TransFormer-APF-MPC-based automatic driving vehicle obstacle avoidance method and device, and medium

Abstract

The invention relates to an automatic driving vehicle obstacle avoidance method and device based on TransFormer-APF-MPC and a medium, and the method comprises the following steps: obtaining pedestrian motion data and interaction scene data in real time, inputting the pedestrian motion data and the interaction scene data into an automatic driving vehicle obstacle avoidance network based on TransFormer-APF-MPC, and outputting an automatic driving vehicle obstacle avoidance scheme; the automatic driving vehicle obstacle avoidance network based on the TransFormer-APF-MPC comprises a TransFormer model and an APF-MPC obstacle avoidance model, the TransFormer model is used for predicting pedestrian trajectories according to input data, the APF-MPC obstacle avoidance model comprises a path planning layer and a trajectory tracking layer, the path planning layer is used for generating an optimal obstacle avoidance path by improving an artificial potential field method, and the trajectory tracking layer is used for tracking the optimal obstacle avoidance path. According to the improved artificial potential field method, the repulsive force between the vehicle and the obstacle and the gravitational force between the vehicle and the target point are calculated according to the improved repulsive force constant function with the adjustment factor introduced; and the trajectory tracking layer is used for calculating a vehicle control instruction through a model prediction control method, so that the vehicle runs along the optimal obstacle avoidance path. Compared with the prior art, the real-time obstacle avoidance planning and control of the vehicle can be better realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of autonomous driving, and in particular, relates to an obstacle avoidance method, device and medium for an autonomous driving vehicle based on TransFormer-APF-MPC. Background Art

[0002] Unmanned driving technology mainly includes two key modules: prediction and control. Among them, the prediction module is mainly responsible for environmental perception and behavior prediction, while the control module is mainly responsible for making control decisions to achieve autonomous obstacle avoidance. However, these two key modules are currently in a relatively independent development stage, resulting in the control module's insufficient consideration of the prediction module results, resulting in insufficient consideration of pedestrian redundancy and safety, which may lead to more uncertainties and risks for unmanned vehicles in actual applications.

[0003] The artificial potential field method (APF) guides the movement of the vehicle by establishing a comprehensive potential field such as the virtual target gravitational potential field and the obstacle repulsive potential field. It has the characteristics of simple model and strong real-time performance. It was widely used in the field of robot path planning in the early days. With the deepening of research, this method has also been introduced into the field of intelligent driving and has achieved good results. Chinese patent application 202211648142.X discloses a path planning method and system based on a tracked vehicle. If the tracked vehicle has unknown dynamic obstacles, APF is used for dynamic collision avoidance, enhancing the repulsive force of the surrounding obstacles and weakening the gravitational force of the target point to avoid obstacles. However, there are two problems with the traditional APF: First, when the distance between the obstacle and the target point is very close, the vehicle approaches the target point, the gravity disappears but the repulsive force still exists, which may cause the target to be unreachable; the gravity disappears but the repulsive force still exists, which may cause the target to be unreachable; second, when the repulsive force of all obstacles is consistent with the direction of gravity and the vehicle does not reach the target, the repulsive force and the gravitational force may be balanced, resulting in a local optimum. The above problems have caused certain limitations on the dynamic collision avoidance capabilities of the vehicle. Therefore, it is necessary to design an obstacle avoidance method for autonomous driving vehicles to better realize the planning and control of real-time obstacle avoidance of the vehicle. Summary of the invention

[0004] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide an obstacle avoidance method, device and medium for autonomous driving vehicles based on TransFormer-APF-MPC, solve the problems of local optimality and unreachable targets, and better realize the planning and control of real-time obstacle avoidance of vehicles.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] The present invention provides an obstacle avoidance method for an autonomous driving vehicle based on TransFormer-APF-MPC, comprising the following steps:

[0007] Obtain pedestrian motion data and interactive scene data in real time, input the trained TransFormer-APF-MPC-based autonomous driving vehicle obstacle avoidance network, and output the autonomous driving vehicle obstacle avoidance solution, including the optimal obstacle avoidance path and vehicle control instructions;

