Research On Dynamic Optimization Problem With Unknown Number Of Peaks Based On Particle Swarm Optimization Algorithm

Many problems in the real world can be modeled as dynamic optimization problems whose characteristics are that the objective function,constraints or input variables change with time.The traditional particle swarm optimization algorithm applied to static optimization problems is no longer applicable.And many strategies such as memory strategy,prediction strategy and resource scheduling strategy focusing on the situation where the number of peaks is fixed can't apply to the case of unknown number of peaks.In view of this,the existing strategies are improved to solve the dynamic optimization problem with unknown number of peaks in this thesis.In this thesis,the existing dual memory strategy is first improved.When introducing the individuals in the long-term memory to the population,all the excellent individuals in the long-term memory are added,rather than randomly selecting several memory individuals.For individuals in the short-term memory set,an individual resource competition strategy based on individual fitness and the improvement of the fitness during the previous local search is proposed to determine the next individual to be searched.The experimental results show that the individual resource competition strategy effectively reduces the local search error and has certain advantages in the global search capability.Then the existing prediction strategy based on neural network is improved.By predicting the severity of environmental changes,the training set is screened to avoid the introduction of abnormal data.By labeling the unpaired solutions in two adjacent environments,the peaks that have just disappeared and appeared are identified to adapt the change of population number.The experimental results show that after the training set is screened,both the training error and the prediction error are reduced to a certain extent,and the population convergence speed is also improved in the problem with unknown number of peaks.At last,a resource scheduling strategy based on stochastic estimator is proposed.This strategy updates the probability that various population are selected according to the estimated value and the response of the environment,and the probability of each population is updated when the number of populations changes to avoid the estimator from converging to poorer populations.The experimental results show that this strategy can help the population to converge faster in the problem with unknown number of peaks.