Student grouping method based on reinforcement learning and improved genetic algorithm
By introducing reinforcement learning and improving genetic algorithms into the student grouping method, and taking into account multi-dimensional factors, the problem of poor effectiveness of the existing grouping method is solved, significantly improving the quality of students' grouping and collaborative learning effect.
Patent Information
- Application Number
- CN202411520692.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-29
- Publication Date
- 2025-02-18
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing student grouping methods usually only consider homogeneity or heterogeneity, lack comprehensive consideration of multi-dimensional factors, resulting in poor coordination among members within the group and poor collaborative learning results.
Using a student grouping method based on reinforcement learning and improved genetic algorithms, generate high-quality by constructing student characteristics data sets containing multi-dimensional factors such as homogeneity, heterogeneity, gender, leadership, and empathy, and designing a minimum cost insertion strategy. The initial solution is to dynamically adjust the key parameters of the genetic algorithm to optimize grouping effects.
It significantly improves the quality of students' groups and the overall effect of collaborative learning. By comprehensively considering multi-dimensional factors, it ensures that the members of the group have complementary skills and knowledge structures, and improves the coordination and learning effect within the group.
Smart Images
Figure CN119476800A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of teaching and education, and in particular to a student grouping method based on reinforcement learning and improved genetic algorithm. Background Art
[0002] Collaborative learning is an educational method that emphasizes interaction and cooperation among students, completing tasks or solving problems through group cooperation. Compared with traditional teaching models, collaborative learning can not only improve students' learning outcomes, but also effectively cultivate their communication skills, teamwork spirit and problem-solving abilities. In collaborative learning, students can deepen their understanding of knowledge by discussing and sharing different viewpoints, and gain richer learning experiences from diverse teams. As a key link in collaborative learning, the rationality of student grouping directly affects the learning effect. Through scientific grouping strategies, it can be ensured that members of the group have complementary skills and knowledge structures, thereby maximizing the results of collaborative learning.
[0003] Existing student grouping methods usually only consider homogeneity or heterogeneity, and lack comprehensive consideration of multi-dimensional factors, such as leadership, gender, and cooperation preference. This single-standard grouping strategy limits the coordination of group members and the overall collaborative learning effect. In addition, many grouping methods using genetic algorithms rely on randomly generated initial solutions, which makes it difficult for the algorithm to find high-quality grouping solutions in the early stages, affecting the final optimization effect. Therefore, developing algorithms that can simultaneously optimize multiple grouping criteria is crucial to improving the effectiveness of collaborative learning. Summary of the invention
[0004] The purpose of the present invention is to provide a student grouping method based on reinforcement learning and improved genetic algorithm to address the shortcomings and defects in the prior art. This method solves the problem of poor effect caused by single standard grouping in the prior art by simultaneously considering multi-dimensional factors such as homogeneity, heterogeneity, gender, leadership and empathy. It not only dynamically adjusts the key parameters of the genetic algorithm, but also constructs a high-quality initial solution by designing a minimum cost insertion strategy, thereby significantly improving the quality of student grouping and the overall effect of collaborative learning.
[0005] To achieve the above purpose, the present invention adopts the following technical scheme: a student grouping method based on reinforcement learning and improved genetic algorithm, which specifically includes the following methods: data set construction: construct a student feature data set containing five dimensions, which include homogeneity, heterogeneity, cooperation preference, leadership and gender, describe and classify students according to these characteristics, and provide data support for the optimization of subsequent grouping algorithms; initial solution generation: in many studies that use genetic algorithms for student grouping, the generation of initial solutions usually relies on randomization methods. Although random initialization can promote the generation of population diversity and help avoid the algorithm from converging to a local optimal solution early, it has obvious limitations when dealing with large-scale problems. Specifically, the randomly generated initial solutions are often of low quality, which may cause the algorithm to require more generations in the subsequent iteration process to gradually improve the quality of the solution, thereby increasing the algorithm's computational cost and time consumption; genetic algorithm operation: use the genetic algorithm to optimize the initial grouping scheme; parameter dynamic adjustment: in each generation of the genetic algorithm, use the reinforcement learning method to dynamically adjust the crossover probability (P c ) and mutation probability (P m ) to achieve a balance between exploration and utilization, and further improve the global search capability and convergence speed of the algorithm.
