Football team training optimization method based on chaos opposition learning

Abstract

The invention discloses a football team training optimization method based on chaos opposition learning, and the method comprises the steps: firstly designing a novel chaos mapping method, achieving the more random and chaos population initialization, and accelerating the initial convergence speed of an algorithm; and then updating the population by using a collective training strategy, a combined training strategy, a grouping training strategy and an individual additional training strategy in sequence. And outputting an optimal solution until the maximum number of iterations is met. Compared with the existing optimization method, the method provided by the invention not only can solve the high-dimensional and non-convex optimization problem, but also has stronger global search capability and faster solution speed, and can solve the complex optimization problem to a certain extent.

Description

Technical Field

[0001] The invention belongs to the field of intelligent computing technology, and in particular relates to a football team training optimization method based on chaotic adversarial learning. Background Art

[0002] Optimization algorithms are fundamental tools in mathematics and computer science that are designed to identify the minimum or maximum of a function subject to various constraints. These algorithms have a wide range of applications in multiple engineering disciplines. For example, in the military, they are essential for optimizing the task allocation and path planning of drones; in the field of intelligent systems, they serve as parameter optimizers, providing strong support for the collaborative control of multi-agent systems; in the financial industry, these algorithms can effectively predict the occurrence of financial crises; in the field of electrical engineering, they significantly improve the operating efficiency of power systems; and in project management, optimization algorithms help us efficiently allocate tasks to team members within the organization.

[0003] Traditional optimization algorithms such as linear programming, quadratic programming, and convex optimization have been widely used in their respective fields. However, in the face of complex problems with high dimensions and nonlinearity, these traditional algorithms often show certain limitations in accuracy and convergence speed. In this context, meta-heuristic algorithms based on random search strategies have gradually attracted widespread attention from researchers and practitioners due to their wide applicability, high efficiency, and flexibility. These meta-heuristic algorithms, in their unique way, provide new ideas and methods for solving complex optimization problems. In recent years, inspired by human social behavior, evolution, biological population characteristics, and physical theory, scholars have conducted in-depth research and proposed various meta-heuristic algorithms, such as particle swarm optimization, gray wolf algorithm, ant colony algorithm, genetic algorithm, whale algorithm, Harris eagle algorithm, football team training algorithm, etc.

[0004] Metaheuristics guide the search through some rules or empirical knowledge, so as to search for the best solution in the solution space within an acceptable time. It is worth noting that in metaheuristic algorithms, the solution to the problem is usually called a population, but this search strategy usually has the following shortcomings:

[0005] 1. It is easy to fall into local optimality. The algorithm only focuses on finding the global optimal solution based on the currently found optimal solution, which usually leads to falling into the local optimal solution. For example, the simulated annealing algorithm, particle swarm algorithm, whale algorithm, etc. all have similar problems;

[0006] 2. Slow convergence. Metaheuristic algorithms often require more iterations to find the optimal solution. Not only do they converge slowly, but they also take up more computing resources. For example, the ant colony algorithm, gray wolf algorithm, Harris Eagle algorithm, etc. all have similar problems. Summary of the invention

[0007] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a football team training optimization method based on chaotic adversarial learning to solve the parameter optimization problem of the collaborative controller. The present invention optimizes the randomly generated population, increases the randomness and unpredictability, and can better simulate the athlete recruitment behavior; then the athlete status is updated according to the collective training strategy, the joint training strategy, the group training strategy and the individual additional training strategy, which improves the global optimization ability and convergence speed of the algorithm.

[0008] The objective of the present invention is achieved through the following technical scheme: A football team training optimization method based on chaotic adversarial learning, comprising the following steps:

[0009] S1. Define the multi-dimensional optimization problem minf(x): For the collaborative control problem of satellite networking, a neural network-based controller is deployed on each satellite. The goal of collaborative control is to make the attitude angles of all satellites consistent. The controller has two custom parameters to be optimized: controller gain and neural network update rate. The goal of this optimization problem is to select a pair of suitable parameters that meet the following indicators: 1) minimize the total control input of the controller; 2) minimize the neural network estimation error; 3) minimize the standard deviation of the attitude angle convergence error; each controller needs to optimize these three indicators, so the optimization problem f(x) of satellite networking collaborative control is defined as follows:

[0010]

