Research On Genetic Algorithm Based On The Principle Of Minimum Entropy Production

Genetic Algorithm(GA)is a heuristic group intelligence algorithm developed by imitating Darwin's "survival of the fittest" evolutionary law.It directly operates on structural objects,has no requirements for execution of the target function,has high robustness and global search capability,and is highly scalable.The genetic algorithm is suitable for the solution of highly nonlinear optimization problems,and can be searched for global optimal solutions in complex multidimensional spaces,and thus has been widely used.However,similar to most evolutionary algorithms,the linear selection strategy of genetic algorithm makes the selection pressure of the algorithm too large,which leads to the loss of population diversity in the search process and easy to fall into local optimum.Aiming at this common problem,this paper takes genetic algorithm as a typical example.Under the inspiration of thermodynamic genetic algorithm TDGA,the evolution process of genetic algorithm is compared with the change process of system under non-equilibrium thermodynamics.The selection strategy based on thermodynamic principle is studied.(1)In the process of population evolution,the principle of minimum entropy increase under thermodynamic non-equilibrium steady state is introduced,and a thermodynamic selection strategy based on the principle of minimum entropy increase is proposed,so that the individual's choice is no longer completely dependent on the size of the fitness value.Individual density is defined within the objective function value space to measure population diversity.Firstly,a selection strategy based on greedy thermodynamics selection mechanism is proposed,but the time complexity is high.In order to reduce the time complexity,a selection strategy based on the component thermodynamic substitution mechanism is proposed according to the superposition of entropy.The population entropy is distributed to each individual,and the individual entropy generation is directly sorted and filtered to make the selection process.Has linear time complexity.(2)Applying the above two selection strategies to the improvement of genetic algorithm,and verifying the two strategies from multiple angles on the simple knapsack problem and the numerical function test problem respectively.The experimental results show that the application of non-The minimum entropy increase principle of thermodynamics in equilibrium can make the population maintain diversity while ensuring convergence speed and precision,which can effectively avoid the population falling into local optimum.It also verifies that the component thermodynamics selection mechanism has higher computational efficiency and can be reduced.time complexity.(3)In order to verify the universality of the selection strategy based on the principle of minimum entropy increase,and further verify the reliability of the selection strategy in diversity preservation,the component thermodynamics selection mechanism based on minimum entropy increase is applied to the multi-objective genetic algorithm.Improvement.In order to calculate the individual density more accurately,the kernel function density estimation is used to fit the individual distribution state in the target space.The population is stratified by non-dominated sorting,and the population diversity is maintained by the principle of minimum entropy increase.Finally,the experimental verification is carried out on the simple two-objective test problem(including the ZDT series problem and the MOP series problem).The experimental results show that the selection strategy based on the minimum entropy increase principle is applicable to the multi-objective optimization problem,especially in the diversity of the population.The effectiveness of sexual preservation.