[0006] Furthermore, a minimum cost insertion strategy is designed in the generation of the initial solution to construct the initial solution, which is specifically as follows: currently there are n students who need to be divided into m groups, the leadership ability of each student is evaluated and ranked, and the top m students with the most leadership ability are selected as leaders and arranged in their respective groups, and the remaining nm students are put into the candidate list to be inserted, and the cost is calculated according to the characteristics described above. Since the leader is inserted into the group in advance, the leader score is not considered here, and the gender score cannot be evaluated when the group is not full. Therefore, the cost calculation here only considers homogeneous characteristics and heterogeneous characteristics as well as cooperation preferences; calculate the cost cost(i,j) of the student i to be inserted and the student j in the group: cost(i,j)=homo(s i ,s j )+hete(s i ,s j )+(p ij +p ji ) / t, where homo(s i ,s j ) represents the score s of student i on a specific homogeneous characteristic i , and the score s of student j on a specific homogeneous characteristic j The degree of similarity, homo(s i ,s j ) is smaller, the higher the homogeneity, hete(si , s j ) represents the score s of student i on a specific heterogeneous feature i , and the difference degree with the score s of student j on a specific heterogeneous feature j . The smaller hete(s i , s j ) is, the higher the heterogeneity is, and p ij , that is, the degree of willingness of student i to cooperate with student j. The smaller p ij is, the higher the willingness to cooperate is. t is the number of people in the group. Therefore, the cost insertcost(s i ) of inserting student s k into group g k (assuming the number of people in the current group g i is c, 1 ≤ c < t) is as follows: k ) is: j ∈ g k , calculate the insertion cost of all students in the candidate list to be inserted candidate = {s1, s2, s3,..., s n-m} inserted into all groups group = {G1, G2,..., G m}: Select the student i corresponding to the minimum insertion cost * , and insert it into group k * . That is, candidate = candidate / i * , k * = k * ∪ i * , and repeat the above steps until all students in candidate are assigned
[0007] Furthermore, the specific operations in the genetic algorithm operations include selection, crossover, and mutation: Selection: Select individuals with higher fitness to enter the next generation through roulette wheel selection, tournament selection, and elite selection methods; Crossover: Generate new individuals through partially mapped crossover (PMX) or order crossover (OX) methods; Mutation: The mutation operation is the key mechanism for introducing new genetic diversity in the algorithm
[0008] Furthermore, the selection operation is specifically as follows: the selection process of the genetic algorithm is a mechanism that simulates natural selection, and its purpose is to select individuals with higher fitness from the current population so that they can reproduce and pass on excellent genetic characteristics to the next generation. The selection process directly affects the performance and search ability of the genetic algorithm. Common selection operations include: Roulette selection: the probability of each individual being selected is proportional to its fitness. The higher the fitness of the individual, the larger the area it occupies in the roulette wheel and the higher the chance of being selected; Tournament selection: a small number of individuals are randomly selected. These individuals compete by comparing their fitness. The individuals with the highest fitness will win the game and be retained to the next generation; Elite selection: at the end of each iteration, a part of the best performing individuals in the current population are directly copied to the next generation.
