A coordinated control method for path tracking and stability of electric vehicles

Abstract

The present invention provides a method for coordinated control of path tracking and stability of an electric vehicle, comprising: constructing a nonlinear ideal vehicle dynamics model based on lateral and yaw; establishing a long short-term memory network prediction model of a nonlinear system, and integrating the long short-term memory network prediction model into a model prediction control algorithm, completing the construction of a path tracking controller, and outputting a front wheel turning angle through the path tracking controller; constructing a stability controller; constructing a torque distribution model for multi-objective coordinated optimization, and outputting an additional yaw moment through the torque distribution model; connecting the front wheel turning angle with a BP neural network, training and outputting phase plane contraction and expansion coefficients and dividing the phase plane area, judging the area where the vehicle is located, and outputting the front wheel turning angle and additional yaw moment after coordinated control according to a coordinated control weight coefficient, so as to finally realize the path tracking and coordinated stability control of a four-wheel independent drive electric vehicle.

Description

Technical Field

[0001] The present invention relates to the technical field of electric vehicles, and in particular to a method for coordinated control of electric vehicle path tracking and stability. Background Art

[0002] As environmental pollution and energy shortage become increasingly prominent, electric vehicles, with their zero emissions and high energy efficiency, have attracted widespread attention from researchers and developers, especially four-wheel independent drive electric vehicles, as a type of electric vehicle.

[0003] In the control process of four-wheel independent drive electric vehicles, path tracking control enables the vehicle to accurately follow the predetermined reference trajectory. However, under some extreme conditions, when the vehicle follows the predetermined trajectory, the stability of the vehicle decreases rapidly, which may cause other safety issues such as vehicle rollover. On the other hand, good stability will result in the vehicle being unable to achieve high accuracy in following the predetermined path. Therefore, how to achieve coordinated control of path tracking and stability, and thus improve the active safety performance of the vehicle, is a technical problem that needs to be solved by those skilled in the art. Summary of the invention

[0004] The object of the present invention is to provide a method for coordinated control of electric vehicle path tracking and stability to improve the active safety performance of the vehicle.

[0005] A method for coordinated control of electric vehicle path tracking and stability comprises the following steps:

[0006] Step S1, based on the nonlinear characteristics of the tires of the four-wheel independent drive electric vehicle, construct a nonlinear ideal vehicle dynamics model based on lateral and yaw;

[0007] Step S2, based on the nonlinear ideal dynamics model of the vehicle constructed in step S1, the ability of the long short-term memory network to predict the future state is connected with the model predictive control algorithm, a long short-term memory network prediction model of the nonlinear system is established, the state quantity, control quantity and objective function are determined, and the long short-term memory network prediction model is integrated into the model predictive control algorithm to complete the construction of the path tracking controller, and the front wheel steering angle is output through the path tracking controller;

[0008] Step S3, based on the nonlinear ideal vehicle dynamics model constructed in step S1 and the front wheel turning angle output in step S2, establish the system tracking error of the center of mass sideslip angle and the yaw angle, introduce the additional yaw moment equation, and construct a stability controller;

[0009] Step S4, based on the stability controller constructed in step S3, a multi-objective particle swarm optimization algorithm improved by a differential evolution mechanism is introduced, tire slip rate and energy utilization rate are determined as optimization targets, a multi-objective collaborative optimization torque distribution model is constructed, and an additional yaw moment is output through the torque distribution model;

[0010] Step S5, connect the front wheel steering angle output in step S2 with the BP neural network, train the output phase plane inward and outward expansion coefficients and divide the phase plane area, judge the area where the vehicle is located, output the front wheel steering angle and additional yaw moment after collaborative control according to the collaborative control weight coefficient, and finally realize the path tracking and stability collaborative control of the four-wheel independent drive electric vehicle.