[0008] Among them, the autonomous driving vehicle obstacle avoidance network based on TransFormer-APF-MPC includes a Transformer model and an APF-MPC obstacle avoidance model. The Transformer model is used to predict pedestrian trajectories based on input data; the APF-MPC obstacle avoidance model includes a path planning layer and a trajectory tracking layer. The path planning layer is used to generate an optimal obstacle avoidance path based on the pedestrian trajectory prediction result through an improved artificial potential field method. The improved artificial potential field method calculates the repulsion between the vehicle and the obstacle and the attraction between the vehicle and the target point according to an improved repulsion constant function that introduces an adjustment factor, ensuring that the repulsion and attraction are reduced to zero at the same time when the vehicle reaches the target point, and describes pedestrian behavior, road boundary risks and destination tendencies through pedestrian repulsion fields, road boundary repulsion fields and road gravity fields respectively, and calculates the optimal obstacle avoidance path based on this, and inputs it into the trajectory tracking layer; the trajectory tracking layer is used to calculate vehicle control instructions through a model predictive control method, so that the autonomous driving vehicle travels along the optimal obstacle avoidance path.

[0009] Furthermore, the pedestrian motion data and interaction scene data are acquired in real time through a multi-sensor information fusion algorithm. The specific process is as follows:

[0010] On streets with mixed traffic of people and vehicles, the self-driving vehicle’s camera and lidar can be used to obtain visual information and laser ranging information respectively.

[0011] Through weighted averaging or Kalman filtering algorithm, the screened and cleaned visual information and laser ranging information are aligned and fused to obtain pedestrian motion data and interactive scene data. The pedestrian motion data includes the current position and speed of the pedestrian, and the interactive scene data includes the current position and speed of the vehicle.

[0012] Furthermore, the Transformer model decouples the uncertainty of pedestrians' future behaviors through a sparse intention anchor mechanism to assist in the prediction of pedestrian trajectories.

[0013] Furthermore, in the improved artificial potential field method, the expression of the improved repulsive force constant function is specifically as follows:

[0014]

[0015] Among them, m is the proportionality coefficient, d(q,q 0 ) is the distance between the vehicle and the obstacle, d 0 is the impact distance of the obstacle, is the adjustment factor, and n is an arbitrary constant.

[0016] Furthermore, the calculated optimal obstacle avoidance path is smoothed and then input into the trajectory tracking layer.

[0017] Furthermore, the specific process of calculating the vehicle control command by the model predictive control method is as follows:

[0018] The vehicle's motion state and behavior are described through the vehicle dynamics model;

[0019] Construct objective function and constraint function, perform optimization and solve, and obtain vehicle control instructions.

[0020] Furthermore, a bicycle model is used as the vehicle dynamics model.

[0021] Furthermore, the expression of the objective function is as follows:

[0022]

[0023] in, is the predicted front wheel turning angle at time t, is the control input predicted at time t-1, ΔU(t) is the control increment sequence required in each control cycle, N p is the length of the prediction time domain, γ(t+i|t) is the predicted vehicle state at time t+i, is the reference trajectory point at time t+i, N c To control the length of the time domain, R and Q are weight matrices, which are used to control the output deviation and the control amount respectively, ρ is the relaxation factor weight matrix, and ε is the relaxation factor;

[0024] The expression of the constraint function is as follows:

[0025] ΔU min ≤ΔU t ≤ΔU max

[0026] U min ≤AΔU t +U t ≤U max

[0027] y hc,min ≤y hc ≤y hc,max

[0028] y sc,min-ε≤y sc ≤y sc,max +ε

[0029] Among them, ΔU min and ΔU max are the minimum and maximum control increment sets, U min and U max are the minimum and maximum control quantity sets, y hc is the hard constraint output, y sc is the soft constraint output, y hc,min and hc,max are the minimum and maximum values ​​of the hard constraint output, respectively. sc,min and sc,max are the minimum and maximum values ​​of the soft constraint output respectively.

[0030] The present invention also provides an electronic device, comprising a memory, a processor, and a program stored in the memory, wherein the processor implements the above method when executing the program.