[0011] Wherein, N is the number of satellites, T is the total number of controller simulation iterations, K is the number of attitude angle convergence error vectors, in this embodiment, N = 4, T = 50001, K = 2. α1, α2, α3 are the weights of the three optimization objectives, which need to satisfy α1+α2+α3=1. In this embodiment, α1=0.3, α2=0.3, α3=0.4; independent variable x={x 1 ,x 2 ,x 3}, x 1 、x 2 、x 3 are the control output of the cooperative controller, the neural network estimation error, and the standard deviation of the attitude angle convergence error, respectively. 1 and x 2 is a matrix of size N×T, x 3 is a matrix of size K×T; For x a The sample data of the jth iteration of the ith satellite in , a=1 or 2; is the sample data of the jth iteration when the kth attitude angle converges, For x 3 The mean of

[0012] S2. In the football team recruitment stage, a more chaotic infinite folding iterative logic-tilted tent mapping method is used to iteratively generate pop athletes;

[0013] S3, conduct group training to identify the best and worst athletes, and then update their status according to the athlete type;

[0014] S4. Conduct joint training to generate and absorb opposing athletes, evaluate and rank all athletes, and retain the top 50% of the top athletes to form a training team;

[0015] S5. Conduct group training: divide all athletes into groups, conduct learning and communication within the group, and update the athlete status according to the four learning and communication effects;

[0016] S6. Individual extra training: Find the best players in the football team and give them individual extra training;

[0017] S7. Record and update the best athlete status and its score. If the maximum number of iterations of the algorithm has not been reached, return to S3; if the maximum number of iterations of the algorithm has been reached, output the best athlete status, that is, the optimal solution obtained by this algorithm.

[0018] The present invention optimizes the randomly generated population by designing a novel chaotic mapping method, increases randomness and unpredictability, and can better simulate the athlete recruitment behavior; then the athlete status is updated according to the collective training strategy, joint training strategy, group training strategy and individual additional training strategy, which improves the global optimization ability and convergence speed of the algorithm. Compared with the prior art, the beneficial effects of the present invention are mainly reflected in the following three aspects:

[0019] 1. A novel football team training optimization method based on chaotic adversarial learning is proposed. Compared with genetic algorithm, gray wolf algorithm, Harris eagle algorithm and other methods, the present invention introduces a newer and better-performing football team training algorithm, which improves the optimization performance of the football team training algorithm;

[0020] 2. In existing related studies, most swarm intelligence optimization algorithms use uniform distribution to initialize the population, which is inconsistent with the population behavior in natural systems, reduces the randomness of the algorithm's global search, and slows down the convergence speed. To address this problem, the present invention innovatively proposes an infinitely folded iterative logic-tilted tent mapping method, which uses the chaotic mapping strategy when initializing the population, increases the randomness and unpredictability of the population, better simulates the recruitment behavior of the football team, and enables the algorithm to converge faster.

[0021] 3. The football team training optimization method based on chaotic adversarial learning of the present invention can be applied to the parameter optimization problem of satellite networking collaborative controller, and can be generalized to solve most optimization problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a flow chart of the football team training optimization method based on chaotic adversarial learning of the present invention;

[0023] Figure 2 This is a convergence curve diagram of the comparative experiment based on the CEC2022 test set of the present invention.

[0024] Figure 3 It is a convergence curve diagram of the comparative experiment on the parameter optimization problem of the satellite networking collaborative controller according to the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation methods described herein are only used to explain the present invention and are not used to limit the present invention.

[0026] like Figure 1 As shown, a football team training optimization method based on chaotic adversarial learning of the present invention comprises the following steps:

[0027] S1. Define the multi-dimensional optimization problem minf(x): For the collaborative control problem of satellite networking, a neural network-based controller is deployed on each satellite. The goal of collaborative control is to make the attitude angles of all satellites consistent. The controller has two custom parameters to be optimized: controller gain and neural network update rate. The goal of this optimization problem is to select a pair of suitable parameters that meet the following indicators: 1) minimize the total control input of the controller; 2) minimize the neural network estimation error; 3) minimize the standard deviation of the attitude angle convergence error; each controller needs to optimize these three indicators, so the optimization problem f(x) of satellite networking collaborative control is defined as follows:

[0028]

[0029] Wherein, N is the number of satellites, T is the total number of controller simulation iterations, K is the number of attitude angle convergence error vectors, in this embodiment, N = 4, T = 50001, K = 2. α1, α2, α3 are the weights of the three optimization objectives, which need to satisfy α1+α2+α3=1. In this embodiment, α1=0.3, α2=0.3, α3=0.4; independent variable x={x 1 ,x 2 ,x 3}, x 1 、x 2、x 3 are the control output of the cooperative controller, the neural network estimation error, and the standard deviation of the attitude angle convergence error, respectively. 1 and x 2 is a matrix of size N×T, x 3 is a matrix of size K×T; For x a The sample data of the jth iteration of the ith satellite in , a=1 or 2; is the sample data of the jth iteration when the kth attitude angle converges, For x 3 The mean of .