[0011] The electric vehicle path tracking and stability coordinated control method provided by the present invention has the following beneficial effects:

[0012] 1. Considering that the traditional model predictive control algorithm predicts the future state through the dynamic model, but the vehicle dynamics model cannot fully and accurately describe the behavior of the system. The present invention introduces the long short-term memory network prediction model, which uses the long short-term memory network to learn historical data and nonlinear relationships, replacing the traditional vehicle dynamics model to predict the future state of the vehicle, thereby improving the robustness and adaptability of the model;

[0013] 2. Different from the general stability controller design, the present invention adopts a new sliding surface function, considers the influence of the sideslip angle and yaw angle on stability, and adaptively controls their weight coefficients through fuzzy algorithm, replaces the fixed gain coefficient with a variable control gain coefficient, and designs a stability controller based on the sideslip angle and yaw angle to reduce the influence of buffeting;

[0014] 3. Based on the phase plane theory, the present invention introduces the shrinkage factor and the expansion factor through the BP neural network algorithm to divide the boundary twice on the basis of the boundary constructed by the double-line method, so as to judge the vehicle state and the area where it is located, and complete the coordinated control of path tracking and stability;

[0015] 4. The present invention combines the designed path tracking controller with the designed stability controller, divides and judges the area using the phase plane method, and outputs the front wheel steering angle and additional yaw moment after cooperative control according to the cooperative control weight coefficient, so as to give full play to the potential of the vehicle stability system in the process of path tracking and improve the vehicle handling stability;

[0016] 5. The present invention can effectively solve the path tracking and stability problems of vehicles under extreme working conditions, improve the driving safety of vehicles, and thus reduce traffic congestion and accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of the electric vehicle path tracking and stability coordinated control method of the present invention;

[0018] Figure 2 A comparison diagram of the actual center of mass sideslip angle and the ideal center of mass sideslip angle obtained by the method provided by the present invention;

[0019] Figure 3 A comparison diagram of the actual trajectory and the reference trajectory obtained by the method provided by the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] See also Figure 1 The electric vehicle path tracking and stability coordinated control method provided by the embodiment of the present invention includes steps S1 to S5.

[0022] Step S1, based on the nonlinear characteristics of the tires of the four-wheel independent drive electric vehicle, a nonlinear ideal vehicle dynamics model based on lateral and yaw is constructed.

[0023] Among them, in step S1, based on the nonlinear characteristics of the tires of the four-wheel independent drive electric vehicle, a nonlinear ideal vehicle dynamics model based on lateral and yaw is constructed, and the specific expression is:

[0024] ;

[0025] ;

[0026] The nonlinear ideal dynamics model of the vehicle is rewritten as a state space equation, which is expressed as follows:

[0027] ;

[0028] in, is the sideslip angular velocity of the center of mass, k f is the front wheel cornering stiffness, k r is the rear wheel cornering stiffness, m is the vehicle mass, v x is the vehicle longitudinal velocity, is the sideslip angle of the center of mass, is the distance from the center of mass to the front axle, b is the distance from the center of mass to the rear axle, is the yaw angular velocity, is the front wheel turning angle, is the yaw angular acceleration, is the moment of inertia about the OZ axis.

[0029] Step S2, based on the nonlinear vehicle ideal dynamics model constructed in step S1, the ability of the long short-term memory network to predict future states is connected to the model predictive control algorithm, a long short-term memory network prediction model of the nonlinear system is established, the state quantity, control quantity and objective function are determined, and the long short-term memory network prediction model is integrated into the model predictive control algorithm to complete the construction of the path tracking controller, and the front wheel steering angle is output through the path tracking controller.

[0030] Wherein, step S2 specifically includes:

[0031] Obtain vehicle status data, perform preprocessing and normalization; then use the deep learning toolbox to define and train the long short-term memory network prediction model; then, use the trained long short-term memory network prediction model to predict the future state of the vehicle; finally, integrate the long short-term memory network model into the model predictive control algorithm framework, complete the construction of the path tracking controller, and output the front wheel steering angle through the path tracking controller;

[0032] Among them, the state update equation of the long short-term memory network prediction model is:

[0033] ;

[0034] ;

[0035] ;

[0036] ;

[0037] ;

[0038] in, Represents the input gate at the current moment; f t Represents the forget gate at the current moment; Represents the output gate at the current moment; The memory unit representing the current moment; h t Indicates the hidden state at the current moment; Indicates the hidden state at the previous moment; Represents the memory unit of the previous moment; represents the input vector; is the sigmoid activation function; tanh is the hyperbolic tangent function; Represents element-wise multiplication operation; W i , W f , W o , W c is the weight of the LSTM prediction model; b i , b f , b o , b c is the bias matrix of the LSTM prediction model.

