Simulation soccer robot evaluation method of LSTM network based on genetic algorithm optimization

Abstract

The invention provides a simulation soccer robot evaluation method based on a genetic algorithm optimized LSTM network, and the method comprises the following steps: building a soccer robot team program based on a simulation soccer platform; constructing a foundation device evaluation module; acquiring original training data of the soccer robot team, and constructing a training set and a test set after preprocessing; building an evaluation network model of a football robot team program based on a long-short term memory network algorithm framework; constructing a genetic algorithm optimization model, and defining a fitness function for guiding genetic manipulation; performing continuous iterative optimization on the evaluation network model by using a genetic algorithm; and an optimized evaluator module is obtained, and the soccer robot is helped to select optimal action parameters in real time. According to the invention, the genetic algorithm is combined with the long and short term memory network, so that the evaluator can reflect the decision-making demand of the soccer robot in a complex dynamic environment more accurately, the overall decision-making precision is improved, and the common problem of insufficient adaptability in a traditional evaluation method is avoided.

Description

Technical Field

[0001] The invention relates to the technical field of simulated soccer robot motion, and more specifically to a simulated soccer robot evaluation method based on an LSTM network optimized by a genetic algorithm. Background Art

[0002] The simulated soccer robot is a complex system that integrates multiple technical fields. In the simulated soccer game, the soccer robot needs to perform multiple tasks such as attack and defense. In the soccer robot team, these tasks are mainly completed by the evaluator. The evaluation mechanism of the traditional evaluator usually relies on pre-set rules, but this method often has problems such as relying on expert systems, being unable to cope with complex situations, lack of adaptability, and limitations in optimization. Therefore, there is an urgent need for an evaluation mechanism that can improve the decision-making ability of the simulated soccer robot in the face of complex situations, enhance adaptability, and is easy to optimize.

[0003] After searching, there is insufficient research on the technical solution for optimizing the simulation soccer robot evaluator. For example, the Chinese patent application number CN201910541349.9, the invention name is: Soccer robot defense strategy based on reinforcement learning. The main research problem of this application is based on the half-court offense platform HFO environment, and there is no in-depth study of the evaluator mechanism that is crucial in soccer robot competitions. Therefore, there is still a need for an evaluation method that is not highly dependent on expert systems, has the ability to cope with complex game situations, is highly adaptable, and is easy to optimize. Summary of the invention

[0004] The purpose of the present invention is to provide a simulation football robot evaluation method based on an LSTM network optimized by a genetic algorithm, which is used to solve the technical problems existing in the current simulation football robot team, such as reliance on expert systems, inability to cope with complex situations, poor adaptability and difficulty in optimization.

[0005] To achieve the above object, the present invention adopts the following technical solution:

[0006] A simulation soccer robot evaluation method based on an LSTM network optimized by a genetic algorithm, the method comprising the following steps:

[0007] S1. Based on the simulated football platform, a football robot team program is established, where the football robot team program consists of two teams, each team includes 11 heterogeneous robot agents;

[0008] S2. Based on the simulated football platform, the football robot team program, and according to the platform characteristics and the original evaluation mechanism, a basic device evaluation module is constructed;

[0009] The input parameters of the basic evaluator module include the currently predicted preferred action and the subsequent possible action sequence, and the output is the evaluation score of the currently predicted preferred action;

[0010] S3, based on the interaction between the simulated football platform, the football robot team program and the basic evaluator module, the original training data of the football robot team is obtained, and the original training data is processed by missing values, dimension reduction and data enhancement to construct a training set and a test set;

[0011] S4. Based on the long short-term memory network algorithm framework, build an evaluation network model of the soccer robot team program, and define the input and output dimensions, number of network layers, number of hidden units, number of fully connected layers, number of neurons in the fully connected layers, learning rate, optimizer, training batch size, number of iterations, loss function, and activation function of the evaluation network model;

