Vehicle transverse and longitudinal cooperative tracking method based on difference innovation search algorithm
By optimizing the weighted matrix of the LQR controller through a differential innovation search algorithm and combining it with a prediction module, the problem of insufficient tracking accuracy and stability of traditional LQR controllers on complex roads is solved, enabling efficient and safe autonomous driving of vehicles in complex environments.
Patent Information
- Application Number
- CN202411050044.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-01
- Publication Date
- 2024-11-22
Smart Images

Figure CN118991819A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of vehicle lateral and longitudinal coordinated tracking, and in particular relates to a vehicle lateral and longitudinal coordinated tracking method based on a difference innovation search algorithm. Background Art
[0002] In recent years, with the development of smart car technology, artificial intelligence algorithms have played an increasingly important role in the automotive field. In order to improve the safety and intelligence level of vehicle driving, auxiliary driving functions such as adaptive cruise control (ACC), automatic parking and forward collision warning (AEB) have been continuously developed;
[0003] However, since the weighting matrix in traditional control algorithms is generally assigned using expert prior knowledge, the accuracy and stability of the controller are greatly reduced; therefore, controllers based on intelligent optimization algorithms are being developed to find the optimal weighting matrix and improve the robustness and accuracy of the vehicle during driving; among them, the linear quadratic regulator (LQR) is a crucial component in the control algorithm, but it is difficult to deploy it in low-cost autonomous driving applications; in the linear quadratic regulator (LQR), an intelligent optimization algorithm is used to optimize its weighting matrix, but it is still difficult to design a controller that can optimize the weighting matrix in real time. Summary of the invention
[0004] The purpose of the present invention is to provide a vehicle lateral and longitudinal collaborative tracking method based on a difference innovation search algorithm, so as to solve the problems existing in the background technology.
[0005] In order to achieve the above technical objectives, the technical solution adopted by the present invention is as follows:
[0006] A vehicle lateral and longitudinal collaborative tracking method based on a difference innovation search algorithm, characterized in that it includes the following steps:
[0007] S1: Through the state observer, the vehicle information in front is detected in real time, and the real-time vehicle data is obtained by the lateral and longitudinal control modules;
[0008] S2: Input the real-time data into the difference innovation search algorithm DCS algorithm module to optimize the gain parameters in the weighting matrix of the LQR controller;
[0009] S3: Based on the optimized gain parameters, the decision control module is used to adjust the lateral and longitudinal movements of the vehicle so that the target vehicle can track the predetermined trajectory.
[0010] Preferably, the state observer comprises a sensor component, a data acquisition unit and a data processing unit;
[0011] Sensor component: The sensor is the input part of the state observer and is used to collect real-time data of the vehicle. The real-time location information of the vehicle is obtained through the global positioning system GPS. The inertial measurement unit IMU is used to measure the real-time acceleration and angular velocity of the vehicle to determine the speed and posture of the vehicle. The wheel encoder is used to measure the rotation speed and direction of the wheel to assist in determining the speed and position of the vehicle. The vehicle bus system provides status information of other subsystems of the vehicle.
[0012] The data acquisition unit is responsible for collecting data from sensors and converting it into the corresponding processing format;
[0013] The data processing unit is the core part of the state observer, which processes the sensor data and estimates the vehicle state by using algorithms.
[0014] Preferably, the longitudinal controller is mainly divided into: an upper controller and a lower controller; the upper controller, as a decision-making module, obtains the expected acceleration or expected torque according to the position deviation, i.e., the speed compensation, the actual speed and the acceleration closed loop; the position controller mainly calculates the position error calculated by the lateral controller to obtain the speed compensation; the speed compensation, the actual speed and the expected speed obtained by the position controller are sent to the speed controller to obtain the acceleration compensation;
[0015] The lower controller, as an execution module, compares the obtained speed and acceleration with the accelerator and brake calibration table, and then converts the data into the throttle opening of the vehicle's accelerator and brake, thereby controlling the acceleration or deceleration of the vehicle; its logical formula is as follows:
[0016]
[0017] Where: thr is the throttle valve opening; T max is the maximum torque of the motor; rpm is the speed; the maximum speed of the car is assumed to be 4523.