[0031] The present invention also provides a computer-readable storage medium on which a computer program is stored, and the program implements the above method when executed by a processor.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] 1. The present invention proposes an obstacle avoidance method for an autonomous driving vehicle based on TransFormer-APF-MPC, wherein an obstacle avoidance scheme for an autonomous driving vehicle is generated through an obstacle avoidance network for an autonomous driving vehicle based on TransFormer-APF-MPC, including an optimal obstacle avoidance path and vehicle control instructions; wherein the obstacle avoidance network for an autonomous driving vehicle based on TransFormer-APF-MPC includes a Transformer model and an APF-MPC obstacle avoidance model, wherein the Transformer model is used to predict pedestrian trajectories based on input data, and assist in the subsequent generation of obstacle avoidance schemes; the APF-MPC obstacle avoidance model includes a path planning layer and a trajectory tracking layer, wherein the path planning layer is used to generate an optimal obstacle avoidance path based on the pedestrian trajectory prediction result by improving the artificial potential field method, and the present invention improves In the artificial potential field method, an improved repulsive constant function with an adjustment factor is designed to calculate the repulsive force between the vehicle and the obstacle and the attractive force between the vehicle and the target point. This can ensure that the repulsive force and attractive force are reduced to zero at the same time when the vehicle reaches the target point, thereby solving the problems of local optimality and unreachable target. The pedestrian repulsive field, road boundary repulsive field and road gravity field are used to describe pedestrian behavior, road boundary risk and destination tendency respectively. While maintaining a safe distance between the vehicle and pedestrians and the boundary, the vehicle can be guided along the center line of the road or a planned path, and the optimal obstacle avoidance path is calculated accordingly. The trajectory tracking layer is used to calculate the vehicle control instructions through the model predictive control method, so that the autonomous driving vehicle can travel along the optimal obstacle avoidance path and realize precise control of the vehicle movement. Through the above method, the planning and control of vehicle real-time obstacle avoidance can be better realized.

[0034] 2. The present invention smoothes the calculated optimal obstacle avoidance path and then inputs it into the trajectory tracking layer, which can reduce the curvature and turning angle of the path and improve driving comfort and safety. The present invention uses a bicycle model as a vehicle dynamics model, which can simplify a four-wheeled vehicle into a two-wheeled vehicle, and describes the vehicle's dynamic characteristics such as yaw angular velocity, steering angle, lateral acceleration, etc., and is suitable for steering and control scenarios.

[0035] 3. The present invention specifically sets control increment constraints and control quantity constraints in the constraint function, which can ensure the smoothness and controllability of the vehicle's dynamic response. At the same time, hard constraints and soft constraints are also set to ensure that the vehicle will not violate safety or operating restrictions under any circumstances. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 Schematic diagram of the obstacle avoidance network structure of an autonomous driving vehicle based on TransFormer-APF-MPC. DETAILED DESCRIPTION

[0037] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0038] Example:

[0039] This embodiment provides an obstacle avoidance method for an autonomous driving vehicle based on TransFormer-APF-MPC, comprising the following steps:

[0040] Acquire pedestrian motion data and interactive scene data in real time, input the trained TransFormer-APF-MPC-based autonomous driving vehicle obstacle avoidance network, and output the autonomous driving vehicle obstacle avoidance plan, including the optimal obstacle avoidance path and vehicle control instructions.

[0041] like Figure 1 As shown, the obstacle avoidance network of an autonomous driving vehicle based on TransFormer-APF-MPC includes a Transformer model and an APF-MPC obstacle avoidance model, wherein the Transformer model is used to predict pedestrian trajectories based on input data, and the APF-MPC obstacle avoidance model is used to generate an obstacle avoidance plan for an autonomous driving vehicle based on the pedestrian trajectory prediction results.

[0042] In the Transformer model, pedestrian motion data and interaction scene data are obtained in real time through a multi-sensor information fusion algorithm. The specific process is as follows:

[0043] First, on streets with mixed pedestrian and vehicle traffic, the camera and lidar of the autonomous driving vehicle are used to obtain visual information and laser ranging information respectively. Then, the filtered and cleaned visual information and laser ranging information are aligned and fused through weighted average or Kalman filter algorithm to obtain pedestrian motion data and interactive scene data. The pedestrian motion data includes the current position and speed of the pedestrian, and the interactive scene data includes the current position and speed of the vehicle.