[0009] Furthermore, the crossover operation is specifically as follows: the crossover operation exchanges genetic information between parent individuals to produce new offspring individuals. The purpose of the crossover operation is to increase the genetic diversity of the population and promote the combination of excellent genetic characteristics, thereby improving the global search ability of the algorithm and the quality of the solution. The single-point crossover cuts at a specific position of the chromosome and exchanges the fragments of the two parent chromosomes; the double-point crossover selects two different points on the chromosome for cutting and exchanges the fragments between the two points; the uniform crossover exchanges genes at corresponding positions with a fixed probability; however, the above crossover methods are not suitable for the problem of forming student groups, because the chromosome encoding in the student grouping problem generally takes the student as one of the points, and the non-repeatability of the student leads to As a result of the unavailability of the above-mentioned crossover operation, the research on student grouping using genetic algorithms generally adopts partial mapping crossover and sequential crossover, but these two methods rely entirely on randomness in the selection of points and cannot effectively optimize in the direction of improving the solution. First, the two groups with the worst adaptability are determined, and then crossover points are randomly selected between the two groups. This crossover point selection mechanism helps to retain those beneficial groups while promoting genetic recombination between the least adaptable groups in order to produce high-quality offspring. After determining the two least adaptable groups and randomly selecting intersection points, the gene sequence of the newborn offspring between the intersection points will be inherited from the parent generation, and as for the rest of the genes, the remaining genes from the other parent generation are filled on the chromosomes in their original order.
[0010] Furthermore, the mutation operation is specifically as follows: mutation helps the algorithm avoid falling into a local optimal solution and enhances the ability to search for a global optimal solution by randomly introducing small changes in the genetic code of an individual.
[0011] Furthermore, the dynamic adjustment of the parameters is specifically as follows: the fitness sum of the population, the population diversity, the fitness value of the best individual in the population and the number of iterations are mainly used as the state, then the state is expressed as follows: state = [f', d', m', current_iters], state is a four-dimensional vector, where current_iters represents the number of iterations of the current genetic algorithm, and the expression of f', d', m' is shown in the following formula: in, represents the fitness of the i-th individual in the population in the t-th iteration, represents the fitness of the i-th individual in the initial solution, Represents the fitness of the best individual in the population in the tth iteration.
[0012] Furthermore, the dynamic adjustment of parameters also includes action setting and reward setting: Action setting: In each evolution of the genetic algorithm, the trained reinforcement learning strategy will select P according to the current population state. c and P m , and then perform the corresponding evolution operation, setting 100 discrete actions, each action involves a pair of crossover probability and mutation probability, for P c , the commonly used value range is 0.1 to 0.6, the interval between each action is 0.05, a total of 10 values, the same method also applies to P m , its value range is 0.01~0.21, and the interval value is 0.02; Reward setting: the reward setting in this study is shown, where constant is 1000, represents the fitness of the best individual in the population in the tth iteration, The policy networks trained in this framework are all three-layer fully connected, with a size of 128×128×100, the activation function is ReLU, and the optimizer uses Adam.
[0013] After adopting the above technical scheme, the beneficial effects of the present invention are as follows: this method solves the problem of poor effect caused by single standard grouping in the prior art by simultaneously considering multi-dimensional factors such as homogeneity, heterogeneity, gender, leadership and empathy. It not only dynamically adjusts the key parameters of the genetic algorithm, but also constructs a high-quality initial solution by designing a minimum cost insertion strategy, thereby significantly improving the quality of student grouping and the overall effect of collaborative learning. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0015] Figure 1 It is the algorithm flow chart of the present invention.
[0016] Figure 2 It is a schematic diagram of the genetic algorithm in the present invention. DETAILED DESCRIPTION
[0017] This method constructs a student feature dataset containing five dimensions: homogeneity, heterogeneity, cooperation preference, leadership, and gender, and uses a minimum cost insertion strategy to generate a high-quality initial solution. The grouping effect is optimized by dynamically adjusting the key parameters of the genetic algorithm, and the effectiveness of the algorithm in student groups of different sizes is verified in experiments. The results show that this grouping strategy has significant advantages in improving intra-group consistency and student performance.