[0018] Preferably, the differential innovation search algorithm DCS algorithm module includes:
[0019] Divide team members into three categories: high performers: excel in divergent thinking and self-driven learning; average performers: use convergent thinking to shape various ideas into feasible solutions; low performers: bring diversity to the team, promote exploration and increase the idea pool;
[0020] Different team members use different learning plans;
[0021] Retrospective evaluations are used to monitor improvements among performers and identify top performers; they provide data for decision making and drive performance improvements across the team;
[0022] The specific steps are as follows:
[0023] S2.1: Initialize the population and set the upper and lower bounds and number of iterations. The initialization formula is as follows:
[0024] x i,d =LB d +U(0,1)*(UB d -LB d )
[0025] Where U(0,1) represents a random number in the interval (0,1), LB d and UB d Lower and upper bounds on the dth dimension specified for the optimization problem;
[0026] S2.2: Use the differentiated knowledge acquisition rate to divide the population into high performers, average performers, and low performers. The formula for calculating the differentiated knowledge acquisition rate is as follows:
[0027]
[0028] in is the variable of individual i at the tth iteration The specific calculation method of the coefficient value is as follows:
[0029]
[0030] Where R i,t is the rank of the i-th individual at the beginning of the t-th iteration, and NP is the population size;
[0031] S2.3: Use creative realism to encourage high performers to adopt divergent thinking, average performers to adopt convergent thinking, and low performers to adopt team diversity;
[0032] The formula for convergent thinking is as follows:
[0033] v i,d =w*x best,d +λ t *(x r2,d -x i,d )+ω i,t *(x r1,d -x i,d )
[0034] where x best,d is the dth position of the best performing individual in the current iteration, w is the best cognitive weight, the default value is 1, ω i,t is the ω coefficient value of individual i at the tth iteration, x r1,d is a random number in {1,2,…,NP} and X r1 ≠X≠Xbest , x r2,d is a random number in {ngs+1,…,NP} and X r2 ≠X r1 ≠X≠X best , ngs is the number of high achievers; λ t is the coefficient value λ of the individual at the tth iteration, and its expression is as follows:
[0035]
[0036] Among them, NFE t Indicates the number of times the current function is evaluated at time t, NFE max Indicates the maximum number of function calculations;
[0037] The formula for divergent thinking is as follows:
[0038] v i,d =x r1,d +Lk(α,σ)
[0039] where x r1,d is a random number in {1,2,…,NP} and X r1 ≠X≠X best , Lk(α,σ) represents the random number generator of Linnik distribution with control parameters α and σ;
[0040] The formula for team diversity is as follows:
[0041] V NP =LB+U(0,1)*(UB-LB)
[0042] Among them, U(0,1) represents a random number in the interval (0,1), L and U are the upper and lower bounds of the optimization problem;
[0043] S2.4: Retrospective Assessment RA is used to establish evaluation criteria as a benchmark for measuring success. RA determines a strategy for continuous improvement so that the team can learn from experience and then optimize collaboration; its main equation is as follows:
[0044]
[0045] Preferably, the LQR controller includes predictive feedback control, front wheel angle compensation and gain adjustment units;
[0046] Predictive feedback control, which is used to calculate control inputs based on the deviation between the current state and the predetermined trajectory;
[0047] Front wheel angle compensation: State feedback cannot make the steady-state error of the system as close to zero as possible. The introduction of front wheel angle compensation can further reduce the steady-state error.
[0048] The gain adjustment unit adjusts the control input according to the gain parameter provided by the difference innovation search algorithm DCS algorithm module.
