the cumulative score at time t-1. When t=1, let =0, represents a standard score;

[0154] Step 6, when and When the chassis domain controller executes the chassis control command, ; Otherwise, the chassis domain controller does not intervene in chassis control and is in silent state;

[0155] when When the chassis domain controller uploads the sensor data of the vehicle during the transition from an unstable state to a stable state to the cloud to supplement the chassis state training database, it returns to step 1 and executes in sequence to further train a better operation evaluation model and a better chassis integrated control model and update them to the chassis domain controller; the specific implementation is that the chassis domain controller reserves a storage space that can store chassis state time series data of N time periods, which are stored in sequence as ,when The evaluation results are efficient operation. is an inefficient operation, and the efficient raw score that satisfies them is higher than the chassis state time series data of the previous time period, that is, ,and Represents the operation evaluation model for chassis status time series data The original scores of efficient operations during the evaluation process are uploaded to the cloud to supplement the chassis status training database and are expressed in chronological order as ; After the training is completed, it is downloaded to the chassis domain controller using over-the-air (OTA) technology.

[0156] Step 7, after t+1 is assigned to t, return to step 5.1 and execute sequentially.

[0157] In this embodiment, an electronic device includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the above method, and the processor is configured to execute the program stored in the memory.

[0158] In this embodiment, a computer-readable storage medium stores a computer program on the computer-readable storage medium, and the computer program executes the steps of the above method when executed by a processor.

Claims

1. An end-to-end vehicle chassis integrated control method, characterized in that: Follow these steps: Step 1: Collect and analyze vehicle sensor signals in real driving scenarios to build a labeled chassis status training database ,in, Indicates the transition of the vehicle from an unstable state to a stable state in a real driving scenario. chassis status dataset, and , express Middle Chassis status time series data set within a time period, and , express Middle The chassis state parameters at each time step are: express The total number of time steps in , and the interval between time steps is ,and ,in, express Middle The front wheel turning angle of the vehicle at each time step; express Middle The wheel speed at each time step is express Middle Suspension parameters at time steps; express Middle The braking force of the wheel in each time step make The label is recorded as ;and ,in, Indicates efficient operation. Indicates inefficient operation; Step 2: Training database based on chassis status , establish chassis control command training library , Indicates Vehicle chassis status dataset The corresponding chassis control instruction set, and , express Middle The chassis control instructions for a period of time, that is, the chassis state time series data set is The chassis control command at the time, and ,in, express Middle Front wheel angular acceleration in a time period; express Middle The rate of change of wheel speed in a time period, and , express Middle The rate of change of the left front wheel speed in a time period is represents the rate of change of the right front wheel speed, express Middle The rate of change of the left rear wheel speed in a time period is express Middle The rate of change of the right rear wheel speed within a time period; express Middle The change rate of the suspension parameters within a time period, and ,in, express Middle The change rate of the stiffness coefficient of the left front suspension in a time period, express Middle The change rate of the suspension stiffness coefficient within a time period, express Middle The change rate of the suspension stiffness coefficient within a time period, express Middle The change rate of the suspension stiffness coefficient within a time period, express Middle The change rate of the damping coefficient of the left front suspension in a time period, express Middle The change rate of the damping coefficient of the right front suspension within a time period, express Middle The change rate of the damping coefficient of the left rear suspension in a time period, express Middle The change rate of the damping coefficient of the right rear suspension within a time period; express Middle The rate of change of the wheel braking force in a time period, and , express Middle The rate of change of the left front wheel braking force in a time period is express Middle The rate of change of the right front wheel braking force in a time period is express Middle The rate of change of the left rear wheel braking force in a time period is express Middle The rate of change of the right rear wheel braking force within a time period; Step 3: Construct an operation evaluation network based on spatiotemporal feature sequences and evaluate any Process and get the corresponding prediction label , thus constructing the loss function , used to train the operation evaluation network to obtain the optimal operation evaluation model; Step 4: Build a chassis integrated control network and use it to control the randomly selected Process and get the corresponding control instructions , thus constructing the loss function , used to train the chassis integrated control network and obtain the optimal chassis integrated control model; Step 5: Build a chassis domain controller based on the joint learning network, including: the trained optimal operation evaluation model, the trained optimal chassis integrated control model and the chassis state scoring module, and perform chassis state time series data at the current time t. Processing, the current operation evaluation result , control instructions and cumulative score ; Step 6, when and When the chassis domain controller executes the chassis control command, ; Otherwise, the chassis domain controller does not intervene in chassis control and is in silent state; when When the chassis domain controller uploads the sensor data of the vehicle during the transition from an unstable state to a stable state to the cloud to supplement the chassis state training database, it returns to step 1 and executes the better operation evaluation model and the better chassis integrated control model further trained and updated in the chassis domain controller; Step 7, after t+1 is assigned to t, return to step 5.1 and execute sequentially.