[0039] The established LSTM prediction model can learn the dynamic behavior of the system after training. In each control step, the current state of the vehicle is provided as input to the LSTM prediction model, which updates its weights by minimizing the error between the predicted output and the actual output. The gradient is calculated using the gradient descent method and the back propagation algorithm.

[0040] Based on the state space equation in step S1, the vehicle path tracking error model is obtained after coordinate transformation, and the expression is as follows:

[0041] ;

[0042] ;

[0043] ;

[0044] in, is the system state matrix The derivative of ; is the lateral error; is the lateral error derivative; is the heading angle error; is the heading angle error derivative; is the input; the front wheel angle As input quantity; and is the time-varying matrix of the linear system;

[0045] Objective function of the path-following controller J for:

[0046] ;

[0047] in, For the prediction time domain, To control the time domain, is from i Time has come The actual trajectory of the vehicle output by the long short-term memory network prediction model within time, is from i Time has come The reference trajectory of the vehicle within time, is the weight matrix of vehicle tracking capability, is from i Time has come -1 moment control increment, is the weight matrix that controls the increment, is the weight coefficient, is the relaxation factor; Reflects the vehicle i Time has come The ability to track the desired trajectory within a certain time frame, It reflects the smoothness of the vehicle during steering, that is, the smooth change of the control amount.

[0048] The optimization problem of the path tracking controller is designed to minimize the objective function J , by optimizing and solving the control input sequence to obtain the minimized objective function.

[0049] In minimizing the objective function J During the process, the following conditions are met:

[0050] ;

[0051] ;

[0052] ;

[0053] in, To predict the control quantity in the time domain, is the maximum value of the control quantity in the prediction time domain, is the minimum value of the control quantity in the prediction time domain, is from i Time has come The maximum value of the control increment at time -1, is from i Time has come The minimum value of the control increment at time -1.

[0054] The established nonlinear system long short-term memory network is introduced into the model predictive control algorithm to obtain the control variable front wheel angle in the path tracking controller.

[0055] Step S3, based on the nonlinear ideal vehicle dynamics model constructed in step S1 and the front wheel turning angle output in step S2, establish the system tracking error of the center of mass sideslip angle and yaw angle, introduce the additional yaw moment equation, and construct a stability controller.

[0056] If the vehicle becomes unstable during path tracking, the yaw torque controller needs to intervene directly to stabilize the vehicle body posture to prevent the vehicle from rolling over or other safety accidents.

[0057] In this embodiment, step S3 specifically includes:

[0058] The nonlinear ideal vehicle dynamics model constructed in step S1 is rewritten to obtain the following formula:

[0059] ;

[0060] in, represents the additional yaw moment;

[0061] The sliding mode control algorithm is used to design the DYC system. In order to take into account the influence of yaw rate and sideslip angle on the lateral stability of the vehicle, the sliding mode surface is Designed for:

[0062] ;

[0063] in, , is a positive coefficient; is the system tracking error derivative;

[0064] ;

[0065] in, is the system tracking error weight coefficient, ; is the ideal yaw rate; is the ideal center-of-mass sideslip angular velocity; is the center of mass sideslip angular velocity error; is the yaw rate error;

[0066] Sliding surface Taking the second-order derivative of the expression, we can get:

[0067] ;

[0068] in, is the second-order derivative of the system tracking error, and are the actual center of mass sideslip angle acceleration and the ideal center of mass sideslip angle acceleration, is the ideal yaw angular acceleration;

[0069] Then, the first-order derivative of the sliding surface is The function expression is:

[0070] ;

[0071] The principle of selecting the convergence law is to select a state point with a larger convergence speed to ensure that the system can converge quickly. However, too large a convergence speed may cause a chattering problem in the system. Therefore, in order to minimize the chattering problem, the present invention introduces an exponential convergence law function based on the constant speed convergence law.