[0018] See also Figure 1 As shown, the technical solution adopted in this specific implementation is: it specifically includes the following methods:
[0019] S1, Dataset construction: Construct a student characteristic dataset containing five dimensions, including homogeneity, heterogeneity, cooperation preference, leadership and gender. Students are described and classified according to these characteristics to provide data support for the optimization of subsequent grouping algorithms.
[0020] S2, initial solution generation: In many studies that use genetic algorithms for student grouping, the generation of initial solutions usually relies on randomization methods. Although random initialization can promote the generation of population diversity and help avoid the algorithm from converging to the local optimal solution early, it has obvious limitations when dealing with large-scale problems. Specifically, the randomly generated initial solutions are often of low quality, which may cause the algorithm to require more generations in subsequent iterations to gradually improve the quality of the solution, thereby increasing the algorithm's computational cost and time consumption. A minimum cost insertion strategy is designed to construct the initial solution, which is as follows:
[0021] 1). Currently, there are n students to be divided into m groups. Evaluate and rank the leadership ability of each student, and select the top m students with the strongest leadership ability as leaders and arrange them into their respective groups. The remaining n - m students are placed in the candidate list for insertion. Calculate the cost according to the characteristics described above. Since the leaders have been inserted into the groups in advance, the scores of the leaders are not considered here. The gender score cannot be evaluated when the group is not full. Therefore, the cost calculation here only considers the homogeneous characteristics, heterogeneous characteristics, and cooperation preferences.
[0022] 2). Calculate the cost cost(i, j) between the student i to be inserted and the student j in the group:
[0023] cost(i, j) = homo(s i , s j ) + hete(s i , s j ) + (p ij + p ji ) / t
[0024] Among them, homo(s i , s j ) represents the similarity between the score s i of student i on a specific homogeneous characteristic and the score s j of student j on the same specific homogeneous characteristic. The smaller homo(s i , s j ) is, the higher the homogeneity. hete(s i , s j ) represents the difference between the score s i of student i on a specific heterogeneous characteristic and the score s j of student j on the same specific heterogeneous characteristic. The smaller hete(s i , s j ) is, the higher the heterogeneity. p ij is the degree of willingness of student i to cooperate with student j. The smaller p ij is, the higher the cooperation willingness. t is the number of people in the group;
[0025] Therefore, the cost insertcost(s i ), g k (assuming the current number of people in group g k is c, 1 ≤ c < t) for inserting the student s i into the group g k is:
[0026]
[0027] Calculate the list to be inserted candidate = {s1, s2, s3, ..., s n-m} into all groups group = {G1, G2, ..., G m The insertion cost in} is:
[0028]
[0029] 3) Select the student i with the minimum insertion cost * , inserted into group k * In
[0030] candidate=candidate / i * , k * =k * ∪i *
[0031] 4) Repeat steps 2 and 3 until all students in the candidate list have been assigned;
[0032] To sum up: an initial solution is obtained, but the initial population of the genetic algorithm contains multiple solutions, and other initial solutions are generated by partially transforming the generated initial solution.
[0033] S3, genetic algorithm operation: Use genetic algorithm to optimize the initial grouping scheme, see Figure 2 As shown, the specific operations include selection, crossover and mutation:
[0034] Selection: Individuals with higher fitness are selected to enter the next generation through roulette selection, tournament selection and elite selection methods, specifically:
[0035] The selection process of the genetic algorithm is a mechanism that simulates natural selection. Its purpose is to select individuals with higher fitness from the current population so that they can reproduce and pass on excellent genetic characteristics to the next generation. The selection process directly affects the performance and search ability of the genetic algorithm. Common selection operations include:
[0036] Roulette selection: The probability of each individual being selected is proportional to its fitness. Individuals with higher fitness occupy a larger area in the roulette wheel and have a higher chance of being selected.
[0037] Tournament selection: A small number of individuals are randomly selected and compete by comparing their fitness. The individual with the highest fitness will win the competition and be retained to the next generation.