[0012] S5. Based on the evaluation network model, a genetic algorithm optimization model is constructed, and the number of network layers, the number of neurons in each layer, and the parameters of the fully connected layer of the evaluation network model are used as optimization model inputs. In the genetic algorithm optimization model, a fitness function is defined to guide genetic operations, including the execution of selection, crossover, and mutation, and the evaluation network model structure parameters are updated by evaluating the fitness of individuals;

[0013] S6, updating the evaluation network model by optimizing the updated network structure parameters based on the genetic algorithm, that is, the evaluation network model is rebuilt and trained according to the updated network structure parameters, and the optimal evaluation network model is continuously updated through set rounds of iterative training;

[0014] S7. Based on the evaluation weights output by the optimal evaluation network model finally selected, the basic evaluator module is interacted with to obtain an optimized evaluator module. The optimized evaluator module will be used in the soccer robot competition to help the soccer robot select the optimal action parameters in real time during the competition.

[0015] Furthermore, the training set described in step S3 includes features including at least real-time position parameters of the simulated football robot, real-time speed parameters of the simulated football robot, real-time position parameters of the ball, real-time speed parameters of the ball, real-time action parameters taken by the simulated football robot, and real-time physical parameters of the simulated football robot.

[0016] Furthermore, the action features in the original training data of the soccer robot team in step S3 need to be processed by a multi-classifier based on a support vector machine (SVM) and a synthetic minority oversampling technique (SMOTE); the minority class samples are oversampled using the synthetic minority oversampling technique to generate new samples to balance the category distribution, and then the balanced data set is subjected to multi-classification operations through the support vector machine; based on the classification results, the training data that best matches the characteristics of the current evaluation network model is selected, and these data are marked as positive samples to enhance the evaluation network's learning ability for the target features, thereby improving the training effect of the model.

[0017] The multi-classifier based on support vector machine is a method for extending the support vector machine algorithm to multi-classification tasks, which is extended by using the "one-vs-many" (OvR) method, and its objective function is as follows:

[0018]

[0019] Among them, i represents the current classifier, j represents the jth data sample in the training set, and y ij is the label of sample j for classifier i, w i is the weight of the i-th class, b i is the bias of the i-th class, and C is the regularization parameter, which is used to weigh the complexity of the model and the penalty for classification errors;

[0020] The final decision function of the multi-classifier based on support vector machine is as follows:

[0021]

[0022] The synthetic minority class oversampling technique generates new samples of the minority class by interpolation, and the specific method is as follows:

[0023] x new =x m +λ·(x n -x m )

[0024] Among them, x m and x n is a minority class sample, x n is x m The nearest neighbor sample of x, λ is a random number. m and x n Linear interpolation is performed on the vectors between them, and new sample points x are generated according to the random proportion of λ. new , thereby expanding the number of minority class samples in the training dataset.

[0025] Furthermore, the evaluation network model described in step S4 includes a short-term memory unit and a long-term memory unit;

[0026] The short-term memory unit is used to capture the short-term time or state changes that occur during the program operation, that is, short-term information centered on the current moment. The short-term information at least includes: the short-term movement of the player, the passing of the ball, the short-term physical changes of the player, and the short-term atomic actions taken by the player, including turning, sprinting, kicking, tackling and diving.

[0027] The long-term memory unit is used to store the accumulated information or trends over a longer period of time during the program operation, that is, the long-term information with a longer time scale from the current moment. The long-term information at least includes: the cumulative running amount of the player, the cumulative physical consumption of the player, and the cumulative number of offensive or defensive cycles, that is, the number of cycles that the player is in the opponent's half and our half.

[0028] Furthermore, the loss function of the evaluation network model described in step S4 adopts a sparse classification cross entropy loss function, which is specifically implemented as follows:

[0029]

[0030] Among them, y j represents the true label of the jth sample, represents the predicted probability distribution of the evaluation network for the jth sample, and N is the number of samples.