[0049] Benefits of the present invention: The present invention obtains the vehicle's front information in real time through a state observer, and performs real-time detection on the front information to obtain real-time data, inputs the real-time data into a lateral controller for extraction to obtain characteristic information, and controls the vehicle based on the characteristic information. The present scheme adopts a differential innovation search algorithm DCS to optimize the weighting matrix of the LQR lateral tracking controller; the differential innovation search algorithm DCS algorithm can be adaptively adjusted according to the actual system operation conditions and feedback information; by updating the weighting matrix in real time, the controller can adapt to the dynamic changes and uncertainties of the system. Using the difference innovation search algorithm DCS to optimize the weighted matrix can improve the response speed, stability and robustness of the system and further improve the control performance; it also makes the controller have a certain tolerance for model errors, parameter changes and environmental changes; this can enhance the stability and reliability of the system and reduce the excessive response and unstable behavior of the controller; in actual driving, the driver can judge the things in front of him through his eyes and brain, and can avoid the factors that may cause danger to a certain extent; when the driver sees that the future path planning is not a straight line, he will turn the steering wheel in advance to change the front wheel angle; however, the algorithm control is lagging, and when encountering special road conditions, it may cause repeated regulation of the front wheel angle, thereby causing the vehicle to become unstable and the safety factor to be greatly reduced; this paper adds a prediction module, which is mainly responsible for analyzing and predicting the behavior of other vehicles, pedestrians, obstacles and other traffic participants in the surrounding environment, so as to better plan and execute automated driving decisions; the prediction module provides prediction results to the path planning module and the decision module to help generate more reliable, safe and efficient driving paths and decision strategies; based on the prediction of the behavior of other traffic participants, the automatic driving system can make reasonable decisions, such as overtaking, following, changing lanes, etc. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] The present invention is further illustrated by means of the non-limiting examples given in the accompanying drawings.
[0051] Figure 1 A flow chart of a vehicle lateral and longitudinal collaborative tracking method based on a difference innovation search algorithm of the present invention;
[0052] Figure 2 is a control flow chart of the present invention;
[0053] Figure 3 It is a two-degree-of-freedom dynamic model;
[0054] Figure 4 is the vehicle tracking error model;
[0055] Figure 5 It is a schematic diagram of the structure of the prediction module;
[0056] Figure 6 It is a schematic diagram of the structure of the LQR module;
[0057] Figure 7 This is a schematic diagram of the DCS optimization module;
[0058] Figure 8 It is the longitudinal control module. DETAILED DESCRIPTION
[0059] In order to enable those skilled in the art to better understand the present invention, the technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.
[0060] In order to solve the problem that the existing LQR controller has poor tracking accuracy and stability on complex roads under fixed weights, a vehicle lateral and longitudinal cooperative tracking method based on a difference innovation search algorithm is provided in the present invention, comprising a path planning module, an LQR prediction feedforward module, a feedback compensation module, a DCS optimization module and a decision control module, wherein the path planning module, the LQR prediction feedforward module, the feedback compensation module, the DCS optimization module and the decision control module are connected in sequence;
[0061] The path planning module plans the path for the starting point and the end point, and calculates the path information in real time to obtain the road information; the LQR prediction feedforward module designs the LQR controller based on the real-time data; the feedback module is used to reduce the steady-state error of the LQR prediction feedforward module to obtain an LQR controller that is as close to zero as possible; the DCS optimization module is used to optimize the weighting matrix of the LQR controller; the decision control module controls the vehicle based on the integrated data;
[0062] The following steps are involved:
[0063] By calculating the planned path, the road information in the expected path is obtained;
[0064] Inputting the road information of the desired path into the lateral and longitudinal control modules of the system for calculation to obtain characteristic information of the vehicle; the characteristic information includes the vehicle heading angle, road curvature, lateral position and longitudinal position;
[0065] Controlling the vehicle based on the characteristic information;
[0066] Among them, the real-time data is input into the lateral and longitudinal control modules for extraction to obtain feature information, including:
[0067] 1. Detect road information in real-time data and extract planning results of vehicle targets;
[0068] 2. Calculate the collected road information and real-time data to obtain the vehicle tracking error model;
[0069] 3. Design the LQR controller using the obtained vehicle tracking error model and obtain the weight matrix of the LQR controller;
[0070] 4. Input the acquired state information into the DCS optimization module and use the DCS algorithm to optimize the weight matrix parameters of the LQR controller;
