Research On Dynamic Optimization Problem Based On Differential Evolution Algorithm

In the real world,many optimization problems exhibit dynamic properties,which the problem to be solved or the optimized objective function will change with time.It is of great practical significance to study algorithms that are suitable for solving dynamic optimization problems that are ubiquitous in the real world.However,the traditional optimization area mainly focuses on static optimization.This type of method is not suitable for solving dynamic optimization problems directly.Differential evolution algorithm has the natural advantage of solving dynamic optimization problems because of it is global optimization algorithm based on population.As the differential evolution algorithm has the advantages of simple principle,less controlled parameters,easy to understand and realize,it is very prominent in the problem of real number optimization.In recent years,it has become a hot topic in the field of dynamic optimization and intelligent computing.Therefore,this paper mainly studies the improved differential evolution algorithm for solving dynamic optimization problems.At present,the difficulty of dynamic optimization is:(1)The random change or partial change of environment causes the optimization algorithm not to be perceived in time.(2)The optimization algorithm can not effectively balance the diversity of search population and quickly track the optimal solution of the movement.In this paper,an improved differential evolution algorithm is proposed for these problems.The main work is as follows:(1)Combining multi-populations strategy with competitive strategy.Aiming at solving dynamic problems.A differential evolution algorithm combining multi-population method and competitive strategy is proposed(DECS).Firstly,a single population is used as a detection population.Using new detection methods.Secondly,the remaining multiple populations are used as search population and the algorithm runs independently and dynamically in search of dynamic environment.At the same time,the exclusion method is introduced.There is only one search population optimization for a locally optimal neighborhood.After several generations of iteration,Improving population diversity through competition between search populations.And make more full use of the limited cost of evaluation.Preserving the evaluation value of the search population with the best performance group.And the generation mechanism of quantum individual generation for its next generation.Re-initialization of other populations.Finally,we use the 49 dynamic problems of the generalized dynamic benchmark generator(GDBG)test set to verify the algorithm.The experimental results are compared with artificial immune algorithm(Dopt-aiNet),reset particle swarm optimization(rPSO)and modified differential algorithm(MDE).(2)Adaptive adjustment of parameters based on individual fitness.Determine the relative position of an individual in the search space based on individual evaluation values.Controls the size of the scalarness factor F depending on the relative position.To control the disturbance of individual genetic material.The size of crossover probability CR is also controlled according to the difference between the individual evaluation value and the average evaluation value of all individuals.Thus controlling the renewal probability of a population.It controls the convergence speed of the algorithm and avoids premature convergence.An re-adaptation differential evolution algorithm with competitive strategy based on multi-population(RDECS)based on this strategy.Finally,Moving peak benchmark(MPB)test set is used to verify the algorithm.The experimental results are compared with DECS algorithm,MDE algorithm and rPSO.The experimental results show that the MPB has no basis function.Performance is superior to all contrast algorithms.But when MPB has a base function,the Performance weaker than DECS.But it better than MDE and rPSO.(3)Adaptive strategy of improved parameters and mutation strategy based on simulated annealing.A new parameter adaptive strategy is proposed to control the values of the shrinkage factor and the crossover probability.Then study the mutation strategy based on simulated annealing.If the evaluation value of the best individual in the current generation is better than the value of the best individual in the previous generation.It shows that the algorithm plays an active role in the optimization of environment.It is possible to increase the weight of the DE/best/1 mutation policy at this time.Speed up convergence of algorithm to enhance local optimization ability.On the other hand,increase the weight of the DE/rand/1 mutation strategy.Increasing population diversity to enhance the algorithm's ability to explore the environment.Based on this,an improved self-adaption differential evolution algorithm(ADECS)is proposed.Finally,verify the algorithm on the MPB test set.The experimental results are compared with DECS,RDECS,MDE and rPSO.The experimental results show that the performance of ADECS algorithm is better than the contrast algorithm when MPB has no basis function and base function.