[0031] Furthermore, the optimizer for evaluating the network model in step S4 adopts the Adam optimizer, wherein the parameter update of the Adam optimizer is implemented as follows:

[0032]

[0033] Mean shift of gradient:

[0034] m t =β1·m t-1 +(1-β1)·g t

[0035]

[0036] Bias correction:

[0037]

[0038] Where t represents the time step, θ t Indicates the model parameters of the network model evaluated at the current moment, g t represents the gradient of the loss function relative to the parameters in the current round of training, a represents the learning rate, which determines the magnitude of the change in network parameters for each update evaluation, and m t 、v tThey are the exponential moving averages of the first moment and the second moment of the calculated gradient at the current moment, respectively, which are used to smooth and adjust the update speed of the gradient.

[0039] Furthermore, the genetic algorithm described in step S5 acts on the hyperparameter optimization of the evaluation network model through selection, crossover, mutation, and population update, so as to find the optimal structure configuration of the evaluation network model;

[0040] The goal of the selection operation is to select individuals with high fitness from the current population according to the value of the fitness function to generate a new population, which is specifically expressed as follows:

[0041]

[0042] where P(k) is the probability that individual k is selected, and DNA k represents the chromosome of the k-th individual. Each DNA k chromosome represents specific network structure parameters. Each chromosome contains several gene loci, and f(DNA k ) is the fitness of individual k, and POP size is the population size;

[0043] The crossover operation is used to combine the chromosomes of two individuals to generate new offspring, and the specific process is expressed as follows:

[0044]

[0045] where DNA child represents the chromosome of the generated offspring individual, which is the network structure parameter. DNA parent1 and DNA parent2 represent two parent individuals used to generate the offspring individual. Mask is a binary vector. The value of the vector being 1 means inheriting the chromosome from parent1, and being 0 means inheriting the chromosome from parent2;

[0046] The mutation operation increases the diversity of the population by randomly changing the gene loci in the individual chromosome, which is specifically expressed as follows:

[0047]

[0048] where rand(range) represents randomly generating a new value within the specified range range. When rand() < MUTATION_RATE, the mutation operation is triggered, and the gene locus of the offspring is set to rand(range). MUTATION_RATE is the mutation probability defined when initializing the genetic algorithm parameters.

[0049] Furthermore, the next generation population in the population update operation is composed of individuals after selection, crossover and mutation. The process is specifically expressed as follows:

[0050] POP new =Selection(POP old )+Crossover(POP selected )+Mutation(POP crossed )

[0051] Among them, POP new Represents the population of the new generation, Selection(POP old ) represents individuals with high fitness selected from the previous generation population, Crossover (POP selected ) means that the selected individuals are cross-operated, paired with each other, and new individuals are generated. crossed ) represents the mutation operation, which randomly changes the gene loci of the generated individuals to increase the diversity of the population.

[0052] Compared with the prior art, the simulation soccer robot evaluation method based on the LSTM network optimized by genetic algorithm described in the present invention has the following beneficial effects:

[0053] (1) The present invention combines a genetic algorithm with a long short-term memory network, thereby effectively optimizing the evaluator, so that the evaluator can more accurately reflect the decision-making needs of the soccer robot in a complex dynamic environment, thereby improving the overall decision-making accuracy.

[0054] (2) By adopting the short-term memory unit and long-term memory unit of the long short-term memory network, the evaluator of the present invention can capture the short-term and long-term information in the simulated football game, ensuring that the decision-making of the football robot in different scenarios has higher adaptability, thus avoiding the lack of adaptability commonly seen in traditional evaluation methods.

[0055] (3) The present invention utilizes a genetic algorithm to continuously iteratively optimize the network structure, so that the parameter configuration of the evaluator model can be automatically adjusted according to changes in the competition environment, thereby simplifying the system optimization process, reducing dependence on expert systems, and solving bottlenecks that are difficult to optimize in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following is a brief introduction to the drawings required for the description of the embodiments. The drawings are not intended to be drawn to scale, and for the sake of clarity, not every component will be marked in each figure. The drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. Among them:

[0057] Figure 1 A flowchart of a simulation soccer robot evaluation method based on an LSTM network optimized by a genetic algorithm provided in an embodiment of the present invention;

[0058] Figure 2 A schematic diagram of the backbone network structure of an LSTM network optimized based on a genetic algorithm in an embodiment of the present invention;

[0059] Figure 3 It is a complete framework diagram of the simulated football platform, simulated football robot program, network model and genetic algorithm optimization module in the embodiment of the present invention. DETAILED DESCRIPTION

[0060] The embodiment of the present invention provides a simulated soccer robot evaluation method based on an LSTM network optimized by a genetic algorithm, and solves the technical problems that the evaluation method of the simulated soccer robot team in the prior art has the characteristics of relying on expert systems, being unable to cope with complex situations, having poor adaptability, and being difficult to optimize.