[0071] 5. Based on the optimized weight matrix, the LQR controller calculates the control input to achieve precise control of the vehicle's lateral and longitudinal directions;
[0072] The collected speed information is input into the longitudinal control module, and the throttle opening of the accelerator and brake of the target vehicle is obtained by calculation; the vehicle is calculated based on the road information of the planning module to obtain the front wheel turning angle; the vehicle is calculated based on the speed information of the planning module to obtain the throttle opening of the accelerator and brake of the target vehicle;
[0073] like Figure 1 As shown, the present invention includes a path planning module, a state observer module, an LQR controller module, a DCS algorithm module, a longitudinal control module and a data processing module, wherein the path planning module is used to plan the path of the starting point and the end point of the vehicle to obtain the optimal path information, the state observer module is used to monitor the real-time information such as the position, speed, acceleration, etc. of the vehicle in the actual driving process in real time, the LQR controller module is used for the lateral control of the vehicle to control the front wheel angle of the vehicle according to the path information and the real-time information, the DCS algorithm module is used to optimize the weighting matrix in the LQR controller in real time, the longitudinal control module is used to control the speed and acceleration of the vehicle according to the path information and the real-time information, and the data processing module is used to perform data processing on the front wheel angle obtained by the final LQR controller and the speed and acceleration obtained by the longitudinal control;
[0074] like Figure 2 As shown, this method introduces feedforward control to eliminate the lateral error and introduces a lateral error e d After calculation through a PID controller and final processing with the front wheel angle, the lateral error is reduced again. This method is aimed at the understeering problem of most vehicles. After this improvement, the problems caused by the large turning radius and lateral error of the vehicle can be greatly reduced; the specific expression is as follows:
[0075]
[0076] Where: δ is the final front wheel steering angle; n1, n2, n3 are the proportional, integral, and differential factors in the PID controller;
[0077] like Figure 3-7 As shown in the figure, the module unit is a lateral control module, and the module introduces a difference innovation search algorithm; the car model used in the module unit is a two-degree-of-freedom dynamics model ( Figure 3 ), the state space equation of the car in the ideal state can be obtained, and then the lateral error and directional error of the target vehicle in tracking the planned path are considered to obtain the vehicle tracking error model ( Figure 4 ), this module unit also adds a prediction module ( Figure 5 ), and send the prediction results to the path planning module and the decision-making module, which can generate safer and more efficient driving paths and decision-making strategies; based on the above modules, the LQR lateral tracking controller is established using the optimal control theory ( Figure 6 ), and on this basis, feedforward control is introduced. Feedforward control can ensure that the lateral error converges to 0, reduce the control input jitter caused by the sudden change of the reference trajectory point, and improve the dynamic response and steady-state performance of the system;
[0078] In the embodiments of the present application, the weighting matrix of the traditional LQR control algorithm is generally assigned using expert prior knowledge, which makes the traditional LQR controller unable to adapt to system dynamic changes and external disturbances, which may cause the target vehicle to deviate during path tracking, thereby causing traffic safety problems; the present application introduces a differential innovation search algorithm to optimize the weighting matrix of the LQR controller. The DCS algorithm can adapt to different working conditions and operating conditions, and provide the LQR controller with more flexible design parameter adjustments, which is more conducive to path tracking in complex environments, so that the robustness and accuracy of the target vehicle in the subsequent tracking process are greatly improved. The LQR controller designed this time also introduces a predictive feedforward module, and its module diagram is shown as follows: Figure 5 As shown in the figure, the module can analyze the road conditions ahead in advance, solving the problem of repeated adjustment of the front wheel angle when the target vehicle encounters special road conditions, which may cause the vehicle to become unstable and the safety factor to be greatly reduced;
[0079] like Figure 3 As shown in the figure, the model is a two-degree-of-freedom dynamic model, which can describe the vehicle behavior in a simple and relatively accurate manner. Considering the movement of the vehicle in two key degrees of freedom, namely lateral movement and longitudinal movement, this helps to more accurately observe the driving and turning performance of the vehicle; the model mainly obtains the state space equation of the vehicle model based on Newton's second law and the torque balance equation, and the formula is as follows:
[0080]
[0081] Among them: m is the vehicle mass, a is y is the lateral acceleration of the target vehicle, Iz is the moment of inertia of the vehicle around the Z axis; is the yaw angular acceleration. y According to the force analysis, it can be expressed as:
[0082] In a vehicle traveling at high speed, the tire will deform and form a side slip angle. When the vehicle reaches stability, the side slip force and the side slip angle will reach a linear relationship as follows:
[0083]
[0084] C f , C r is the cornering stiffness of the front and rear wheels of the vehicle, α f and α r It can also be approximated as:
[0085]
[0086]
[0087] The front wheel turning angle is generally small and can be ignored, sinδ≈0, cosδ≈1, and the above can be summarized as follows:
[0088]
[0089] Arranging the equation into a state space expression yields:
[0090]
[0091] In the example of this application, the two-degree-of-freedom dynamic model used simplifies the analysis process by ignoring the longitudinal motion of the vehicle, making the model easier to understand and handle; despite the simplification of the model, the model can still capture the key dynamic characteristics of the vehicle in lateral and yaw motion, which is crucial for understanding the maneuverability and stability of the vehicle; the model is applicable to a variety of working conditions, especially when driving at high speeds, and has a strong ability to capture the dynamic lateral and yaw motion of the vehicle; the model unit makes the design and analysis of the control system more direct, and helps the difference innovation search algorithm to optimize the weighting matrix of the LQR controller;
[0092] like Figure 4 As shown, the model is a vehicle tracking error model. Based on the dynamic model, the model introduces the error between the expected path and the actual path to obtain the vehicle tracking error model. The application introduces a natural coordinate system to obtain the vehicle information of the expected position and the actual position during driving, which makes the vehicle more accurate when tracking the path. The error model is mainly based on the shortest distance e between the actual and expected path centroids of the vehicle. dand the deviation between the actual yaw angle of the vehicle and the desired yaw angle The heading angle error formula is as follows:
[0093]
[0094] in: is the actual yaw angle; θ r is the desired heading angle;
[0095] The lateral error e d Based on the first-order derivative, the lateral velocity error e can be obtained by rearranging d :
[0096]
[0097] in: is the lateral velocity error; is the heading error;
[0098] In summary, d and By taking derivatives, we get the transformation formula between the dynamic model and the error model:
[0099]
[0100] For θ r After two guides Can be ignored. Combining the dynamic model, we get the state space expression of the error model:
[0101]
[0102] In the example of this application, the vehicle tracking error model used makes the vehicle lateral and longitudinal coordinated tracking control system and method based on the difference innovation search algorithm of the present invention more standardized and systematic; the model can be used in a real-time feedback control system to dynamically adjust the control input according to the error to achieve better tracking performance; and through the tracking error model, it is used to evaluate the robustness of the control system to external disturbances and model uncertainties, and to improve the safety of the vehicle by reducing the tracking error, especially in emergency obstacle avoidance and complex traffic environments;
[0103] like Figure 5As shown in the figure, the module is a prediction module. In actual driving, the driver uses his eyes and brain to judge the things in front of him, and can avoid the factors that may cause danger to a certain extent. When the path planning seen by the driver is not a straight line, he will turn the steering wheel in advance to change the front wheel angle. However, the algorithm control has a lag. When encountering special road conditions, it may cause repeated adjustment of the front wheel angle, which will cause the vehicle to become unstable and the safety factor will be greatly reduced. Therefore, it is necessary to introduce the prediction function on the basis of the original LQR. Figure 5 It can be seen that the location information of the prediction point is:
[0104]
[0105] Where: T is the prediction time; X pre , Y pre , v xpre 、v ypre , is the predicted point's horizontal coordinate, vertical coordinate, yaw angle, lateral velocity, longitudinal velocity, and yaw angular velocity; X, Y, vx, vy, They are the vehicle's horizontal coordinate, vertical coordinate, lateral speed, longitudinal speed, yaw angle, and yaw angular velocity at the current moment respectively;
[0106] In the example of this application, the prediction module used can predict the future position and behavior of other vehicles and road conditions, and can respond in advance, thereby improving driving safety; T is the prediction time, and the prediction time can be set in advance. A reasonable prediction time can improve the robustness and stability of the system; if the prediction time is too long, it will cause a delay in the system response. Under rapidly changing working conditions, a long prediction time cannot adapt to emergencies;
[0107] like Figure 6 As shown, the module is an LQR module. The vehicle error model can be simplified as:
[0108]
[0109] in:
[0110] For the design of LQR controller, firstly, the Then discretize the A matrix and the B matrix, Ad = (I-AT / 2) -1 (I+AT / 2)(Midpoint Euler method); B d = BT (forward Euler method); discrete period T = 0.01s. The state equation after discretization is obtained:
[0111] X(k+1)=A d X(k)+Bd U(k)
[0112] The performance index of the LQR controller is used to measure the performance of the system; the performance index is a quadratic function of the state and control input, and the form is as follows:
[0113]