[0072] Specifically, the exponential reaching law function is introduced to obtain the following formula:

[0073] ;

[0074] Then we get:

[0075] ;

[0076] The additional yaw moment The expression is:

[0077] ;

[0078] in, is the symbolic function, and is a coefficient greater than 0;

[0079] System tracking error weight coefficient It should be able to perform adaptive adjustment according to the dynamic state of the vehicle. The present invention adopts fuzzy control to adjust the system tracking error weight coefficient ,The tracking errors of the fuzzy processed input variables sideslip angle and yaw angle are inferred through fuzzy control rules to obtain the stability controller.

[0080] Step S4, based on the stability controller constructed in step S3, introduces a multi-objective particle swarm optimization algorithm improved by a differential evolution mechanism, determines tire slip rate and energy utilization rate as optimization targets, constructs a multi-objective collaborative optimization torque distribution model, and outputs an additional yaw moment through the torque distribution model.

[0081] Wherein, step S4 specifically includes:

[0082] Based on the stability controller constructed in step S3, the differential evolution search algorithm is introduced with tire slip rate and energy utilization as optimization targets to construct a torque distribution model for multi-objective collaborative optimization, where tire slip rate The calculation formula is:

[0083] ;

[0084] in, is the wheel radius, For the e The angular velocity of each wheel;

[0085] Minimize the sum of the slip ratios It is expressed as:

[0086] ;

[0087] in, For the e The slip rate of each wheel;

[0088] The energy utilization rate is determined based on the motor output power, and the expression is as follows:

[0089] ;

[0090] ;

[0091] in, is the energy utilization rate, is the motor output power, Indicates e The input power of the motor corresponding to each wheel, For the e The weight value corresponding to each wheel is dt Indicates time t The differential of is the motor efficiency;

[0092] The expression of the fitness function of the improved multi-objective particle swarm optimization algorithm is:

[0093] ;

[0094] in, J total is the total fitness objective function, J 1 is the objective function of minimizing the slip rate, J 2 is the objective function of vehicle energy utilization, and They are J 1 and J A weight factor of 2;

[0095] During the iteration process of the multi-objective particle swarm optimization algorithm, the differential evolution scaling factor and the differential evolution crossover probability are introduced. After each particle update, the differential evolution operation is performed to obtain the global optimal solution and distribute the torque to the four wheels.

[0096] Step S5, connect the front wheel steering angle output in step S2 with the BP neural network, train the output phase plane inward and outward expansion coefficients and divide the phase plane area, judge the area where the vehicle is located, output the front wheel steering angle and additional yaw moment after collaborative control according to the collaborative control weight coefficient, and finally realize the path tracking and stability collaborative control of the four-wheel independent drive electric vehicle.

[0097] Wherein, step S5 specifically includes:

[0098] Divide the phase plane region, where:

[0099] The stable region satisfies the following equation:

[0100] ;

[0101] The critical stability region satisfies the following equation:

[0102] ;

[0103] The unstable region satisfies the following equation:

[0104] ;

[0105] in, is the intercept of the stability boundary, , is the shrinkage coefficient of the stable region boundary, B 1. B 2 is the original boundary coefficient, is the intercept of the unstable boundary, , is the expansion coefficient of the stable region boundary;

[0106] For the shrinkage coefficient of the stable region boundary and the expansion coefficient of the stable region boundary , using a BP neural network model, the BP neural network model includes an input layer, a hidden layer, and an output layer, and the BP neural network model is trained and then solved. The specific steps are as follows:

[0107] The input is set as the front wheel angle, longitudinal speed and road adhesion coefficient, and the number of hidden layer neurons is set to 64; then, the variable structure sigmoid function is used as the activation function, and the chain derivative rule is used to back propagate the error during the training process to achieve the purpose of updating the weights and biases. Finally, after the neural network is trained, the shrinkage coefficient of the stable region boundary is output. and the expansion coefficient of the stable region boundary ;

[0108] Among them, the input layer matrix And the output layer matrix Defined as:

[0109] ;

[0110] in, is the road adhesion coefficient;

[0111] According to the results of dividing the phase plane area, the working areas of the path tracking controller and the stability controller are divided as follows:

[0112] Only the path tracking area satisfies the following equation:

[0113] ;

[0114] The path tracking area and the stability control area coordinate control area satisfy the following formula:

[0115] ;

[0116] Only the stability control region satisfies the following equation:

[0117] ;

[0118] The coordinated switching function is further designed to use the hyperbolic tangent function, and the coordinated control weight coefficient is obtained q Satisfy the following formula:

[0119] ;

[0120] Finally, the front wheel steering angle and additional yaw moment after coordinated control satisfy the following equation:

[0121] ;

[0122] ;

[0123] in, represents the front wheel steering angle after cooperative control, represents the additional yaw moment after cooperative control.

[0124] By introducing the BP neural network to divide the phase plane area, the front wheel angle and additional yaw moment after collaborative control are output according to the weight coefficient, ultimately achieving path tracking and stability collaborative control of four-wheel independent drive new energy vehicles.

[0125] Figure 2 The figure is a comparison diagram of the actual center of mass sideslip angle and the ideal center of mass sideslip angle obtained by the method provided by the present invention. Figure 3 A comparison diagram of the actual trajectory and the reference trajectory obtained by the method provided by the present invention. Figure 2 and Figure 3The test conditions are: vehicle speed 90km / h, road adhesion coefficient 0.5. Figure 2 It can be seen that the maximum center of mass side slip angle of the vehicle is close to 0.008rad, which makes the vehicle stable. Figure 3 It can be seen that an error occurs between the vehicle and the actual trajectory at the turning point of the double-lane-changing condition. The reason for the error is to prevent the vehicle from losing stability and other safety issues at this position, but a certain tracking control accuracy is lost. This phenomenon is in line with the design purpose of the present invention.

[0126] In summary, the electric vehicle path tracking and stability coordinated control method provided by the present invention has the following beneficial effects:

[0127] 1. Considering that the traditional model predictive control algorithm predicts the future state through the dynamic model, but the vehicle dynamics model cannot fully and accurately describe the behavior of the system. The present invention introduces the long short-term memory network prediction model, which uses the long short-term memory network to learn historical data and nonlinear relationships, replacing the traditional vehicle dynamics model to predict the future state of the vehicle, thereby improving the robustness and adaptability of the model;

[0128] 2. Different from the general stability controller design, the present invention adopts a new sliding surface function, considers the influence of the sideslip angle and yaw angle on stability, and adaptively controls their weight coefficients through fuzzy algorithm, replaces the fixed gain coefficient with a variable control gain coefficient, and designs a stability controller based on the sideslip angle and yaw angle to reduce the influence of buffeting;

[0129] 3. Based on the phase plane theory, the present invention introduces the shrinkage factor and the expansion factor through the BP neural network algorithm to divide the boundary twice on the basis of the boundary constructed by the double-line method, so as to judge the vehicle state and the area where it is located, and complete the coordinated control of path tracking and stability;

[0130] 4. The present invention combines the designed path tracking controller with the designed stability controller, divides and judges the area using the phase plane method, and outputs the front wheel steering angle and additional yaw moment after cooperative control according to the cooperative control weight coefficient, so as to give full play to the potential of the vehicle stability system in the process of path tracking and improve the vehicle handling stability;

[0131] 5. The present invention can effectively solve the path tracking and stability problems of vehicles under extreme working conditions, improve the driving safety of vehicles, and thus reduce traffic congestion and accidents.

