CN104401232B - Electric automobile torque optimization method based on data-driven PREDICTIVE CONTROL - Google Patents

Abstract

A kind of electric automobile torque optimization method based on data-driven PREDICTIVE CONTROL, belongs to electric vehicle engineering field。It is an object of the invention to can effectively carry out electric automobile drive and braking moment optimized distribution, to realize the electric automobile torque optimization method based on data-driven PREDICTIVE CONTROL of vehicular longitudinal velocity tracking and controlling method。The present invention is first against the suitable excited data of the dynamic properties design of system, to ensure the abundant excitation to system；Secondly, the inputoutput data that excitation obtains is constructed the prediction output equation of system；Then, it is considered to executor's firm constraints of motor, set of cells and brake, the cost function of Model Predictive Control Algorithm structure moment of torsion optimal control is utilized；Finally, corresponding by solving cost function optimal problem, it is thus achieved that control input and act on system, thus realizing the control to system。Having only to when obtaining excited data in control algolithm of the present invention needs some off-line simulations to test, and development cost is relatively low。

Description

Torque optimization method for electric vehicles based on data-driven predictive control

technical field

The invention belongs to the technical field of electric vehicles.

Background technique

With the increasing environmental pollution and energy crisis, pure electric vehicles have become a major direction for the development of the automotive industry due to their advantages such as high energy efficiency, low emissions, low noise, and energy recovery. The country has also issued a large number of preferential policies to support enterprise research. Electric vehicles, during the "Tenth Five-Year Plan" period, the National 863 Program "Major Science and Technology Projects for Electric Vehicles" received strong support from the state for two consecutive phases. After the efforts of many companies, colleges and research institutes, my country's research on pure electric vehicles has There have been many technological innovations and breakthroughs in the research and development of core components and the design of vehicle systems. Whether it is a traditional car or an electric car, the problem of torque optimization is involved. How to design a controller to reasonably generate driving and braking torque signals to achieve good performance of the vehicle has always been one of the important issues in the research of the automotive industry. For electric vehicles, due to its relatively simple structure, the control problem can be appropriately simplified. At the same time, the response speed of the motor is fast, and the torque and speed of the motor are easy to obtain, which provides a good basic condition for the torque optimization control of electric vehicles. For centralized electric vehicle torque optimization control, the main problems are as follows:

1. Since the research object is a centralized pure electric vehicle, which has only one motor and is driven by the front wheels, there is no distribution optimization problem, and the torque distribution optimization design cannot be carried out like the four-wheel hub motor driven electric vehicle. Consider the generation of driving/braking torque in the vehicle controller.

2. The mechanism model of electric vehicle torque optimization control system is complex, nonlinear and difficult to model. At the same time, the output driving and braking torque signals of the controller must meet the rigid constraints of the actuator motor and brake, that is, the output driving and braking torque signals of the controller cannot exceed the actual maximum output torque of the motor and the maximum braking torque of the brake .

3. Electric vehicles need a power source. Lithium battery packs are commonly used now to supply power to the motor. The power supply voltage of the motor also affects the maximum output torque of the motor. Therefore, the output voltage of the battery pack must also be considered when considering the maximum output torque of the motor. Impact.

Contents of the invention

The purpose of the present invention is to effectively optimize the distribution of electric vehicle driving and braking torque, so as to realize the data-driven predictive control-based electric vehicle torque optimization method of the vehicle longitudinal speed tracking control method.

The steps of the present invention are:

a. Construction of centralized electric vehicle simulation model: A modularized method is used to build a centralized electric vehicle vehicle model in AMESim. The entire electric vehicle simulation model mainly includes electric drive modules, transmission modules, tire modules and vehicle longitudinal dynamics. The vehicle model parameters are shown in Table 1:

Table 1 Electric Vehicle Parameters

;

b. Torque optimization data-driven predictive controller design: The data-driven predictive control algorithm adopted is composed of subspace identification and model predictive control. The specific steps are as follows:

①Selection of excitation data: select the driving torque command and braking torque command as the excitation input signal, in the AMESim vehicle model, give the designed driving torque and braking torque data to the motor and brake, and run the model in an open loop to obtain the measurement The actual vehicle speed, the driving torque command, the braking torque command and the measured vehicle speed are the final required system excitation data; according to these data, the input Hankel matrix Up and Uf of the system, and the output Hankel matrix Yp and Yf are obtained:

;

②Prediction equation:

The predicted output equation of the system constructed by inputting and outputting the Hankel matrix is:

, (1)

That is, the future output value of the system estimated by the subspace identification method, by solving the least squares problem, the two prediction matrix coefficients Lw and Lu in the prediction equation are obtained

(2)

The solutions of the two prediction matrix coefficients Lw and Lu are obtained by the following equations:

(3)

After the Hankel matrix of the input and output information of the system is known, and Lw and Lu are obtained, the future output value of the system can be obtained by using the subspace prediction equation (1);

The final output prediction equation is as formula (4), is the Hankel matrix first column of

, (4)

③Data-driven predictive controller design: The optimized driving torque and braking torque signals are obtained by solving the target cost function. The mathematical expression of the cost function is as follows:

(5)

Rewrite the prediction equation (4) output by the system into an incremental form:

; (6)

The following model prediction output equation is obtained by accumulating the incremental prediction output:

(7)

in,

;

The F term is the free response of the system, and the corresponding S?uf term is called the control response of the system;

The cost function of the following form is constructed by using the incremental prediction output model:

(8)

in,

, ; (9)

Among them, Γu and Γy are the weighting matrices of input items and output items respectively; the control time domain is Nu, and the prediction time domain is Np; the weighting coefficient Λ=Γu/Γy, the selected Λ value is:

The maximum torque constraint of the motor, the maximum braking torque constraint of the brake and the constraint of the actual vehicle speed in the torque optimization system are described by the following mathematical expressions:

(10).

The invention mainly aims at the torque optimization problem of the centralized front-wheel drive electric vehicle, and utilizes a data-driven predictive control algorithm to design a controller to realize the optimization of the driving and braking torque of the electric vehicle. The data-driven predictive control algorithm is a perfect combination of data-driven subspace identification and model predictive control. It directly uses the input and output data collected offline and the characteristics obtained after data processing to design the controller, effectively avoiding complex mechanism construction. mold. Through the collected input and output data, the torque prediction equation can be obtained, and then the cost function is constructed by using the model predictive control method, and the constraints are considered, and the optimal solution is obtained to obtain the optimized driving and braking torque command signals. The present invention can well solve the three problems of the prior art, and the data-driven predictive control algorithm can effectively avoid complex system mechanism modeling, and the torque optimization controller designed by the present invention can effectively avoid complex system mechanism modeling, and at the same time Consider actuator hard constraints for electric motors, battery packs, and brakes. The beneficial effects are:

1. Most of the traditional control algorithms are based on the mechanism model of the control system, but in the actual torque optimization process, the mechanism modeling is complicated and the workload is heavy. The data-driven predictive control algorithm designs the controller directly through the input and output data collected offline and the characteristics obtained through data processing, effectively avoiding complex mechanism modeling and eliminating the intermediate solution of the state space equation of the model. On the one hand, it reduces the amount of calculation, and on the other hand, it avoids the problem of model error introduced from data identification to state space model.

2. The torque optimization controller system designed in the present invention is a nonlinear system, and takes into account the rigid constraints of the actuators of the motor, battery pack and brake. The traditional control algorithm cannot effectively deal with the constraints of the system, and the data-driven prediction The control algorithm can effectively deal with the multi-input and multi-output control problems with constraints. When solving, the constraints are directly compiled into the S_function in simulink for online solution.

3. The development process of traditional automotive electronic control units requires a lot of experience and calibration tests. In the control algorithm of the present invention, only some off-line simulation experiments are needed when obtaining the excitation data, and the development cost is relatively low.