[0114] Among them, J is the performance index, X k is the system state vector, U k is the control input vector, Q is the state weight matrix. The larger the elements in the Q matrix, the faster the response speed. It is usually a semi-positive definite matrix. R is the control input weight matrix. The larger the elements in the R matrix, the smoother the control of the system. It is usually a positive definite matrix. The purpose of LQR is to obtain a control input vector U k and the system state vector X k Under the condition of linear constraints, the performance index J reaches the minimum value;
[0115] Find the minimum value of the performance index J under the linear constraints of the vehicle tracking error model X(k+1), and construct the Hamiltonian equation as shown below:
[0116]
[0117] By taking the derivative of the above formula, we can get:
[0118]
[0119] After iterative calculation, the optimal control law U k =-KX k ,in Where P is the solution of the Riccati equation, which has the following form:
[0120] P=Q+A T PA-A T PB(R+B T PB) -1 B T PA
[0121] The optimal control law U k =-KX k Substituting into the state space equation, we can get:
[0122]
[0123] When the vehicle is tracking the path, the path will definitely change. Therefore, no matter how the optimal control law K is adjusted, the steady-state error of the system cannot be 0. Therefore, a feedforward controller is needed to make the steady-state error of the system as close to 0 as possible. The final control law of this invention is:
[0124] U=-Ke rr +δ e
[0125] Where: e is the front wheel steering angle in feedforward control;
[0126] Substitute the final control law into the state space equation and make We can get:
[0127]
[0128] In order to select a suitable e , so that It may be 0, and we can get it by calculation:
[0129]
[0130] Where ρ is the road curvature, k3 is the third column element of the feedback gain matrix K;
[0131] In the embodiment of the present application, a feedforward is introduced when designing the LQR controller for the vehicle because the traditional LQR controller cannot make its steady-state error reach 0. After adding the feedforward, it becomes possible for the LQR controller to reduce the steady-state error to 0; the feedforward control is combined with the LQR feedback control of the vehicle to ensure the stability of the system while optimizing the control performance, achieving faster response speed and smaller overshoot; the feedforward control of the LQR module designed in this application is combined with the state observer to make decisions quickly based on real-time measurement and predicted information, which will significantly improve the accuracy and safety factor of the vehicle when tracking the path. In addition to introducing feedforward control to eliminate lateral errors, this application also introduces lateral errors through a PID controller to solve the problem of understeering; understeering will reduce the maneuverability of the vehicle, and it will be difficult for the driver to perform precise steering control as expected, increasing the risk of traffic accidents, especially in emergency avoidance or high-speed driving. The introduction of lateral errors to compensate for the front wheel angle can significantly solve the problem of understeering;
[0132] like Figure 7As shown in the figure, the module is the DCS calculation module. The inspiration of DCS is to use the concepts of differentiated knowledge acquisition and creative realism. The initialization of DCS starts with a set of candidate members. The solution is randomly generated between the upper bound U and the lower bound L of the optimization problem. The initialized members are divided into high performers, average performers, and low performers through differentiated knowledge acquisition (DKA). The effect of the DKA process on each individual X can be performed using the following formula:
[0133] j rand =randint(1,D)
[0134]
[0135] Where v i,d is the element v at the dth position of the experimental member it , v it is the experimental member v of the tth iteration i ;x i,d represents the element at the dth position in individual X; U(0,1) represents the uniform distribution on the interval (0,1); η i,t represents the quantitative knowledge acquisition rate (QKR) of the individual at the tth iteration; j rand is an integer randomly selected from 1 to dimension D and generated once for each sample;
[0136] Differentiated Knowledge Acquisition (DKA) randomly divides individuals into three categories, with the goal of tailoring different learning plans for each type of team member; high performers adopt divergent thinking, average performers adopt convergent thinking, and low performers adopt team diversification; the algorithm incorporates the concept of creative realism, balancing exploration and development through iterative cycles of divergent and convergent thinking to prevent premature identification of suboptimal solutions;
[0137] Retrospective evaluation helps teams learn from past performance and make improvements. In the DCS algorithm, retrospective evaluation focuses on two aspects: the selection process and best performance tracking; in each iteration of the selection process, the algorithm ranks individuals according to their objective value, and lower rankings indicate better performance; the equation for the selection process is as follows:
[0138]
[0139] Where, X i,t+1 represents the i-th individual of individual X at the t+1 iteration;
[0140] Retrospective evaluation also involves tracking and identifying top performers, which helps future decisions and drives performance improvements for the entire team. Through this evaluation, teams are able to recognize their strengths, weaknesses, decision-making, individual contributions, and communication dynamics, thereby fostering a culture of continuous improvement.