[0030] The present invention solves the above optimization problem by using football team training optimization based on chaotic adversarial learning, which specifically includes the following steps:

[0031] S2. In the football team recruitment stage, a more chaotic infinite folding iterative logic-tilted tent mapping method is used to iteratively generate pop athletes; there are the following sub-steps:

[0032] S21, the parameter pair consisting of two controller parameters to be optimized is regarded as a player; the number of players of the football team training optimization algorithm is defined as pop = 30, and the maximum number of iterations of the optimization algorithm is IterMax = 200; when the number of iterations is Iter, the solution on the jth dimension is

[0033] S22. Define the iterative logic of infinite folding - tilted tent mapping method. The iterative update formula of the t+1th athlete in one dimension is:

[0034]

[0035] where t∈[1,pop]; x t ∈[l b ,u b ] is a randomly generated athlete state, i.e., a random solution to the optimization problem (the athlete state in the present invention is the specific value of the parameter pair); μ∈(0,4],α∈[0,1] are control parameters;

[0036] S23, randomly generate x1, and then use the infinite folding iterative logic defined in S13 - the tilted tent mapping method to iteratively generate pop athletes.

[0037] S3, conduct group training, find the best and worst athletes, and then update their status according to the athlete type; there are the following sub-steps:

[0038] S31. Calculate the function value f(x) of the generated pop athletes, and find the best athlete for each dimension of the solution to the optimization problem. With the worst athlete

[0039] S32. Use random interval sampling to determine the type of each athlete, i.e., follower, discoverer, thinker, and volatile;

[0040] S33. If the athlete is a follower, he or she is more inclined to follow the footsteps of the best athlete. The update calculation formula is:

[0041]

[0042] Where rand∈[0,1] is a random number; is the solution of the optimization algorithm in the jth dimension when the number of iterations is i;

[0043] S34. If the athlete is a discoverer, then while discovering the gap between himself and the best athlete, he will also avoid learning from the worst athlete. The update calculation formula is:

[0044]

[0045] Where rand1∈[0,1] and rand2∈[0,1] are two different random numbers;

[0046] S35. If the athlete is a thinker, he will pay attention to the difference between the best and the worst athletes. The update calculation formula is:

[0047]

[0048] S36. If the athlete is a fluctuator, his behavior is not affected by others, and his state The update calculation formula is:

[0049]

[0050] Where Iter is the current iteration number, and t(·) is the probability density function that conforms to the t distribution.

[0051] S4, conduct joint training, generate and absorb opposing athletes, evaluate and rank all athletes, and retain the top 50% of the top athletes to form a training team; there are the following sub-steps:

[0052] S41. Based on the players updated by the iteration in S3, an adversarial learning rule is used to generate an adversarial player:

[0053]

[0054] in is the state of the ith opposing athlete; rand1∈[0,1] and rand2∈[0,1] are two different random numbers; k∈(0,∞) is the control parameter of the opposing learning, in this embodiment, k=12000;

[0055] S42. Absorb the opposing athletes into our team, calculate the status of all athletes, and retain the top 50% of athletes to form a training team.

[0056] S5, conduct group training: divide all athletes into groups, conduct learning and communication within the group, and update the athlete status according to the four learning and communication effects; there are the following sub-steps:

[0057] S51. Group the training teams according to the Gaussian mixture model expectation maximization method or the uniform random assignment method, and divide all athletes into four groups: forwards, centers, defenders, and goalkeepers;

[0058] S52. When training in groups, first identify the best athlete in each group;

[0059] S53. In the process of group training, four situations will occur: optimal learning, random learning, random communication and random error. If athlete i can make his ability in a certain dimension reach the level of the best athlete in the group through learning, then it is optimal learning, and its state The update calculation formula is:

[0060]