[0044] In view of the problem that existing methods cannot well predict the future multimodal trajectories of pedestrians, this embodiment proposes a learnable sparse intention anchor mechanism in the encoder-decoder architecture of the Transformer model to decouple the uncertainty of pedestrians' future behaviors and achieve stable and accurate prediction of pedestrians' future multimodal trajectories. Specifically, a set of sparse intention anchors are first extracted from the data set through cluster analysis, which represent the possible future motion patterns or intentions of pedestrians. Then, these anchors are converted into high-dimensional feature vectors in the encoder stage, and position encoding is added so that the model can capture the temporal order and position relationship of pedestrian trajectories. In the decoder stage, these intention anchors interact with the historical trajectory context features extracted by the encoder, and the uncertainty of pedestrians' future behaviors is gradually decoupled and refined through iterative refinement of the multi-layer Transformer structure. This process allows the model to decompose complex future uncertainty distributions into a set of simple unimodal distributions, thereby simplifying the optimization process and improving the model's ability to predict pedestrians' future behaviors. Finally, the output layer of the decoder can output the complete predicted future trajectory at one time. This sparse intent anchor-based approach not only stabilizes the training process, but also effectively covers all future modalities by adaptively refining the intent anchors, reducing the reliance on dense anchors and significantly reducing computational and memory costs.

[0045] After training, the Transformer model can obtain a set of weights and parameters with the best performance, so that the model performs best on the validation set. The optimized weights and parameters are used to predict pedestrian trajectories in the future, and the current and historical data are used to generate possible movement paths of pedestrians in the future.

[0046] The APF-MPC obstacle avoidance model includes a path planning layer and a trajectory tracking layer. The path planning layer is used to generate the optimal obstacle avoidance path based on the pedestrian trajectory prediction results by improving the artificial potential field method. Its input information includes the current position and speed of the vehicle, the target position, and environmental obstacles (pedestrians, road boundaries). The output result is a series of path points containing position and direction information. The trajectory tracking layer is used to calculate the vehicle control instructions through the model predictive control method to make the autonomous driving vehicle travel along the optimal obstacle avoidance path. Its input data includes the path points output by the local path planning layer, the current state of the vehicle (position, speed, acceleration, etc.) and the vehicle dynamics model parameters. The output control instructions include the steering wheel angle and acceleration.

[0047] Specifically, the improved artificial potential field method first initializes the initial position of the vehicle, the target position, the position of the obstacle and the road boundary; then, according to the improved repulsion constant function, the repulsion between the vehicle and the obstacle and the attraction between the vehicle and the target point are calculated; subsequently, the pedestrian repulsion field, the road boundary repulsion field and the road gravity field are used to describe the pedestrian behavior, the road boundary risk and the destination tendency respectively, and the optimal obstacle avoidance path is calculated based on this, and smoothing is performed to reduce the curvature and turning angle of the path, thereby improving the driving comfort and safety; finally, the planned optimal obstacle avoidance path is output in the form of a series of path points and input into the trajectory tracking layer.

[0048] Improve the repulsive field function by introducing a regulating factor That is, the distance between the car and the target point. In this way, when the car is driving towards the target, the gravitational force and repulsive force are reduced to a certain extent at the same time, ensuring that the repulsive force and gravitational force are reduced to zero at the same time when the vehicle reaches the target point, thus solving the problem of local optimality and unreachable target. The expression of the improved repulsive constant function is as follows:

[0049]

[0050] Where m is the proportionality coefficient, d(q,q 0 ) is the distance between the vehicle and the obstacle, d 0 is the impact distance of the obstacle, is the distance between the vehicle and the target point, n is an arbitrary constant, and its value is 2 after multiple simulation experiments.