[0132] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0133] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A method for coordinated control of electric vehicle path tracking and stability, characterized in that: include: Step S1, based on the nonlinear characteristics of the tires of the four-wheel independent drive electric vehicle, construct a nonlinear ideal vehicle dynamics model based on lateral and yaw; Step S2, based on the nonlinear ideal dynamics model of the vehicle constructed in step S1, the ability of the long short-term memory network to predict the future state is connected with the model predictive control algorithm, a long short-term memory network prediction model of the nonlinear system is established, the state quantity, control quantity and objective function are determined, and the long short-term memory network prediction model is integrated into the model predictive control algorithm to complete the construction of the path tracking controller, and the front wheel steering angle is output through the path tracking controller; Step S3, based on the nonlinear ideal vehicle dynamics model constructed in step S1 and the front wheel turning angle output in step S2, establish the system tracking error of the center of mass sideslip angle and the yaw angle, introduce the additional yaw moment equation, and construct a stability controller; Step S4, based on the stability controller constructed in step S3, a multi-objective particle swarm optimization algorithm improved by a differential evolution mechanism is introduced, tire slip rate and energy utilization rate are determined as optimization targets, a multi-objective collaborative optimization torque distribution model is constructed, and an additional yaw moment is output through the torque distribution model; Step S5, connecting the front wheel steering angle output in step S2 with the BP neural network, training and outputting the phase plane inward and outward expansion coefficients and dividing the phase plane area, judging the area where the vehicle is located, and outputting the front wheel steering angle and additional yaw moment after cooperative control according to the cooperative control weight coefficient, and finally realizing the path tracking and stability cooperative control of the four-wheel independent drive electric vehicle; Step S5 specifically includes: Divide the phase plane region, where: The stable region satisfies the following equation: ； The critical stability region satisfies the following equation: ； The unstable region satisfies the following equation: ； in, is the sideslip angular velocity of the center of mass, is the sideslip angle of the center of mass, is the intercept of the stability boundary, , is the shrinkage coefficient of the stable region boundary, B 1. B 2 is the original boundary coefficient, is the intercept of the unstable boundary, , is the expansion coefficient of the stable region boundary; The shrinkage coefficient of the stable region boundary and the expansion coefficient of the stable region boundary , using a BP neural network model, the BP neural network model includes an input layer, a hidden layer, and an output layer, and the BP neural network model is trained and then solved. The specific steps are as follows: The input is set as the front wheel angle, longitudinal speed and road adhesion coefficient, and the number of hidden layer neurons is set to 64; then, the variable structure sigmoid function is used as the activation function, and the chain derivative rule is used to back propagate the error during the training process to achieve the purpose of updating the weights and biases. Finally, after the neural network is trained, the shrinkage coefficient of the stable region boundary is output. and the expansion coefficient of the stable region boundary ; Among them, the input layer matrix And the output layer matrix Defined as: ； in, is the road adhesion coefficient, is the front wheel turning angle, v x is the vehicle longitudinal speed; According to the results of dividing the phase plane area, the working areas of the path tracking controller and the stability controller are divided as follows: Only the path tracking area satisfies the following equation: ； The path tracking area and the stability control area coordinate control area satisfy the following formula: ； Only the stability control region satisfies the following equation: ； The coordinated switching function is further designed to use the hyperbolic tangent function, and the coordinated control weight coefficient is obtained q Satisfy the following formula: ； Finally, the front wheel steering angle and additional yaw moment after coordinated control satisfy the following equation: ； ； in, represents the front wheel steering angle after cooperative control, represents the additional yaw moment after cooperative control, represents the additional yaw moment.

2. The electric vehicle path tracking and stability coordinated control method according to claim 1, characterized in that: In step S1, based on the nonlinear characteristics of the tires of the four-wheel independent drive electric vehicle, a nonlinear ideal vehicle dynamics model based on lateral and yaw is constructed, and the expression is: ； ； The nonlinear ideal dynamics model of the vehicle is rewritten as a state space equation, which is expressed as follows: ； in, k f is the front wheel cornering stiffness, k r is the rear wheel cornering stiffness, m is the vehicle mass, is the distance from the center of mass to the front axle, b is the distance from the center of mass to the rear axle, is the yaw angular velocity, is the yaw angular acceleration, is the moment of inertia about the OZ axis.