[0061] in represents the state of the best player in the group in the jth dimension, p study ∈[0,1] is the learning probability;

[0062] S54, if athlete i learns from a random athlete in the group, it is random learning, and its state The update calculation formula is:

[0063]

[0064] in represents the state of a random athlete k in the group in the jth dimension;

[0065] S55. If athlete i randomly communicates with athlete k on the jth dimension, it is a random communication. The states of the two athletes are and The update calculation formula is:

[0066]

[0067] Where randn represents a random number that follows a normal distribution, p comn ∈[0,1] is the probability of communication;

[0068] S56. If athlete i accidentally learns the wrong dimension, it is a random error and its state The update calculation formula is:

[0069]

[0070] in is the random state of the kth athlete in the lth dimension, p error ∈[0,1] is the error probability.

[0071] S6. Individual additional training: Find the best player in the football team and give him individual additional training; the sub-steps are as follows:

[0072] S61. Evaluate all athletes after group training and identify the best athletes after group training;

[0073] S62. Update the status of the best athlete according to the following update formula:

[0074]

[0075] Among them, randn and randc are random numbers that conform to Gaussian distribution and Cauchy distribution respectively. It is the state of the best athletes after additional training alone.

[0076] S7. Record and update the best athlete status and its score. If the maximum number of iterations of the algorithm has not been reached, return to S3; if the maximum number of iterations of the algorithm has been reached, output the best athlete status, that is, the optimal solution obtained by this algorithm.

[0077] In this embodiment, the CEC2022 test set is first used to test the algorithm performance, and then a simulation experiment is performed on the controller parameter optimization problem proposed in this patent. During the test, genetic algorithms (GA), grey wolf optimization (GWO), whale optimization algorithm (WOA), Harris Hawks optimization (HHO), and the original football team training algorithm (FTTA) are used for comparative simulation experiments. The CEC2022 test set has a total of 12 benchmark functions. The comparison results are shown in the figure below. Figure 2 As shown, Figure 3 The figure is a convergence curve diagram of the comparative experiment for the parameter optimization problem of the satellite networking collaborative controller. IFTTA is the convergence curve of the algorithm proposed in the present invention. The simulation results confirm that compared with the mainstream meta-heuristic optimization algorithm, the football team training optimization method based on chaotic adversarial learning has stronger global search ability and convergence speed.

[0078] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A football team training optimization method based on chaotic adversarial learning, characterized in that: The following steps are involved: S1. Define the multi-dimensional optimization problem minf(x): For the collaborative control problem of satellite networking, a neural network-based controller is deployed on each satellite. The goal of collaborative control is to make the attitude angles of all satellites consistent. The controller has two custom parameters to be optimized: controller gain and neural network update rate. The goal of this optimization problem is to select a pair of suitable parameters that meet the following indicators: 1) minimize the total control input of the controller; 2) minimize the neural network estimation error; 3) minimize the standard deviation of the attitude angle convergence error; each controller needs to optimize these three indicators, so the optimization problem f(x) of satellite networking collaborative control is defined as follows: Wherein, N is the number of satellites, T is the total number of controller simulation iterations, K is the number of attitude angle convergence error vectors, in this embodiment, N = 4, T = 50001, K = 2. α1, α2, α3 are the weights of the three optimization objectives, which need to satisfy α1+α2+α3=1. In this embodiment, α1=0.3, α2=0.3, α3=0.4; independent variable x={x 1 ,x 2 ,x 3 }, x 1 、x 2 、x 3 are the control output of the cooperative controller, the neural network estimation error, and the standard deviation of the attitude angle convergence error, respectively. 1 and x 2 is a matrix of size N×T, x 3 is a matrix of size K×T; For x a The sample data of the jth iteration of the ith satellite in , a=1 or 2; is the sample data of the jth iteration when the kth attitude angle converges, For x 3 The mean of S2. In the football team recruitment stage, a more chaotic infinite folding iterative logic-tilted tent mapping method is used to iteratively generate pop athletes; S3, conduct group training to identify the best and worst athletes, and then update their status according to the athlete type; S4. Conduct joint training to generate and absorb opposing athletes, evaluate and rank all athletes, and retain the top 50% of the top athletes to form a training team; S5. Conduct group training: divide all athletes into groups, conduct learning and communication within the group, and update the athlete status according to the four learning and communication effects; S6. Individual extra training: Find the best players in the football team and give them individual extra training; S7. Record and update the best athlete status and its score. If the maximum number of iterations of the algorithm has not been reached, return to S3; if the maximum number of iterations of the algorithm has been reached, output the best athlete status.

