Research And Application Of Particle Swarm Optimization Based On Diversity Control Strategy

Particle Swarm Optimization(PSO)is a swarm intelligence optimization algorithm based on simulating the foraging behavior of bird swarms,and it is an emerging branch in the field of optimization.It has the advantages of simple implementation,easy to understand,and few control parameters.Therefore,as soon as the algorithm was proposed,it attracted wide attention of researchers.However,PSO has shortcomings of premature convergence.In view of this problem,it analyzes from the perspective of diversity maintenance and control,and proposes two improved PSO algorithms.This paper mainly includes the following research work:Firstly,aiming at the problem of particle swarm falling into premature convergence,a Random Drift Swarm Optimization with Dynamic Opposition based learning(DO-RDPSO)algorithm is proposed,which combines random drift strategy and opposition learning strategy.In the DO-RDPSO algorithm,when a random drift strategy is used for search,the number of individual particle optimal values stagnation exceeds a predetermined threshold,indicating that the Swarm diversity was too low and jumped out by random drift motion.The probability of local optimization is very small.At this time,it is not necessary to update by a random drift strategy but to perform opposition learning,construct the opposition solution of the particle,and increase Swarm diversity.It can be seen from the test of 14 functions that DO-RDPSO can jump out of the local optimum and find the global optimal solution.Secondly,aiming at the problem that particle swarm optimization is easy to fall into local optimization at high latitude or multi-extreme functions,a Crossover Operator of Random Drift Particle Swarm Optimization(CO-RDPSO)algorithm is proposed.The algorithm is based on PSO,adding random drift strategy and crossover operator.First,randomly select the two historically optimal positions of individuals to cross,so that the operation can improve the diversity of the Swarm and retain excellent particle information.Second,the threshold is used to determine whether to apply the crossover to obtain the new particle position instead of the original individual historical optimal position.Finally,the random drift strategy is used to update the speed and position.It can be drawn from the convergence accuracy and the convergence graph of the 14 test functions that CO-RDPSO not only quickly finds the global optimal solution,but also improves the convergence speed.Thirdly,to further verify the effectiveness of the CO-RDPSO algorithm,the short-term power load is predicted.CO-RDPSO has the characteristics of high solution accuracy and fast convergence speed,while avoiding falling into local optimum,and can optimize the SVM model quickly.By comparing the two models of SVM and CO-RDPSO-SVM,this paper proposes that CO-RDPSO performs hyperparameter optimization on the SVM regression estimation method to obtain the model,which can better reflect the power load situation.Aiming at the problem that particle swarm optimization is easy to fall into local optimization,this paper improves the particle swarm from the perspective of diversity maintenance and control,and proposes DO-RDPSO algorithm and CO-RDPSO algorithm respectively.At the same time,the two proposed algorithms are applied to 14 test functions.and compared with other improved particle swarm optimization algorithms to analyze the results.The experimental results show that both DO-RDPSO algorithm and CO-RDPSO algorithm have good optimization effects.