[0051] The pedestrian repulsion field means that during the driving process of the vehicle, pedestrians are regarded as an obstacle, and the vehicle needs to maintain a safe distance from the pedestrian to avoid collision. For this reason, a virtual "repulsion field" is defined around the pedestrian, and the strength of the field increases as the distance between the vehicle and the pedestrian decreases. The road boundary repulsion field is similar to the pedestrian repulsion field, but the object is the boundary of the road (such as the shoulder, guardrail, etc.). In order to keep the vehicle within a safe driving range, when approaching the road boundary, the boundary will exert a "repulsion force" on the vehicle, forcing the vehicle to stay away from the boundary area to avoid crossing the line or causing an accident. Usually, this repulsion also increases as the distance between the vehicle and the boundary decreases. The road gravitational potential field is opposite to the repulsion field. It is an attraction field designed to guide the vehicle to travel along the center line of the road or a planned path. The road gravitational field will urge the vehicle to move closer to the target path or target direction to ensure that the vehicle is driving in the appropriate lane. Its function is to make the vehicle tend to the target when there are no obstacles.

[0052] In the trajectory tracking layer, the specific process of calculating the vehicle control command through the model predictive control method is as follows:

[0053] First, the vehicle's motion state and behavior are described through the vehicle dynamics model; then, the objective function and constraint function are constructed, optimized and solved to obtain the vehicle control instructions.

[0054] In this embodiment, a bicycle model is used as the vehicle dynamics model. This model simplifies a four-wheeled vehicle into a two-wheeled vehicle, and can describe the dynamic characteristics of the vehicle, such as yaw angular velocity, steering angle, and lateral acceleration, and is suitable for steering and control scenarios.

[0055] The expression of the trajectory tracking layer objective function is as follows:

[0056]

[0057] in, is the predicted front wheel turning angle at time t, is the control input predicted at time t-1, ΔU(t) is the control increment sequence required in each control cycle, N p is the length of the prediction time domain, γ(t+i|t) is the predicted vehicle state at time t+i, is the reference trajectory point at time t+i, N c To control the length of the time domain, R and Q are weight matrices, which are used to control the output deviation and the control amount respectively. ρ is the relaxation factor weight matrix, and ε is the relaxation factor.

[0058] The expression of the constraint function is as follows:

[0059] ΔU min ≤ΔU t ≤ΔU max

[0060] U min ≤AΔU t +U t ≤U msx

[0061] y hc,min ≤y hc ≤y hc,max

[0062] y sc,min -ε≤y sc ≤y sc,max +ε

[0063] Among them, ΔU min and ΔU max are the minimum and maximum control increment sets, U min and U max are the minimum and maximum control quantity sets, y hc is the hard constraint output, y scis the soft constraint output, y hc,min and hc,max are the minimum and maximum values ​​of the hard constraint output, respectively. sc,min and sc,max are the minimum and maximum values ​​of the soft constraint outputs, respectively. The control increment constraint and the control amount constraint ensure the smoothness and controllability of the vehicle dynamic response, while the hard constraint and the soft constraint ensure that the vehicle will not violate safety or operating restrictions under any circumstances.

[0064] To verify the effectiveness of the above method, a corresponding model can be built in carsim and tested jointly with the obstacle avoidance model built in simulink.

[0065] If the above method is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0066] The above description of the embodiments is to facilitate the understanding and use of the invention by those skilled in the art. It is obvious that those skilled in the art can easily make various modifications to these embodiments and apply the general principles described herein to other embodiments without creative work. Therefore, the present invention is not limited to the above embodiments, and improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the present invention should be within the scope of protection of the present invention.

Claims

1. An obstacle avoidance method for an autonomous driving vehicle based on TransFormer-APF-MPC, characterized in that: The following steps are involved: Obtain pedestrian motion data and interactive scene data in real time, input the trained TransFormer-APF-MPC-based autonomous driving vehicle obstacle avoidance network, and output the autonomous driving vehicle obstacle avoidance solution, including the optimal obstacle avoidance path and vehicle control instructions; Among them, the autonomous driving vehicle obstacle avoidance network based on TransFormer-APF-MPC includes a Transformer model and an APF-MPC obstacle avoidance model. The Transformer model is used to predict pedestrian trajectories based on input data; the APF-MPC obstacle avoidance model includes a path planning layer and a trajectory tracking layer. The path planning layer is used to generate an optimal obstacle avoidance path based on the pedestrian trajectory prediction result through an improved artificial potential field method. The improved artificial potential field method calculates the repulsion between the vehicle and the obstacle and the attraction between the vehicle and the target point according to an improved repulsion constant function that introduces an adjustment factor, ensuring that the repulsion and attraction are reduced to zero at the same time when the vehicle reaches the target point, and describes pedestrian behavior, road boundary risks and destination tendencies through pedestrian repulsion fields, road boundary repulsion fields and road gravity fields respectively, and calculates the optimal obstacle avoidance path based on this, and inputs it into the trajectory tracking layer; the trajectory tracking layer is used to calculate vehicle control instructions through a model predictive control method, so that the autonomous driving vehicle travels along the optimal obstacle avoidance path.