Description of drawings

Fig. 1 is the electric vehicle torque optimization control block diagram based on data-driven predictive control according to the present invention;

Fig. 2 is the centralized electric vehicle AMESim vehicle model of the present invention;

Fig. 3 is the expected vehicle speed under the NEDC working condition of the present invention, and the unit is m/s, and the abscissa is time, and the unit is s;

Fig. 4 is a block diagram of the data-driven predictive control algorithm used in the present invention;

Fig. 5 is the excitation data selected by the design torque optimization controller in the present invention, and the excitation input is driving and braking torque command, unit Nm; The output data obtained off-line is actual vehicle speed, unit is m/s;

Fig. 6 is the identification result of a section of excitation data arbitrarily extracted from the excitation data in order to verify the prediction model of the present invention, the abscissa is time, and the unit is s;

Fig. 7 is the identification result obtained by redesigning a group of excitations in order to verify the prediction model in the present invention, the abscissa is time, and the unit is s;

Fig. 8 is the torque optimization data-driven predictive controller that the present invention builds in simulink, mainly realizes the data-driven predictive control algorithm by the M file compiling program in Matlab;

Fig. 9 is the comparison between the actual vehicle speed and the expected vehicle speed after the action of the controller under the initial parameters of the present invention, wherein the dotted line indicates the actual vehicle speed, and the solid line indicates the expected vehicle speed, the unit is m/s, and the abscissa is time, the unit s;

Figure 10 is the comparison between the actual vehicle speed and the expected vehicle speed after the action of the controller under the condition that only the slope is changed and other parameters remain unchanged according to the present invention, wherein the dotted line indicates the actual vehicle speed, and the solid line indicates the expected vehicle speed, and the units are m/s, the abscissa is time, the unit is s;

Figure 11 is the comparison between the actual vehicle speed and the expected vehicle speed after the action of the controller under the condition that only the vehicle load is changed and other parameters remain unchanged according to the present invention, wherein the dotted line indicates the actual vehicle speed, and the solid line indicates the expected vehicle speed, the unit Both are m/s, the abscissa is time, the unit is s;

Fig. 12 is the comparison between the actual vehicle speed and the expected vehicle speed after the action of the controller under the condition that only the wind speed is changed and other parameters remain unchanged according to the present invention, wherein the dotted line indicates the actual vehicle speed, and the solid line indicates the expected vehicle speed, and the units are m/s, the abscissa is time, the unit is s.

detailed description

The research method of the present invention is based on data-driven model predictive control, comprising the following steps:

Firstly, appropriate excitation data is designed for the dynamic characteristics of the system to ensure sufficient excitation of the system; secondly, the input and output data obtained by the excitation are used to construct the predictive output equation of the system; then, the actuators of the motor, battery pack and brake are considered With hard constraints, the model predictive control algorithm is used to construct the cost function of torque optimization control; finally, by solving the optimal problem corresponding to the cost function, the control input is obtained and acts on the system, so as to realize the control of the system.

In order to solve the problems of the technologies described above, the present invention is achieved by adopting the following technical solutions:

The electric vehicle torque optimization based on data-driven predictive control described in the present invention is realized through joint simulation of software systems. The software system consists of Matlab/Simulink and AMESim advanced simulation software. Wherein the AMESim software is a commercial complex system modeling and simulation platform, and its main function in the present invention is to provide a high-fidelity centralized motor type pure electric vehicle model. In the simulation experiment, this model replaces the real vehicle as the implementation of the control method The object; MATLAB/Simulink software is used to build the simulation model of the controller, that is, to complete the construction of the torque optimization controller based on data-driven predictive control through Simulink programming.

Functionally speaking, the present invention may include the following parts: a controller simulation module and a centralized pure electric vehicle model. The role of each part is explained in detail below:

The main role of the centralized pure electric vehicle model in the design of the controller is to obtain the input and output excitation data that can reflect the dynamic characteristics of the system through offline simulation, and the controller constructs the predictive output equation of the system based on these data.

The main function of the controller module is to collect various state information of the electric vehicle model, such as vehicle speed, battery output voltage, motor output torque, etc., and then perform table lookup and calculation, and produce control signals—motor drive torque signals and mechanical control signals. Power torque signal and send the signal to the actuator of the electric vehicle - the motor and the traditional mechanical brake mechanism;

The electric vehicle model includes several parts such as the electric drive system model, the drive shaft model, the vehicle dynamics model, and the tire model. Its function is to generate corresponding driving or braking torque according to the control signal given by the torque optimization controller, and change the dynamics of the vehicle—motor output torque, mechanical braking torque, vehicle speed, etc., to achieve good performance of the vehicle.