[0141] In the embodiment of the present application, the DCS algorithm is very suitable for vehicle path tracking using LQR control. By mathematically modeling the vehicle, the weighted matrix of the LQR controller is obtained. DCS is optimized in a wide search space and is suitable for multi-objective optimization. The fitness function in this application is:
[0142]
[0143] Where: T is the sampling time; e d (t) δ(t) are the lateral error, heading error, lateral velocity error, heading angular velocity error, and front wheel steering angle error of the vehicle at time t;
[0144] According to the above fitness function, each member can find its best position so that the fitness function value is the minimum and the member meets the weight matrix of the LQR controller; that is, all members find the position and weight matrix with the minimum fitness value; the above fitness function takes into account the five aspects that most affect the path tracking accuracy, and optimizing them can achieve better accuracy;
[0145] like Figure 8 As shown in the figure, the module is a longitudinal control module. The longitudinal position controller and the longitudinal speed controller are both PID controllers. The first PID controller is mainly used to control the error between the planned position and the actual position, and the speed compensation is obtained and compensated to the second PID controller together with the planned speed. The second controller is mainly used to control the vehicle speed, and its control target is the speed error e v , the speed error and the speed compensation obtained by the position controller are calculated together to obtain the acceleration compensation. The relationship between the input and output of the PID control method is as follows:
[0146]
[0147] Where: u t is the output variable, e t is the input variable, K P is the proportionality coefficient, K I is the integration coefficient, K D is the differential coefficient;
[0148] In the embodiment of the present application, the PID controller is widely used in the field of path tracking because of its simple structure, easy implementation and adjustment, strong adaptability, high stability, and good cost-effectiveness; it can quickly respond to system errors and achieve precise control of the target vehicle's motion through intuitive parameter adjustment methods; the speed and acceleration information obtained after calculation by the longitudinal dual PID controller is passed to the lower-level controller again, which calculates the throttle opening of the target vehicle's accelerator and brake to achieve speed control of the vehicle.
[0149] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by those with ordinary knowledge in the technical field without departing from the spirit and technical ideas disclosed by the present invention should still be covered by the claims of the present invention.
Claims
1. A vehicle lateral and longitudinal collaborative tracking method based on a difference innovation search algorithm, Features: The steps include: S1: Through the state observer, the vehicle information in front is detected in real time, and the real-time vehicle data is obtained by the lateral and longitudinal control modules; S2: Input the real-time data into the difference innovation search algorithm DCS algorithm module to optimize the gain parameters in the weighting matrix of the LQR controller; S3: Based on the optimized gain parameters, the decision control module is used to adjust the lateral and longitudinal movements of the vehicle so that the target vehicle can track the predetermined trajectory.
2. According to claim 1, a vehicle lateral and longitudinal collaborative tracking method based on a difference innovation search algorithm, Features: The state observer includes a sensor component, a data acquisition unit and a data processing unit; Sensor component: The sensor is the input part of the state observer and is used to collect real-time data of the vehicle; the real-time location information of the vehicle is obtained through the global positioning system GPS; The inertial measurement unit (IMU) is used to measure the real-time acceleration and angular velocity of the vehicle to determine the speed and attitude of the vehicle. The wheel encoder is used to measure the rotation speed and direction of the wheel to assist in determining the speed and position of the vehicle. The vehicle bus system provides status information of other subsystems of the vehicle. The data acquisition unit is responsible for collecting data from sensors and converting it into the corresponding processing format; The data processing unit is the core part of the state observer, which processes the sensor data and estimates the vehicle state by using algorithms.