2. The obstacle avoidance method for an autonomous driving vehicle based on TransFormer-APF-MPC according to claim 1, characterized in that: The multi-sensor information fusion algorithm is used to obtain pedestrian motion data and interaction scene data in real time. The specific process is as follows: On streets with mixed traffic of people and vehicles, the self-driving vehicle’s camera and lidar can be used to obtain visual information and laser ranging information respectively. Through weighted averaging or Kalman filtering algorithm, the screened and cleaned visual information and laser ranging information are aligned and fused to obtain pedestrian motion data and interactive scene data. The pedestrian motion data includes the current position and speed of the pedestrian, and the interactive scene data includes the current position and speed of the vehicle.

3. The obstacle avoidance method for an autonomous driving vehicle based on TransFormer-APF-MPC according to claim 1, characterized in that: The Transformer model decouples the uncertainty of pedestrians' future behaviors through a sparse intention anchor mechanism to assist in the prediction of pedestrian trajectories.

4. The obstacle avoidance method for an autonomous driving vehicle based on TransFormer-APF-MPC according to claim 1, characterized in that: In the improved artificial potential field method, the expression of the improved repulsive force constant function is as follows: Among them, m is the proportional coefficient, d(q,q0) is the distance between the vehicle and the obstacle, d0 is the impact distance of the obstacle, is the adjustment factor, and n is an arbitrary constant.

5. The obstacle avoidance method for an autonomous driving vehicle based on TransFormer-APF-MPC according to claim 1, characterized in that: The calculated optimal obstacle avoidance path is smoothed and then input into the trajectory tracking layer.

6. The obstacle avoidance method for an autonomous driving vehicle based on TransFormer-APF-MPC according to claim 1, characterized in that: The specific process of calculating vehicle control instructions through model predictive control method is as follows: The vehicle's motion state and behavior are described through the vehicle dynamics model; Construct objective function and constraint function, perform optimization and solve, and obtain vehicle control instructions.

7. The method for avoiding obstacles of an autonomous driving vehicle based on TransFormer-APF-MPC according to claim 6, characterized in that: A bicycle model is used as the vehicle dynamics model.

8. The method for avoiding obstacles of an autonomous driving vehicle based on TransFormer-APF-MPC according to claim 6, characterized in that: The expression of the objective function is as follows: in, is the predicted front wheel turning angle at time t, is the control input predicted at time t-1, ΔU(t) is the control increment sequence required in each control cycle, N p is the length of the prediction time domain, γ(t+i|t) is the predicted vehicle state at time t+i, is the reference trajectory point at time t+i, N c To control the length of the time domain, R and Q are weight matrices, which are used to control the output deviation and the control amount respectively, ρ is the relaxation factor weight matrix, and ε is the relaxation factor; The expression of the constraint function is as follows: ΔU min ≤ΔU t ≤ΔU max Y min ≤AΔU t +U t ≤U max and hc,min ≤y hc ≤y hc,max y sc,min -ε≤y sc ≤y sc,max +e Among them, ΔU min and ΔU max are the minimum and maximum control increment sets, U min and U max are the minimum and maximum control quantity sets, y hc is the hard constraint output, y sc is the soft constraint output, y hc,min and hc,max are the minimum and maximum values ​​of the hard constraint output, y sc,min and sc,max are the minimum and maximum values ​​of the soft constraint output respectively.

9. An electronic device comprising a memory, a processor, and a program stored in the memory, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images