3. The electric vehicle path tracking and stability coordinated control method according to claim 2, characterized in that: Step S2 specifically includes: Obtain vehicle status data, perform preprocessing and normalization; then use the deep learning toolbox to define and train the long short-term memory network prediction model; then, use the trained long short-term memory network prediction model to predict the future state of the vehicle; finally, integrate the long short-term memory network model into the model predictive control algorithm framework, complete the construction of the path tracking controller, and output the front wheel steering angle through the path tracking controller; Among them, the state update equation of the long short-term memory network prediction model is: ； ； ； ； ； in, Represents the input gate at the current moment; f t Represents the forget gate at the current moment; Represents the output gate at the current moment; The memory unit representing the current moment; h t Indicates the hidden state at the current moment; Indicates the hidden state at the previous moment; Represents the memory unit of the previous moment; represents the input vector; is the sigmoid activation function; tanh is the hyperbolic tangent function; Represents element-wise multiplication operation; W i , W f , W o , W c is the weight of the LSTM prediction model; b i , b f , b o , b c is the bias matrix of the LSTM prediction model; Based on the state space equation in step S1, the vehicle path tracking error model is obtained after coordinate transformation, and the expression is as follows: ； ； ； in, is the system state matrix The derivative of ; is the lateral error; is the lateral error derivative; is the heading angle error; is the heading angle error derivative; is the input; the front wheel angle As input quantity; and is the time-varying matrix of the linear system; Objective function of the path-following controller J for: ； in, For the prediction time domain, To control the time domain, is from i Time has come The actual trajectory of the vehicle output by the long short-term memory network prediction model within time, is from i Time has come The reference trajectory of the vehicle within time, is the weight matrix of vehicle tracking capability, is from i Time has come -1 moment control increment, is the weight matrix that controls the increment, is the weight coefficient, is the relaxation factor; The optimization problem of the path tracking controller is designed to minimize the objective function J ; In minimizing the objective function J During the process, the following conditions are met: ； ； ； in, To predict the control quantity in the time domain, is the maximum value of the control quantity in the prediction time domain, is the minimum value of the control quantity in the prediction time domain, is from i Time has come The maximum value of the control increment at time -1, is from i Time has come The minimum value of the control increment at time -1.

4. The electric vehicle path tracking and stability coordinated control method according to claim 3 is characterized in that: Step S3 specifically includes: The nonlinear ideal vehicle dynamics model constructed in step S1 is rewritten to obtain the following formula: ； The sliding surface Designed for: ； in, , is a positive coefficient; is the system tracking error derivative; ； in, is the system tracking error weight coefficient, ; is the ideal yaw rate; is the ideal center-of-mass sideslip angular velocity; is the center of mass sideslip angular velocity error; is the yaw rate error; Sliding surface Taking the second-order derivative of the expression, we can get: ； in, is the second-order derivative of the system tracking error, and are the actual center of mass sideslip angle acceleration and the ideal center of mass sideslip angle acceleration, is the ideal yaw angular acceleration; Then, the first-order derivative of the sliding surface is The function expression is: ； Introducing the exponential reaching law function, we get the following formula: ； Then we get: ； The additional yaw moment The expression is: ； in, is the symbolic function, and is a coefficient greater than 0; Using fuzzy control to adjust the system tracking error weight coefficient , the tracking errors of the fuzzy processed input variables sideslip angle and yaw angle are inferred through fuzzy control rules to obtain the stability controller.

5. The electric vehicle path tracking and stability coordinated control method according to claim 4, characterized in that: Step S4 specifically includes: Based on the stability controller constructed in step S3, the differential evolution search algorithm is introduced with tire slip rate and energy utilization as optimization targets to construct a torque distribution model for multi-objective collaborative optimization, where tire slip rate The calculation formula is: ； in, is the wheel radius, For the e The angular velocity of each wheel; Minimize the sum of the slip ratios It is expressed as: ； in, For the e The slip rate of each wheel; The energy utilization rate is determined based on the motor output power, and the expression is as follows: ； ； in, is the energy utilization rate, is the motor output power, Indicates e The input power of the motor corresponding to each wheel, For the e The weight value corresponding to each wheel is dt Indicates time t The differential of is the motor efficiency; The expression of the fitness function of the improved multi-objective particle swarm optimization algorithm is: ； in, J total is the overall fitness objective function, J 1 is the objective function of minimizing the slip rate, J 2 is the objective function of vehicle energy utilization, and They are J 1 and J A weight factor of 2; During the iteration process of the multi-objective particle swarm optimization algorithm, the differential evolution scaling factor and the differential evolution crossover probability are introduced. After each particle update, the differential evolution operation is performed to obtain the global optimal solution and distribute the torque to the four wheels.

Images