[0038] Elite selection: At the end of each iteration, a portion of the best performing individuals in the current population are directly copied to the next generation. This study uses an elite strategy, retaining the top 10% of individuals to the next generation at the end of each iteration;
[0039] Crossover: Generate new individuals through partial mapping crossover (PMX) or sequential crossover (OX) methods, specifically:
[0040] The crossover operation exchanges genetic information between parent individuals to produce new offspring individuals. The purpose of the crossover operation is to increase the genetic diversity of the population and promote the combination of excellent genetic characteristics, thereby improving the global search ability of the algorithm and the quality of the solution. Single-point crossover cuts at a specific position of the chromosome and exchanges the fragments of the two parent chromosomes; double-point crossover selects two different points on the chromosome for cutting and exchanges the fragments between the two points; uniform crossover exchanges genes at corresponding positions with a fixed probability;
[0041] However, the above crossover method is not applicable to the problem of forming student groups. This is because the chromosome encoding in the student grouping problem generally takes students as one of the points, and the non-repeatability of students makes the above crossover operation unavailable. The research on student grouping using genetic algorithms generally adopts partial mapping crossover and sequential crossover, but these two methods completely rely on randomness in the selection of points and cannot effectively optimize in the direction of improving the solution. First, the two worst-adapted groups are determined, and then crossover points are randomly selected in these two groups. This crossover selection mechanism helps to retain those beneficial groups while promoting gene recombination between the least adapted groups in order to produce high-quality offspring. After determining the two least adapted groups and randomly selecting intersection points, the gene sequence of the newborn offspring between the intersection points will be inherited from the parent generation. As for the remaining part of the gene, the remaining genes from the other parent generation are filled on the chromosome in their original order. This method not only ensures genetic diversity, but also increases the possibility of producing offspring that are more adaptable to the environment.
[0042] Mutation: Mutation is the key mechanism for introducing new genetic diversity into the algorithm, specifically:
[0043] Mutation helps the algorithm avoid falling into local optimal solutions and enhances the ability to search for global optimal solutions by randomly introducing small changes in the genetic code of individuals. This method uses two-point exchange as mutation.
[0044] S4, dynamic parameter adjustment: In each generation of the genetic algorithm, the crossover probability (P c ) and mutation probability (P m ) to achieve a balance between exploration and utilization, and further improve the algorithm's global search capability and convergence speed. Specifically:
[0045] The state is mainly the sum of the fitness of the population, the population diversity, the fitness value of the best individual in the population, and the number of iterations. The state is expressed as follows:
[0046] state=[f',d',m',current_iters]
[0047] state is a four-dimensional vector, where current_iters represents the number of iterations of the current genetic algorithm, and the expressions of f', d', and m' are as follows:
[0048]
[0049] in, represents the fitness of the i-th individual in the population in the t-th iteration, represents the fitness of the i-th individual in the initial solution, Represents the fitness of the best individual in the population in the tth iteration.
[0050] It also includes action settings and reward settings:
[0051] Action setting: In each evolution of the genetic algorithm, the trained reinforcement learning strategy will select P according to the current population state. c and P m , and then perform the corresponding evolution operation, setting 100 discrete actions, each action involves a pair of crossover probability and mutation probability, for P c , the commonly used value range is 0.1 to 0.6, the interval between each action is 0.05, a total of 10 values, the same method also applies to P m , its value range is 0.01~0.21, and the interval value is 0.02;
[0052] Reward setting: The reward setting in this study is shown below, where the constant is 1000. represents the fitness of the best individual in the population in the tth iteration,
[0053]
[0054] The policy networks trained in this framework are all three-layer fully connected, with a size of 128×128×100, the activation function is ReLU, and the optimizer uses Adam.
[0055] The above description is only used to illustrate the technical solution of the present invention rather than to limit it. Other modifications or equivalent substitutions made to the technical solution of the present invention by ordinary technicians in this field should be included in the scope of the claims of the present invention as long as they do not depart from the spirit and scope of the technical solution of the present invention.