3. According to claim 1, a vehicle lateral and longitudinal collaborative tracking method based on a difference innovation search algorithm, Features: The longitudinal controller is mainly divided into: upper controller and lower controller; the upper controller is a decision-making module, which obtains the expected acceleration or expected torque according to the position deviation, i.e., speed compensation, actual speed and acceleration closed loop; the position controller mainly calculates the position error calculated by the lateral controller to obtain the speed compensation; the speed compensation, actual speed and expected speed obtained by the position controller are sent to the speed controller to obtain the acceleration compensation; The lower controller, as an execution module, compares the obtained speed and acceleration with the accelerator and brake calibration table, and then converts the data into the throttle opening of the vehicle's accelerator and brake, thereby controlling the acceleration or deceleration of the vehicle; its logical formula is as follows: Where: thr is the throttle valve opening; T max is the maximum torque of the motor; rpm is the speed; the maximum speed of the car is assumed to be 4523.
4. The vehicle lateral and longitudinal collaborative tracking method based on the difference innovation search algorithm according to claim 1, Features: The DCS algorithm module includes: Divide team members into three categories: high performers: excel in divergent thinking and self-driven learning; average performers: use convergent thinking to shape various ideas into feasible solutions; low performers: bring diversity to the team, promote exploration and increase the idea pool; Different team members use different learning plans; Retrospective evaluations are used to monitor improvements among performers and identify top performers; they provide data for decision making and drive performance improvements across the team; The specific steps are as follows: S2.1: Initialize the population and set the upper and lower bounds and number of iterations. The initialization formula is as follows: x i,d =LB d +U(0,1)*(UB d -LB d ) Where U(0,1) represents a random number in the interval (0,1), LB d and UB d Lower and upper bounds on the dth dimension specified for the optimization problem; S2.2: Use the differentiated knowledge acquisition rate to divide the population into high performers, average performers, and low performers. The formula for calculating the differentiated knowledge acquisition rate is as follows: in is the variable of individual i at the tth iteration The specific calculation method of the coefficient value is as follows: Where R i,t is the rank of the i-th individual at the beginning of the t-th iteration, and NP is the population size; S2.3: Use creative realism to encourage high performers to adopt divergent thinking, average performers to adopt convergent thinking, and low performers to adopt team diversity; The formula for convergent thinking is as follows: v i,d =w*x best,d +λ t *(x r2,d -x i,d )+ω i,t *(x r1,d -x i,d ) where x best,d is the dth position of the best performing individual in the current iteration, w is the best cognitive weight, the default value is 1, ω i,t is the ω coefficient value of individual i at the tth iteration, x r1,d is a random number in {1,2,…,NP} and X r1 ≠X≠X best , x r2,d is a random number in {ngs+1,…,NP} and X r2 ≠X r1 ≠X≠X best , ngs is the number of high achievers; λ t is the coefficient value λ of the individual at the tth iteration, and its expression is as follows: Among them, NFE t Indicates the number of times the current function is evaluated at time t, NFE max Indicates the maximum number of function calculations; The formula for divergent thinking is as follows: v i,d =x r1,d +P(α,σ) where x r1,d is a random number in {1,2,…,NP} and X r1 ≠X≠X best , Lk(α,σ) represents the random number generator of Linnik distribution with control parameters α and σ; The formula for team diversity is as follows: V NP =LB+U(0,1)*(UB-LB) Among them, U(0,1) represents a random number in the interval (0,1), L and U are the upper and lower bounds of the optimization problem; S2.4: Retrospective Assessment RA is used to establish evaluation criteria as a benchmark for measuring success. RA determines a strategy for continuous improvement so that the team can learn from experience and then optimize collaboration; its main equation is as follows:
5. The vehicle lateral and longitudinal collaborative tracking method based on the difference innovation search algorithm according to claim 1, Features: The LQR controller includes predictive feedback control, front wheel angle compensation and gain adjustment unit; Predictive feedback control, which is used to calculate control inputs based on the deviation between the current state and the predetermined trajectory; Front wheel angle compensation: State feedback cannot make the steady-state error of the system as close to zero as possible. The introduction of front wheel angle compensation can further reduce the steady-state error. The gain adjustment unit adjusts the control input according to the gain parameter provided by the difference innovation search algorithm DCS algorithm module.
Citation Information
Cited By
Closed-loop automobile crosswind stability dynamic evaluation system based on MPC control algorithm
CN119882422A
Vehicle lane changing trajectory tracking error elimination method and device and storage medium
CN120096601A
