Based On Double Populations Of Improved Particle Swarm Optimization Algorithm

Because of its simple principle and easy realization, particle swarmoptimization (PSO) has become hot topic of swarm intelligence and hasbeen successfully applied into some aspects. In swarm intelligence theory,particle has stochastic feature and the accuracy of searching results cannot be assured. Besides, because it is simple, basic PSO is easily trappedinto local optima in the solution of complex problems and prematureconvergence easily happens.According to deficiency of PSO in optimization problems, a novelimproved PSO based on two swarms is proposed Main contributions ofthis thesis are given as following:(1) Based on basic PSO model, the influence of parameters on thealgorithm is analyzed in detail. Then, the idea of a two-swarm PSOalgorithm, which is based on different parameters, is proposed.(2) Basic PSO algorithm is susceptible to being trapped into localoptimum and premature convergence happens. According to deficiency ofPSO, a new two-swarm based PSO algorithm (TSPSO) with roulettewheel selection is proposed. Its mathematic description and computationprocedure are also given. The mechanics of information exchangebetween swarms is designed and parameters selection of the algorithm arestudied and tested. The proposed algorithm is tested on three benchmarktest functions. The results show that the proposed algorithm is superior toPSO and GA in the solution of complex optimization problems.(3) To further test engineering application of TSPSO, TSPSO isproposed to solve optimal power flow problem. To deal with constraints,adaptive penalty coefficients are designed, which can effectively balanceobjective function and constraints in the process of swarm evolution andmake particles searching from infeasible region to feasible region. Theproposed algorithm is tested on IEEE 30 bus system and the results showthat it can effectively solve optimal power flow problem. Because ofTSPSO based optimal power flow algorithm is a general procedure, it canbe used to solve large-scale optimal power flow problems.