Research On Adaptive Differential Evolution Algorithm Based On Variable Coefficient And Pseudo Gradient Learning

Differential evolution algorithm is an evolutionary algorithm based on the difference of population,through the cooperation and competition among individuals in the population to solve the optimization problem.For the single-objective continuous optimization problem,the differential evolution algorithm has its own advantages.However,the DE algorithm also has some shortcomings,such as: strong individual similarity in the later stage of the search,poor population diversity,lack of effective evidence to judge the search stalled,weak local ability of mutation strategy,and slow algorithm convergence.Based on the existing research,this paper proposes a mutation strategy based on direction information and a mutation strategy based on pseudo-gradient learning.The mutation coefficient is introduced as an effective basis for judging whether the search is stuck or premature convergence,and the control parameters are adaptively adjusted.The main research contents are as follows:In view of the shortcomings of existing mutation strategies,excellent individuals cannot effectively use the direction information between population individuals to conduct mutation-guided evolution.When processing multimodal functions and complex functions,the algorithm often has difficulty converging to the global optimal solution.A new type of differential evolution algorithm(Adaptive Differential Evolution Algorithm Based on Restart Mechanism and Direction Information,abbreviated as ADERD)is used to search for the global optimal solution in continuous space.ADERD proposes a new mutation strategy based on direction information,which can make excellent individuals make full use of the direction information among individuals to mutate and guide evolution to find a better solution.In ADERD,the population is divided into two subpopulations based on individual error values,and only the top ranked individuals are mutated using the new mutation strategy.At the same time,the coefficient of variation is used for the first time to determine whether the search process has stalled or the algorithm hasprematurely converged.In order to verify the extensibility of the improvement measures proposed by ADERD,the above measures are integrated into the other two adaptive differential evolution algorithms(JADE_rcr and JADE_sort)to improve their performance.The improved JADE_sort algorithm is expressed as ADERD_sort.Compared with other influential international differential evolution algorithms,the ADERD_sort algorithm is competitive in the quality and stability of optimal solutions.In view of the existing mutation strategy can not use the effective movement between the individuals of the population to accelerate the convergence speed,there is a disadvantage of slow convergence speed,for some specific optimization problems,the algorithm is easy to premature convergence,this study proposes a new differential evolution algorithm Evolution Algorithm Based on Pseudo Gradient(abbreviated as ADEPG)is used to solve single-objective continuous optimization problems.In ADEPG,a new mutation strategy is constructed with the forward movement in the pseudo-gradient to accelerate the convergence rate and reduce the error when processing multi-peak and mixed functions.At the same time,ADEPG also uses a restart mechanism based on the coefficient of variation to try to avoid stagnation in search.The experimental results show that the mutation strategy based on pseudo-gradient and the restart mechanism based on mutation coefficient in ADEPG can also improve the performance of JADE_rcr algorithm,and ADEPG has a better optimization effect than other classical algorithms in solving global optimization problems.