The control block diagram of the implementation of the electric vehicle torque optimization method based on data-driven predictive control in the present invention is shown in Figure 1, the torque optimization controller in the figure is built in Simulink, the input of the controller is the desired vehicle speed, and the actual vehicle speed is Feedback to the controller in real time, Tmax is the maximum driving torque of the motor, which is determined by the mechanical characteristics of the motor and the output voltage of the battery. Time increases the voltage to reduce the impact on vehicle performance. The driving torque obtained by the vehicle controller must be less than or equal to Tmax, so Tmax is given to the controller as a constraint. The centralized electric vehicle model in the figure is built in AMESim to simulate the operation of the real vehicle. The driving and braking torque signals optimized by the controller are respectively sent to the motor and braking module to control the operation of the vehicle, and the actual speed of the vehicle is fed back to the controller as a feedback signal.

The control objective of the present invention is that the torque optimization controller compares the actual vehicle speed and the expected vehicle speed signal fed back in real time, and under the premise of satisfying the constraint conditions, optimizes the driving and braking torque signals Tm and Tb, and gives them to the vehicle model The motor and brake module control the operation of the vehicle, and finally let the actual speed track the desired speed.

The present invention provides a set of devices based on the above operation principle and operation process. That is, a PC-based off-line electric vehicle torque optimization design test platform. The build and run process is as follows:

1. Software selection

The simulation models of the controlled object and the controller of the control system are built by the software Matlab/Simulink and AMESim respectively. The software versions are MatlabR2009a and AMESimR10 respectively, and the solvers are ode3 and Euler respectively. The simulation step size is a fixed step size, and the step size is selected as 0.1s.

2. Co-simulation settings

In order to realize the co-simulation of Matlab/Simulink and AMESim, the environment variables of the PC must be set according to the requirements first, so that the two are related to each other. Then add an interface module for communicating with simulink in the AMESim interface, and connect the variables that need to communicate between Matlab/Simulink and AMESim to this module; finally, after system compilation, the model information in AMESim is retained in Simulink in the form of S-function , so as to realize the joint simulation and communication of the two. When running the Simulink simulation model, the AMESim model is also being calculated and solved at the same time. During the simulation process, data is exchanged between the two continuously. If the model structure or parameter settings in AMESim are modified, recompilation is required. It is worth noting that the simulation step size of the two must be the same.

Detailed steps of the present invention are:

a. Construction of centralized electric vehicle simulation model: A modularized method is used to build a centralized electric vehicle vehicle model in AMESim, as shown in Figure 2. The whole electric vehicle simulation model mainly includes electric drive module, transmission module, tire module and vehicle longitudinal dynamics. The parameters of the whole vehicle model are shown in Table 1:

Table 1 Electric Vehicle Parameters

;

The electric drive system provides power for the vehicle, including the battery part and the motor part. The battery is mainly used to supply power to the motor. The battery part of a pure electric vehicle is composed of multiple single batteries connected in series and parallel. Lithium battery packs are commonly used. The terminal voltage output by the battery pack is the sum of the output voltage of a single battery. The output terminal voltage of the battery system is the voltage provided by the battery pack to the motor. In this model, the rated terminal voltage of the battery system is 320V; the permanent magnet synchronous motor is used in the present invention , the control method of the motor is the vector control method, and the maximum output torque of the motor is 300Nm.

The transmission module mainly includes the transmission and the drive shaft. The torque output by the motor passes through the transmission to produce different speed ratios through different gear radii to reduce the speed and increase the output torque. The transmission ratio here is 5:1, that is, the speed becomes 1/5 of the original, and the torque increases to 5 of the original. times, just to make up for the high speed of the motor itself and the problem of insufficient torque. The present invention focuses on the linear running of the vehicle, the output torque and rotational speed of the left and right wheels are the same without differential, and thus there is no differential.

There are four tire models in the picture, and the tires use the MagicFormula magic formula model.

The vehicle longitudinal dynamics part, which considers the driving force, braking force and driving resistance of the vehicle during driving, where the driving resistance includes air resistance, rolling resistance and frictional resistance. In this module, parameters such as the overall quality of the vehicle, slope, and wind speed can be set.