2. A football team training optimization method based on chaotic adversarial learning according to claim 1, characterized in that: In S2, there are the following sub-steps: S21, the parameter pair consisting of two controller parameters to be optimized is regarded as an athlete; the number of athletes in the football team training optimization algorithm is defined as pop, and the maximum number of iterations of the optimization algorithm is IterMax; when the number of iterations is Iter, the solution on the jth dimension is S22. Define the iterative logic of infinite folding - tilted tent mapping method. The iterative update formula of the t+1th athlete in one dimension is: where t∈[1,pop]; x t ∈[l b ,u b ] is a randomly generated athlete state, i.e., a random solution to the optimization problem; μ∈(0,4],α∈[0,1] are control parameters; S23, randomly generate x1, and then use the infinite folding iterative logic defined in S13 - the tilted tent mapping method to iteratively generate pop athletes.

3. A football team training optimization method based on chaotic adversarial learning according to claim 1, characterized in that: In S3, there are the following sub-steps: S31. Calculate the function value f(x) of the generated pop athletes, and find the best athlete for each dimension of the solution to the optimization problem. With the worst athlete S32. Use random interval sampling to determine the type of each athlete, i.e., follower, discoverer, thinker, and volatile; S33. If the athlete is a follower, he or she is more inclined to follow the footsteps of the best athlete. The update calculation formula is: Where rand∈[0,1] is a random number; is the solution of the optimization algorithm in the jth dimension when the number of iterations is i; S34. If the athlete is a discoverer, then while discovering the gap between himself and the best athlete, he will also avoid learning from the worst athlete. The update calculation formula is: Where rand1∈[0,1] and rand2∈[0,1] are two different random numbers; S35. If the athlete is a thinker, he will pay attention to the difference between the best and the worst athletes. The update calculation formula is: S36. If the athlete is a fluctuator, his behavior is not affected by others, and his state The update calculation formula is: Where Iter is the current iteration number, and t(·) is the probability density function that conforms to the t distribution.

4. A football team training optimization method based on chaotic adversarial learning according to claim 1, characterized in that: In S4, there are the following sub-steps: S41. Based on the players updated by the iteration in S3, an adversarial learning rule is used to generate an adversarial player: in is the state of the ith opposing athlete; rand1∈[0,1] and rand2∈[0,1] are two different random numbers; k∈(0,∞) is the control parameter of the opposing learning, in this embodiment, k=12000; S42. Absorb the opposing athletes into our team, calculate the status of all athletes, and retain the top 50% of athletes to form a training team.

5. A football team training optimization method based on chaotic adversarial learning according to claim 1, characterized in that: In S5, there are the following sub-steps: S51. Group the training teams according to the Gaussian mixture model expectation maximization method or the uniform random assignment method, and divide all athletes into four groups: forwards, centers, defenders, and goalkeepers; S52. When training in groups, first identify the best athlete in each group; S53. In the process of group training, four situations will occur: optimal learning, random learning, random communication and random error. If athlete i can make his ability in a certain dimension reach the level of the best athlete in the group through learning, then it is optimal learning, and its state The update calculation formula is: in represents the state of the best player in the group in the jth dimension, p study ∈[0,1] is the learning probability; S54, if athlete i learns from a random athlete in the group, it is random learning, and its state The update calculation formula is: in represents the state of a random athlete k in the group in the jth dimension; S55. If athlete i randomly communicates with athlete k on the jth dimension, it is a random communication. The states of the two athletes are and The update calculation formula is: Where randn represents a random number that follows a normal distribution, p comn ∈[0,1] is the probability of communication; S56. If athlete i accidentally learns the wrong dimension, it is a random error and its state The update calculation formula is: in is the random state of the kth athlete in the lth dimension, p error ∈[0,1] is the error probability.

6. A football team training optimization method based on chaotic adversarial learning according to claim 1, characterized in that: In S6, there are the following sub-steps: S61. Evaluate all athletes after group training and identify the best athletes after group training; S62. Update the status of the best athlete according to the following update formula: Among them, randn and randc are random numbers that conform to Gaussian distribution and Cauchy distribution respectively. It is the state of the best athletes after additional training alone.

Images