2. The end-to-end vehicle chassis integrated control method according to claim 1, characterized in that: The step 3 is performed according to the following steps: Step 3.1, the operation evaluation network includes: a gated recurrent unit layer, a fully connected layer, and an activation function layer; The gated recurrent unit layer is Process and output Middle Hidden state in time period ; The fully connected layer is based on , output Middle The raw score set for the time period ;in, represents the weight matrix of the fully connected layer, represents the bias term of the fully connected layer, and , express Middle The raw score of efficient operations in a time period; express Middle The raw score of inefficient operations in a time period, express Middle The raw score of transition operations in the time period; The activation function layer Perform nonlinear transformation to obtain Middle Predicted labels for time periods ,in, express The original score of the i-th type The predicted probability after nonlinear transformation, and , They represent efficient operation, inefficient operation and transition operation respectively; Step 3.2: Use formula (1) to construct the loss function of the operation evaluation network ; (1) In formula (1), represents the weighting coefficient; Step 3.3: Use formula (2) to construct the loss function The gradient update formula is: (2) Step 3.4: Use equations (3) and (4) to get a new weight matrix and the new bias term : (3) (4) In formula (3) and formula (4), represents the learning rate; Step 3.5, use formula (5) to construct the parameter update formula of the gated recurrent unit layer: (5) In formula (5), represents the updated parameters of the gated recurrent unit layer, represents the parameters of the gated recurrent unit layer before updating, represents the learning rate; Step 3.6: Based on equations (3), (4) and (5), the operation evaluation network is trained by gradient descent method to obtain the optimal operation evaluation model after training.

3. The end-to-end vehicle chassis integrated control method according to claim 2, characterized in that: The step 4 is performed according to the following steps: Step 4.1, the chassis integrated control network is composed of a gated recurrent unit layer and a deep reinforcement learning layer; The gated recurrent unit layer is Chassis status time series data set for the nth time period and Chassis status time series data set for the n+1th time period Process and output The corresponding low-dimensional chassis state vector and The corresponding low-dimensional chassis state vector ; The deep reinforcement learning layer receives and , and use the DDPG algorithm to Process and generate Corresponding predicted chassis control commands ,Right now Indicates chassis integrated control network based on Output the predicted change rate of chassis parameters in the next time period, express The predicted rate of change of the front wheel angle in the n+1th time period, that is, S, express The predicted rate of change of wheel speed in the n+1th time period, express The predicted change rate of the suspension parameters in the n+1th time period, express The predicted rate of change of the wheel braking force in the n+1th time period, and ; express The predicted rate of change of the left front wheel speed in the n+1th time period, express The predicted rate of change of the right front wheel speed in the n+1th time period, express The predicted rate of change of the left rear wheel speed in the n+1th time period, express The predicted rate of change of the right rear wheel speed in the n+1th time period; express Middle The predicted change rate of the suspension parameters within a time period, , express The predicted change rate of the left front suspension stiffness coefficient in the n+1th time period, express The predicted change rate of the right front suspension stiffness coefficient in the n+1th time period, express The predicted change rate of the left rear suspension stiffness coefficient in the n+1th time period, express The predicted change rate of the right rear suspension stiffness coefficient in the n+1th time period, express The predicted change rate of the left front suspension damping coefficient in the n+1th time period, express The predicted change rate of the right front suspension damping coefficient in the n+1th time period, express The predicted change rate of the left rear suspension damping coefficient in the n+1th time period, express The predicted change rate of the right rear suspension damping coefficient in the n+1th time period; express Middle The predicted rate of change of the braking force of the wheel within a time period, and , express The predicted change rate of the left front wheel braking force in the n+1th time period, express The predicted change rate of the right front wheel braking force in the n+1th time period, express The predicted change rate of the left rear wheel braking force in the n+1th time period, express The predicted rate of change of the right rear wheel braking force in the n+1th time period; Step 4.2: Calculate the vehicle chassis control prediction command according to formula (6): The chassis status prediction dataset for the n+1th time period after : (6) In formula (5), express The chassis state prediction parameters at the t-th time step in ; Step 4.3, use formula (7) to establish The reward function : (7) In formula (7), and Represents 2 weight coefficients; To make predictive control instructions The reward function of the predicted chassis state is obtained, and: (8) In formula (8), and Respectively The raw scores of efficient and inefficient operations in the operation evaluation network; In formula (7), Produce predictive control instructions for the chassis integrated control network The reward function is: (9) In formula (9), Predictive control instructions The steering error reward is: (10) In formula (9), Predictive control instructions The wheel speed error reward is: (11) In formula (9), Predictive control instructions The braking force error reward is: (12) In formula (9), Predictive control instructions The suspension parameter error reward is: (13) Step 4.4, use formula (14) to construct Chassis status dataset The corresponding loss function of the gated recurrent unit layer : (14) In formula (14), and are 2 weight coefficients, express The corresponding value network loss function of the deep reinforcement learning layer, express The corresponding policy network loss function of the deep reinforcement learning layer; Step 4.5, use formula (15) to construct the parameter update formula of the gated recurrent unit layer: (15) In formula (15), represents the updated parameters of the gated recurrent unit layer, represents the parameters of the gated recurrent unit layer before updating, represents the learning rate; Step 4.6: Based on formula (15), the parameters of the gated recurrent unit layer in the chassis integrated control network are adjusted by the gradient descent method to obtain the optimal chassis integrated control model after training.

4. The end-to-end vehicle chassis integrated control method according to claim 3, characterized in that: Step 5 is to follow these steps: Step 5.1: The chassis domain controller receives the sensor data in real time and analyzes it to form the chassis status time series data at the current time t. ; Step 5.2: The optimal operation evaluation model and the optimal chassis integrated control model respectively calculate the chassis state time series data at the current time t. Processing is performed to obtain the operation evaluation result at the current moment. and control instructions ; Step 5.3, evaluate the results based on the operation , the chassis status score module accumulates the scores ; like For efficient operation, the cumulative score at time t ; like is the transition operation, the cumulative score at time t ; like is an inefficient operation, and the cumulative score at time t is ,in, represents the cumulative score at time t-1. When t=1, let =0, Represents a standard score.

5. An electronic device, comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the automobile chassis integration control method described in any one of claims 1-4, and the processor is configured to execute the program stored in the memory.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the vehicle chassis integration control method according to claim 1 are executed.

Images