Claims
1. A student grouping method based on reinforcement learning and improved genetic algorithm, characterized by: It specifically includes the following methods: S1, Dataset Construction: Construct a student characteristic dataset containing five dimensions, including homogeneity, heterogeneity, cooperation preference, leadership and gender. Students are described and classified according to these characteristics to provide data support for the optimization of subsequent grouping algorithms; S2, initial solution generation: In many studies that use genetic algorithms for student grouping, the generation of initial solutions usually relies on randomization methods. Although random initialization can promote the generation of population diversity and help avoid the algorithm from converging to the local optimal solution early, it has obvious limitations when dealing with large-scale problems. Specifically, the randomly generated initial solutions are often of low quality, which may cause the algorithm to require more generations in the subsequent iteration process to gradually improve the quality of the solution, thereby increasing the algorithm's computational cost and time consumption; S3, genetic algorithm operation: using genetic algorithm to optimize the initial grouping scheme; S4, dynamic parameter adjustment: In each generation of the genetic algorithm, the crossover probability (P c ) and mutation probability (P m ) to achieve a balance between exploration and utilization, and further improve the global search capability and convergence speed of the algorithm.
2. The student grouping method based on reinforcement learning and improved genetic algorithm according to claim 1, characterized in that: In S2, a minimum cost insertion strategy is designed to construct an initial solution, which is specifically: 1) Currently, there are n students who need to be divided into m groups. The leadership ability of each student is evaluated and ranked, and the top m students with the best leadership ability are selected as leaders and arranged in their respective groups. The remaining nm students are put into the candidate list to be inserted. The cost is calculated according to the characteristics described above. Since the leader is inserted into the group in advance, the leader score is not considered here. The gender score cannot be evaluated when the group is not full. Therefore, the cost calculation here only considers homogeneous characteristics and heterogeneous characteristics as well as cooperation preference; 2) Calculate the cost cost(i,j) of the student i to be inserted and the student j in the group: cost(i,j)=homo(s i ,s j )+hete(s i ,s j )+(p ij +p ji ) / t Among them, homo(s i ,s j ) represents the score s of student i on a specific homogeneous characteristic i , and the score s of student j on a specific homogeneous characteristic j degree of similarity, homo(s i ,s j ) is smaller, the higher the homogeneity, hete(s i ,s j ) represents the score s of student i on a specific heterogeneous feature i , and the score s of student j on a specific heterogeneous feature j The degree of difference, hete(s i ,s j ) is smaller, the higher the heterogeneity, p ij That is, the intensity of student i's willingness to cooperate with student j, p ij The smaller it is, the higher the willingness to cooperate, and t is the number of people in the group; Therefore, the student s to be inserted i is inserted into the group g k (assuming the current number of people in group g k is c, 1 ≤ c < t), the cost insertcost(s i , g k ) is as follows: Calculate the list to be inserted candidate = {s1, s2, s3, ..., s n-m } into all groups group = {G1, G2, ..., G m The insertion cost in} is: 3) Select the student i corresponding to the minimum insertion cost * , inserted into group k * In candidate=candidate / i * ,k * =k * ∪i * 4) Repeat steps 2 and 3 until all students in the candidate list have been assigned.
3. The student grouping method based on reinforcement learning and improved genetic algorithm according to claim 1, characterized in that: The specific operations in S3 include selection, crossover and mutation: Selection: Roulette wheel selection, tournament selection and elite selection methods are used to select individuals with higher fitness to enter the next generation; Crossover: Generate new individuals through partial mapping crossover (PMX) or sequential crossover (OX) methods; Mutation: Mutation is the key mechanism for introducing new genetic diversity into the algorithm.