[0061] 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 and embodiments. It should be understood that the specific embodiments described herein are only used for the present invention and do not limit the present invention.

[0062] The embodiment of the present invention provides a simulation soccer robot evaluation method based on an LSTM network optimized by a genetic algorithm. Figure 1-3 , the method comprising:

[0063] S1. Based on the simulated football platform, a football robot team program is established, wherein the robot team program consists of two teams, and each team includes 11 heterogeneous robot players.

[0064] The simulated football platform is RCSSSERVER, and a simulated football robot team program is built based on the basic parameter interface of the football field and players provided by the simulated football platform.

[0065] S2, based on the simulated football platform and the football robot team program, and in accordance with the platform characteristics and the original evaluation mechanism, builds a basic evaluation module.

[0066] Among them, the scoring mechanism of the evaluation module of the football robot includes at least the following scenarios: 1. When the ball is in the opponent's half and the current football robot player is in control of the ball, a positive evaluation score is returned based on the judgment of the position parameters of the opponent's football robot player and the own football robot player on the field at this time; 2. When the ball is in the own half and the current football robot player is defending the player in control of the ball, a negative evaluation score is returned based on the judgment of the position parameters of the own football robot player as the goalkeeper and other own football robot players participating in the defense on the field at this time; 3. When the ball is in front of the opponent's goal and the current football robot player is in control of the ball, a very large positive evaluation score is returned based on the judgment of the position parameters of the opponent's football robot player and the own football robot player on the field at this time; 4. When the ball is in front of the own goal and the current football robot player is defending the player in control of the ball, a very small negative evaluation score is returned based on the judgment of the position parameters of the own football robot player as the goalkeeper and other own football robot players participating in the defense on the field at this time.

[0067] S3, the simulated football platform interacts with the football robot team program and the basic evaluation module to obtain original training data, perform missing value processing, dimension reduction and data enhancement processing on the original data, and construct a training set and a test set.

[0068] Among them, the multi-classifier based on support vector machine (SVM) and synthetic minority oversampling technology (SMOTE) are used to process the action features in the training data. The SMOTE technology is used to oversample the minority class samples and generate new samples to balance the class distribution. Then, the balanced data set is multi-classified by SVM. According to the classification results, the training data that best matches the current evaluation network features are selected and marked as positive samples. By increasing the proportion of positive samples in the training data set, the evaluation network's ability to learn the target features can be enhanced, thereby improving the training effect of the model.

[0069] S4, based on the long short-term memory network algorithm framework, build an evaluation network model of the soccer robot team program, and define the input and output dimensions, number of network layers, number of hidden units, number of fully connected layers, number of neurons in the fully connected layers, learning rate, optimizer, training batch size, number of iterations, loss function and activation function of the network model.

[0070] Among them, the number of network layers, the number of hidden units, the number of fully connected layers, and the number of neurons in the fully connected layers are initialized in the chromosomes in the genetic algorithm respectively. The number of network layers is initially a random number, the optimizer uses the Adam optimizer, the loss function uses the sparse classification cross entropy loss function, and the activation function uses the ReLU activation function.

[0071] S5, based on the evaluation network model, a genetic algorithm optimization model is constructed, and the number of network layers, the number of neurons in each layer, and the parameters of the fully connected layer of the evaluation network are used as model inputs. During the execution of the genetic optimization algorithm, a fitness function is defined to evaluate the performance of each candidate network structure. The output of the fitness function determines the retention and reproduction probability of an individual in genetic operations (selection, crossover, mutation). Specifically, by evaluating the fitness of each individual, the genetic algorithm determines which network structure parameters are more likely to be retained for the next generation of iterative training. Ultimately, the genetic algorithm continuously optimizes and updates the network structure parameters.