The cycle condition selected by the model is the cycle condition NEDC (New European Driving Cycle) of European light-duty vehicles. This condition contains several times of stopping and constant speed movement for a long time, and the starting and braking curves are relatively smooth. It reflects the steady-state performance of the vehicle. This working condition simulates the driving state of the car in urban road conditions, and has strong practicability. The EUDC driving cycle takes 1200 seconds. The expected speed of the vehicle under this condition is shown in Figure 3, which is the expected speed of the torque optimization controller.

b. The control target of the controller in the present invention is to meet the speed requirements of the NEDC working condition while optimizing the torque, and at the same time, the driving torque and braking torque should meet the constraints of the maximum torque of the motor and the battery in practice. These constraints are in Embodied in constraint handling when designing data-driven predictive controllers. The design of the torque-optimized data-driven predictive controller of the present invention: the adopted data-driven predictive control algorithm is composed of the combination of subspace identification and model predictive control, and its composition schematic diagram is shown in Figure 4. The main difference between this method and other control methods is that it does not need to use the traditional transfer function or state space model to design the controller, but to use the subspace linear prediction equation to predict the future output value of the system. The coefficients of the subspace prediction equation are directly passed The off-line input and output data of the system are calculated, based on the prediction equation, combined with the basic principle of the model predictive control algorithm, a data-driven predictive control method with constraints is obtained. The specific steps are as follows:

① Selection of excitation data: Before designing the controller, it is first necessary to design appropriate excitation data for the dynamic characteristics of the vehicle model built earlier, and then obtain the system prediction output equation based on these excitation data. Here, the driving torque command and the braking torque command are selected as the excitation input signals. In the AMESim vehicle model, the designed driving torque and braking torque data are given to the motor and brake, and the model is run in an open loop to obtain the measured actual vehicle speed. Here The simulation step size is 0.1s. The driving torque command, braking torque command and the measured vehicle speed are the final required system excitation data, as shown in Figure 5. In the selection of excitation, since the driving torque range is: -300Nm-300Nm, the braking torque signal range: 0-1, and the vehicle speed range: 0-40m/s, in order to ensure the accuracy of the excitation model, the excitation data is normalized Normalization, the normalization coefficient of driving torque is 1/100, the normalization coefficient of braking torque is 1, and the normalization coefficient of vehicle speed is 1/40. It is worth noting that before the controller gives the driving and braking torque commands after the optimization solution to the vehicle model, denormalization is required. After getting the excitation data, save it as a data file, and get the input Hankel matrix Up, Uf of the system according to these data, and output the Hankel matrix Yp, Yf:

The subscripts p and f of the matrix represent "past" and "future" respectively. In order to ensure that the data can contain sufficient information about the state of the system, the number of columns of the Hankel matrix often needs to be large enough. For a single-input single-output system, the element in the Hankel matrix is a value, while for a multiple-input multiple-output system, the element in the Hankel matrix is a column matrix of the corresponding dimension. The input of the present invention is the driving torque command and the braking torque command, and the output is the vehicle speed, so the elements in the input Hankel matrix Up, Uf are 2-dimensional column matrices, and the elements in the output Hankel matrix Yp, Yf are a numerical value;

.

② Prediction equation and its verification:

The predictive output equation of the system constructed by the input and output Hankel matrix obtained above is:

, (1)

That is, the future output value of the system estimated by the subspace identification method, by solving the least squares problem, the two prediction matrix coefficients Lw and Lu in the prediction equation are obtained

(2)

The solutions of the two prediction matrix coefficients Lw and Lu are obtained by the following equations:

(3)

After the Hankel matrix of the input and output information of the system is known, and Lw and Lu are obtained, the future output value of the system can be obtained by using the subspace prediction equation (1); in the actual application of the controller, only the Hankel matrix is output The first column of is used as the estimated value of the system in the future, which on the one hand greatly reduces the amount of calculation, and on the other hand prepares for the application of model predictive control. Therefore, the final output prediction equation is as in formula (4), is the Hankel matrix first column of

, (4)

The predictive output model of the system is obtained according to the excitation data. For the data-driven predictive control algorithm, whether the predictive model can accurately predict the future output of the system determines the quality of the control performance. Next, the present invention verifies the prediction equation obtained through several sets of simulation experiments.