4. The student grouping method based on reinforcement learning and improved genetic algorithm according to claim 3 is characterized by: The selection operation is specifically as follows: The selection process of the genetic algorithm is a mechanism that simulates natural selection. Its purpose is to select individuals with higher fitness from the current population so that they can reproduce and pass on excellent genetic characteristics to the next generation. The selection process directly affects the performance and search ability of the genetic algorithm. Common selection operations include: Roulette selection: The probability of each individual being selected is proportional to its fitness. Individuals with higher fitness occupy a larger area in the roulette wheel and have a higher chance of being selected. Tournament selection: A small number of individuals are randomly selected and compete by comparing their fitness. The individual with the highest fitness will win the competition and be retained to the next generation. Elite selection: At the end of each iteration, a subset of the best performing individuals in the current population are directly copied to the next generation.
5. The student grouping method based on reinforcement learning and improved genetic algorithm according to claim 3 is characterized by: The crossover operation is specifically as follows: The crossover operation exchanges genetic information between parent individuals to produce new offspring individuals. The purpose of the crossover operation is to increase the genetic diversity of the population and promote the combination of excellent genetic characteristics, thereby improving the global search ability of the algorithm and the quality of the solution. Single-point crossover cuts at a specific position of the chromosome and exchanges the fragments of the two parent chromosomes; double-point crossover selects two different points on the chromosome for cutting and exchanges the fragments between the two points; uniform crossover exchanges genes at corresponding positions with a fixed probability; However, the above crossover method is not applicable to the problem of student group formation. This is because the chromosome encoding in the student grouping problem generally takes the student as one of the points, and the non-repeatability of the student makes the above crossover operation unavailable. The research on student grouping using genetic algorithms generally adopts partial mapping crossover and sequential crossover, but these two methods rely entirely on randomness in the selection of points and cannot effectively optimize in the direction of improving the solution. First, the two groups with the worst adaptability are determined, and then crossover points are randomly selected between the two groups. This crossover point selection mechanism helps to retain those beneficial groups while promoting genetic recombination between the least adaptable groups in order to produce high-quality offspring. After determining the two least adaptable groups and randomly selecting intersection points, the gene sequence of the newborn offspring between the intersection points will be inherited from the parent generation, and as for the rest of the genes, the remaining genes from the other parent generation are filled on the chromosomes in their original order.
6. The student grouping method based on reinforcement learning and improved genetic algorithm according to claim 3, characterized in that: The mutation operation is specifically as follows: mutation helps the algorithm avoid falling into a local optimal solution and enhances the ability to search for a global optimal solution by randomly introducing small changes in the genetic code of an individual.
7. The student grouping method based on reinforcement learning and improved genetic algorithm according to claim 1, characterized in that: The S4 is specifically: The state is mainly the sum of the fitness of the population, the population diversity, the fitness value of the best individual in the population, and the number of iterations. The state is expressed as follows: state=[f',d',m',current_iters] state is a four-dimensional vector, where current_iters represents the number of iterations of the current genetic algorithm, and the expressions of f', d', and m' are as follows: in, represents the fitness of the i-th individual in the population in the t-th iteration, represents the fitness of the i-th individual in the initial solution, Represents the fitness of the best individual in the population in the tth iteration.
8. The student grouping method based on reinforcement learning and improved genetic algorithm according to claim 1, characterized in that: The S4 also includes action settings and reward settings: Action setting: In each evolution of the genetic algorithm, the trained reinforcement learning strategy will select P according to the current population state. c and P m , and then perform the corresponding evolution operation, setting 100 discrete actions, each action involves a pair of crossover probability and mutation probability, for P c , the commonly used value range is 0.1 to 0.6, the interval between each action is 0.05, a total of 10 values, the same method also applies to P m , its value range is 0.01~0.21, and the interval value is 0.02; Reward setting: The reward setting in this study is shown below, where the constant is 1000. represents the fitness of the best individual in the population in the tth iteration, The policy networks trained in this framework are all three-layer fully connected, with a size of 128×128×100, the activation function is ReLU, and the optimizer uses Adam.