[0072] Among them, the selection operation adopts the roulette method. According to the fitness value of each individual, the probability of being selected is calculated, and then individuals are randomly selected from the population to form a new population. The probability of being selected is proportional to the fitness. The crossover operation adopts a single-point crossover, randomly selects a chromosome for crossover, and then uses the Boolean value True or False to indicate whether the corresponding gene site on the chromosome is exchanged. The probability of crossover is determined by the parameter CROSS_RATE. The mutation operation adopts random mutation. Each gene site has a certain probability of mutation. The probability of occurrence is determined by the parameter MUTATION_RATE. The specific operation of mutation is to randomly change the value of the gene site.

[0073] S6, updating the evaluation network based on the network structure parameters updated by genetic algorithm optimization, the evaluation network is reconstructed and trained according to these network structure parameters, and the optimal evaluation network is continuously updated according to the performance on the test set each time by setting rounds of iterative training.

[0074] Among them, the reconstruction and training of the evaluation network are mainly completed by the fitness function. In each generation, the genetic algorithm generates a new set of network structure parameters, that is, the final output chromosome. For each individual in each generation, an evaluation network is reconstructed and retrained using its corresponding network structure parameters. The network is trained using the training set provided in step 3. The goal of the training is to minimize the loss function. After the training is completed, the network performance is evaluated using the test set provided in step 3 to obtain the fitness value. With multiple iterations, the network structure parameters are continuously optimized, and the performance of the evaluation network is gradually improved.

[0075] S7, based on the evaluation weights output by the optimal evaluation network finally selected, the basic evaluator module is interacted with to obtain an optimized evaluator module. The optimized evaluator module will act on the soccer robot game to help the soccer robot select the optimal action parameters in real time during the game, thereby providing more accurate decision support for the soccer robot.

[0076] In summary, the simulation football robot evaluation method based on the LSTM network optimized by genetic algorithm provided by the embodiment of the present invention establishes a football robot team program including multiple heterogeneous football robot players through a simulation football platform, and constructs a basic evaluation module of the football robot according to the simulation football platform, the football robot team program, the platform characteristics and the basic evaluation mechanism, the simulation football platform interacts with the football robot team to obtain the original data, and obtains the training data set and the test data set through missing value processing, dimensionality reduction and data enhancement processing, and builds an evaluation network model based on the long short-term memory network model, and defines the input and output dimensions, the number of network layers, the number of hidden units, the number of fully connected layers, the number of neurons in the fully connected layers, the learning rate, the optimizer, the training batch size, the number of iterations, the loss function and the activation function of the evaluation network model, and constructs a genetic algorithm optimization model based on the evaluation network model, and uses the number of network layers, the number of neurons in each layer and the parameters of the fully connected layer of the evaluation network as the model input, and adopts an appropriate The fitness function evaluates the performance of each candidate network structure, and the output of the fitness function determines the probability of retention and reproduction of individuals in genetic operations (selection, crossover, mutation), thereby continuously optimizing and updating the network structure parameters, and updating the evaluation network based on the network structure parameters optimized and updated by the genetic algorithm. The evaluation network is rebuilt and trained according to these network structure parameters, and the optimal evaluation network is continuously updated according to the performance on the test set each time by setting rounds of iterative training. The evaluation weights output by the optimal evaluation network finally selected are interacted with the basic evaluator module to obtain the optimized evaluator module. The optimized evaluator module will be used in the football robot game to help the football robot select the optimal action parameters in real time during the game, thereby providing more accurate decision support for the football robot, simplifying the optimization difficulty of the football robot evaluator, and solving the technical problems of the simulation football robot evaluation method in the prior art, such as dependence on expert systems, inability to cope with complex situations, poor adaptability and difficulty in optimization.