Figure 6 is the identification result of a section of incentive data randomly extracted from the incentive data. It can be seen that the predicted system output is very close to the actual system output. In addition, we redesigned a set of incentives as shown in Figure 7, and the predicted system output can also track the actual system output well. This shows that the accuracy of the designed prediction model meets the requirements.

③Data-driven predictive controller design: After obtaining the predictive output equation of the system output, predictive control algorithm based on the model. The optimized driving torque and braking torque signals are obtained by solving the objective cost function, and the mathematical expression of the cost function is as follows:

(5)

It can be seen from the predictive control theory that the incremental predictive controller is used, which can make the controller have an integral function, thereby reducing the steady-state error of the system and improving the control effect. Therefore, the prediction equation (4) output by the system is rewritten as an incremental form:

; (6)

The following model prediction output equation is obtained by accumulating the incremental prediction output:

(7)

in,

;

The F term is the free response of the system, and the corresponding S?uf term is called the control response of the system;

The cost function of the following form is constructed by using the incremental prediction output model:

(8)

in,

, ; (9)

Wherein, Γu and Γy are the weighted matrices of the input and output items respectively; the control time domain is Nu, and the prediction time domain is Np; in the present invention, after a large number of experimental simulations in the later stage, the selected prediction time domain is 20, The control time domain is 4. The reference value Ref here is the expected vehicle speed v_q of the NEDC cycle condition, which is normalized and updated to the controller in real time. The weighting coefficient Λ=Γu/Γy, after a large number of simulation experiments and adjustments, the selected value of Λ is:

The maximum torque constraint of the motor, the maximum braking torque constraint of the brake and the constraint of the actual vehicle speed in the torque optimization system are described by the following mathematical expressions:

(10).

The final torque-optimized data-driven predictive controller model is shown in Fig. 8, mainly through Matlab

The M file compiler in the program is implemented. Through the joint simulation technology of AMESim and Simulink, after debugging, the vehicle speed simulation result obtained under the initial parameters is shown in Figure 9. It can be seen from the figure that the actual vehicle speed basically coincides with the expected vehicle speed, and the control effect is good.

Experimental verification

In order to further verify the control performance of the torque-optimized data-driven predictive controller, the performance of the torque-optimized controller was tested under the three conditions of changing the slope, vehicle load, and wind speed in the centralized electric vehicle model, respectively, while other parameters remained unchanged. verify.

1) Different slopes

The climbing ability requirement of electric vehicles is generally greater than or equal to 20% (that is, 9 degrees). Through experiments, it is known that the maximum torque of the motor directly affects the climbing performance (the climbing performance here refers to the ENDC desired speed that can be achieved under this slope. Tracking), when the maximum torque is 100Nm, the climbing ability is 10%, and the speed tracking cannot be realized after exceeding 10%; when the maximum torque is 200Nm, it is 15%, and the speed tracking cannot be realized after exceeding 15%; the maximum slope can not be realized when the maximum torque is 300Nm Can meet the 20% requirement. In the present invention, the maximum torque of the motor is set to 300Nm, and other quantities are kept unchanged, the wind speed is 0m/s, and the mass of the vehicle is 1000kg. Only the slope is changed, and they are respectively set to 5%, 10%, and 20%. The obtained simulation results are shown in the figure Ten shown.

It can be seen from the figure that due to the slope, the vehicle has a tendency to slide when it starts to start on a hill. The greater the slope, the greater the tendency to slide. In practice, it is necessary to step on the brakes when starting on a hill, so in the model this is reasonable. As the slope increases, the overall speed tracking effect is still very good. However, due to the rigid condition of the actuator with the maximum torque of the motor, when the slope is very large, the vehicle speed cannot continue to increase after reaching the upper limit. When the speed is high, the vehicle speed can only be maintained at the upper limit and cannot continue to track, but the speed tracking below the upper speed limit works well. Among them, the upper limit of speed is about 18m/s (64.8km/h) at a slope of 20%, which is also reasonable in reality. It shows that the torque optimization data-driven predictive controller still has good control performance under different slope conditions.

