Self-Feedback Differential Evolution Algorithm Based On Fitness Landscape And Its Application

The Optimization issues in the field of intelligent computing have always been a research topic for many scholars.However,in the development of society,many optimization problems have become more and more complex,they are more or less with non-continuous,non-linear,non-differentiable,multi-peak and so on.Because of this,a lot of traditional optimization algorithms are difficult to solve them.As a heuristic algorithm combined with computer science and biology,differential evolution algorithm has the advantages of fast convergence speed,less control parameters and so on which make it become one of the best algorithms to solve the optimization problems.Because of its sample principle,high performance,and soon a large number of scholars have been attracted to research and improvement,and now successfully applied to a number of areas.However,the traditional differential evolution algorithm also has some shortcomings,such as: 1)The convergence rate will slow down in the late generation and the local search ability is lacking,which make the algorithm cannot converge to the optimal solution of the problem in finite iteration.2)Setting different scaling factors F and hybridization probability C R has great influence on the algorithm.When the F value is small or the C R value is close to 1,the convergence rate is fast,but it is easy to fall into the local optimum.When the F value is large or the CR value is close to 0,although the diversity is guaranteed and the probability of jumping out of the local optimal is higher,but will reduce the convergence rate of the algorithm.3)There are many kinds of variation strategies,the characteristics of each variation strategy are not the same,therefore,according to different optimization problems to choose the appropriate mutation strategy has some difficulties.Aiming at the shortcomings of the differential evolutio n algorithm on the optimization problems,this paper proposes a new self-feedback differential evolution based on the fitness landscape(FLDE).The fitness landscape is the relationship between the solutions and the fitness value.When the evolutionary algorithm solves the complex optimization problem,the corresponding fitness landscape is usually very complex,generally composed of different local fitness landscape.Thus the variability of the local fitness landscape used by the mutation and crossover strategy should also be different.The proposed FLDE is to select the optimal crossover and mutation strategies by extracting the local fitness landscape features of each generation,in order to improve the search ability,accuracy and stability of the algorithm on different problems.The main achievements and innovations of this paper are as follows:(1)Using the Section Skew Tent chaotic map to initialize the population,so that individuals can be more evenly distributed in the solution space which improves the gene diversity of the initial population.(2)A new differential evolution algorithm based on fitness landscape is proposed.It uses a new hybrid mutation operator based on fitness landscape,which by combining the mutation operator of standard differential e volution algorithm and the fitness landscape.By extracting the local fitness landscape characteristics of each generation population and combining the characteristics of single or multi-peak in the fitness landscape,the differential variation strategy is selected so that the population can converge to the global optimal solution more quickly.(3)The scaling factor F and hybridization probability C R of each generation population were updated by using self-feedback techniques.This technique mainly updates them by collecting the mutation and the hybridization informations of each generation population,and then selecting the Gaussian distribution or Cauchy distribution to adjust them with the fitness features which improve the performance of this algorithm.(4)Using the FLDE in soil moisture equation optimization,from the simulation results,we consider that this algorithm convergence fast,has high precision and robustness,thus,it has some advantages to solve the complexity of the optimization problems.