Claims

1. A simulation soccer robot evaluation method based on LSTM network optimized by genetic algorithm, characterized in that: The method comprises the following steps: S1. Based on the simulated football platform, a football robot team program is established, where the football robot team program consists of two teams, each team includes 11 heterogeneous robot agents; S2. Based on the simulated football platform, the football robot team program, and according to the platform characteristics and the original evaluation mechanism, a basic device evaluation module is constructed; The input parameters of the basic evaluator module include the currently predicted preferred action and the subsequent possible action sequence, and the output is the evaluation score of the currently predicted preferred action; S3, based on the interaction between the simulated football platform, the football robot team program and the basic evaluator module, the original training data of the football robot team is obtained, and the original training data is processed by missing values, dimension reduction and data enhancement to construct a training set and a test set; S4. Based on the long short-term memory network algorithm framework, build an evaluation network model of the soccer robot team program, and define the input and output dimensions, number of network layers, number of hidden units, number of fully connected layers, number of neurons in the fully connected layers, learning rate, optimizer, training batch size, number of iterations, loss function, and activation function of the evaluation network model; S5. Based on the evaluation network model, a genetic algorithm optimization model is constructed, and the number of network layers, the number of neurons in each layer, and the parameters of the fully connected layer of the evaluation network model are used as optimization model inputs. In the genetic algorithm optimization model, a fitness function is defined to guide genetic operations, including the execution of selection, crossover, and mutation, and the evaluation network model structure parameters are updated by evaluating the fitness of individuals; S6, updating the evaluation network model by optimizing the updated network structure parameters based on the genetic algorithm, that is, the evaluation network model is rebuilt and trained according to the updated network structure parameters, and the optimal evaluation network model is continuously updated through set rounds of iterative training; S7. Based on the evaluation weights output by the optimal evaluation network model finally selected, the basic evaluator module is interacted with to obtain an optimized evaluator module. The optimized evaluator module will be used in the soccer robot competition to help the soccer robot select the optimal action parameters in real time during the competition.

2. The simulation soccer robot evaluation method based on LSTM network optimized by genetic algorithm according to claim 1, characterized in that: The training set described in step S3 includes at least the following features: real-time position parameters of the simulated football robot, real-time speed parameters of the simulated football robot, real-time position parameters of the ball, real-time speed parameters of the ball, real-time action parameters taken by the simulated football robot, and real-time physical parameters of the simulated football robot.

3. The simulation soccer robot evaluation method based on LSTM network optimized by genetic algorithm according to claim 2, characterized in that: The action features in the original training data of the soccer robot team in step S3 need to be processed by a multi-classifier based on a support vector machine (SVM) and a synthetic minority class oversampling technique (SMOTE); The synthetic minority class oversampling technique is used to oversample the minority class samples to generate new samples to balance the class distribution, and then the balanced data set is subjected to multi-classification operation through the support vector machine. According to the classification results, the training data that best matches the characteristics of the current evaluation network model is selected, and these data are marked as positive samples to enhance the evaluation network's ability to learn the target features, thereby improving the training effect of the model.

4. The simulation soccer robot evaluation method based on LSTM network optimized by genetic algorithm according to claim 3 is characterized in that: The objective function of the multi-classifier based on support vector machine is as follows: Among them, i represents the current classifier, j represents the jth data sample in the training set, and y ij is the label of sample j for classifier i, w i is the weight of the i-th class, b i is the bias of the i-th class, and C is the regularization parameter, which is used to weigh the complexity of the model and the penalty for classification errors; The final decision function of the multi-classifier based on support vector machine is as follows: The synthetic minority class oversampling technique generates new samples of the minority class by interpolation, and the specific method is as follows: x new =x m +λ·(x n -x m ) Among them, x m and x n is a minority class sample, x n is x m The nearest neighbor sample of x, λ is a random number. m and x n Linear interpolation is performed on the vectors between them, and new sample points x are generated according to the random proportion of λ. new , thereby expanding the number of minority class samples in the training dataset.