2) Different vehicle loads

Keep other quantities unchanged, set the slope to 0, the wind speed to 0m/s, and only change the mass of the vehicle m, assuming that the average weight of passengers is 75kg, set the mass of the vehicle to 1500kg, (1500+2*75)kg respectively , (1500+4*75)kg, the obtained speed curve is shown in Figure 11.

It can be seen from the figure that the overall control effect is very good under these masses, but as the mass increases, the deviation will increase, and there is some overshoot when the mass is (1500+4*75) kg, but the overall tracking effect is very good. Well, it shows that the data-driven rolling optimization vehicle controller still has good control performance under different quality conditions.

3) Different wind speeds

Keeping other variables unchanged, set the slope to 0, the mass of the vehicle to 1000kg, and only change the ambient wind speed, which is set to 10m/s, 20m/s, and 30m/s respectively. The obtained vehicle speed comparison curves are shown in Figure 11. It can be seen from the figure that after changing the ambient wind speed to 30m/s, the tracking effect of the vehicle speed is still very good, which shows that the torque optimization data-driven predictive controller still has good control performance under different wind speeds.

The present invention designs a torque optimization controller based on data-driven predictive control for centralized electric vehicles. This method cleverly combines data-driven subspace identification and model predictive control methods, and obtains the system's performance based on the input and output data of the system. The predictive equations are then optimized by model predictive control to obtain the control input of the system, while explicitly considering the constraints of the system. In order to verify the effectiveness of the torque optimization controller, a centralized front-wheel drive electric vehicle model was built in the AMESim advanced simulation software, and the performance of the controller was verified under different slopes, different vehicle loads, and different wind speed conditions. The simulation results show that the data-driven predictive controller vehicle controller still has good control performance under the conditions of different masses, different slopes, and no ventilation speed.

Claims (1)

1. A method for optimizing electric vehicle torque based on data-driven predictive control, characterized in that: its steps are: a. Construction of centralized electric vehicle simulation model: A modularized method is used to build a centralized electric vehicle vehicle model in AMESim. The entire electric vehicle simulation model mainly includes electric drive modules, transmission modules, tire modules and vehicle longitudinal dynamics. The vehicle model parameters are shown in Table 1: Table 1 Electric Vehicle Parameters ; b. Torque optimization data-driven predictive controller design: The data-driven predictive control algorithm adopted is composed of subspace identification and model predictive control. The specific steps are as follows: ①Selection of excitation data: select the driving torque command and braking torque command as the excitation input signal, in the AMESim vehicle model, give the designed driving torque and braking torque data to the motor and brake, and run the model in an open loop to obtain the measurement The actual vehicle speed, the driving torque command, the braking torque command and the measured vehicle speed are the final required system excitation data; according to these data, the input Hankel matrix Up and Uf of the system, and the output Hankel matrix Yp and Yf are obtained: ; ②Prediction equation: The predicted output equation of the system constructed by inputting and outputting the Hankel matrix is: , (1) That is, the future output value of the system estimated by the subspace identification method, by solving the least squares problem, the two prediction matrix coefficients Lw and Lu in the prediction equation are obtained (2) The solutions of the two prediction matrix coefficients Lw and Lu are obtained by the following equations: (3) After the Hankel matrix of the input and output information of the system is known, and Lw and Lu are obtained, the future output value of the system can be obtained by using the subspace prediction equation (1); The final output prediction equation is as formula (4), is the Hankel matrix first column of , (4) ③Data-driven predictive controller design: The optimized driving torque and braking torque signals are obtained by solving the target cost function. The mathematical expression of the cost function is as follows: (5) Rewrite the prediction equation (4) output by the system into an incremental form: ; (6) The following model prediction output equation is obtained by accumulating the incremental prediction output: (7) in, ; The F term is the free response of the system, and the corresponding S?uf term is called the control response of the system; The cost function of the following form is constructed by using the incremental prediction output model: (8) in, , ; (9) Among them, Γu and Γy are the weighting matrices of input items and output items respectively; the control time domain is Nu, and the prediction time domain is Np; the weighting coefficient Λ=Γu/Γy, the selected Λ value is: The maximum torque constraint of the motor, the maximum braking torque constraint of the brake and the constraint of the actual vehicle speed in the torque optimization system are described by the following mathematical expressions: (10).