5. The simulation soccer robot evaluation method based on LSTM network optimized by genetic algorithm according to claim 1, characterized in that: The evaluation network model described in step S4 includes a short-term memory unit and a long-term memory unit; The short-term memory unit is used to capture the short-term time or state changes that occur during the program running, that is, the short-term information centered on the current moment, wherein the short-term information at least includes: the short-term movement of the player, the passing of the ball, the short-term physical changes of the player, and the short-term atomic actions taken by the player, and the atomic actions include turning, sprinting, kicking, tackling and diving; The long-term memory unit is used to store the accumulated information or trends over a longer period of time during program operation, that is, long-term information with a longer time scale from the current moment, wherein the long-term information at least includes: the player's cumulative running amount, the player's cumulative physical energy consumption and the cumulative number of cycles in offense or defense, that is, the number of cycles that the player is in the opponent's half and our half.

6. The simulation soccer robot evaluation method based on LSTM network optimized by genetic algorithm according to claim 5, characterized in that: The loss function of the evaluation network model described in step S4 adopts the sparse classification cross entropy loss function, which is specifically implemented as follows: Among them, y j represents the true label of the jth sample, represents the predicted probability distribution of the evaluation network for the jth sample, and N is the number of samples.

7. The simulation soccer robot evaluation method based on LSTM network optimized by genetic algorithm according to claim 6, characterized in that: The optimizer for evaluating the network model in step S4 adopts the Adam optimizer, wherein the parameter update of the Adam optimizer is implemented as follows: Mean shift of gradient: m t =β1·m t-1 +(1-β1)·g t Bias correction: Where t represents the time step, θ t Indicates the model parameters of the network model evaluated at the current moment, g t represents the gradient of the loss function relative to the parameters in the current round of training, a represents the learning rate, which determines the magnitude of the change in network parameters for each update evaluation, and m t 、v t They are the exponential moving averages of the first-order moment and the second-order moment of the gradient calculated at the current moment, which are used to smooth and adjust the update speed of the gradient.

8. The simulation soccer robot evaluation method based on LSTM network optimized by genetic algorithm according to claim 1, characterized in that: The genetic algorithm described in step S5 works together to optimize the hyperparameters of the evaluation network model through selection, crossover, mutation and population update, so as to find the optimal structural configuration of the evaluation network model; The goal of the selection operation is to select individuals with high fitness from the current population according to the value of the fitness function to generate a new population, which is specifically expressed as follows: Among them, P(k) is the probability of individual k being selected, DNA k represents the chromosome of the kth individual, each DNA k Chromosomes encode specific network structure parameters. Each chromosome contains several gene loci, f(DNA k ) is the fitness of individual k, POP size is the population size; The crossover operation is used to combine the chromosomes of two individuals to generate new offspring. The specific representation of the process is as follows: Among them, DNA child Represents the chromosome of the generated offspring individual, the chromosome is the network structure parameter, DNA parent1 and DNA parent2 Represents the two parent individuals used to generate offspring individuals. Mask is a binary vector. The vector value is 1, indicating that the chromosome is inherited from parent1, and 0, indicating that the chromosome is inherited from parent2. The mutation operation increases the diversity of the population by randomly changing the gene loci of the chromosomes of individuals, and is specifically expressed as follows: Among them, rand(range) represents randomly generating a new value within the specified range range. When rand() < MUTATION_RATE, the mutation operation is triggered, and the chromosome gene locus of the offspring is set to rand(range). MUTATION_RATE is the mutation probability defined when initializing the genetic algorithm parameters.

9. The simulation football robot evaluation method based on the LSTM network optimized by the genetic algorithm according to claim 8, wherein The next-generation population in the population update operation consists of individuals after selection, crossover, and mutation, and this process is specifically expressed as follows: POP new =Selection(POP old )+Crossover(POP selected )+Mutation(POP crossed ) Among them, POP new Represents the population of the new generation, Selection(POP old ) represents individuals with high fitness selected from the previous generation population, Crossover (POP selected ) means that the selected individuals are cross-operated, paired with each other, and new individuals are generated. crossed ) represents the mutation operation, which randomly changes the gene loci of the chromosomes of the generated individuals to increase the diversity of the population.

Images