Study On Particle Swarm Optimization Based With Dimensional Learning Strategy

In traditional particle swarm optimization algorithm(PSO),each particle updates its velocity and position with a learning mechanism based on its personal best experience and the population best experience.The learning mechanism in traditional PSO is simple and easy to implement,but it suffers some potential problems,such as the phenomena of “oscillation” and “two steps forward,one step back”.Therefore,designing an effective learning strategy to avoid these two phenomena and improve the search efficiency is an urgent issue for PSO research.In order to discover and protect the promising information of the population best solution,this paper proposes a dimensional learning strategy(DLS)for discovering and integrating the promising information of the population best solution according to the personal best experience of each particle.Dimensional learning strategy constructs a learning exemplar for each particle through each dimension of the particles' personal best solution learning from corresponding dimension of the population best solution.Therefore,the learning exemplar learns only from the dimension of the population best position that could improve its fitness,which ensures that the learning exemplar will not be degraded and hence,weaken the phenomenon of “two steps forward,one step back”.In dimensional learning strategy,each particle learns from gbest,which leads all particles to being close to gbest with strong exploitation ability,potentially resulting in premature convergence.To this end,we have introduced a diversity enhancement mechanism-comprehensive learning strategy(CLS).In comprehensive learning strategy,each dimension of particles learns from the optimal experience of different particles according to a certain random selection mechanism.This strategy enhances the population diversity and help the particles escape from local optima.Finally,a two-swarm learning particle swarm optimization(TSLPSO)algorithm with heterogeneous learning strategy based on dimension learning strategy and comprehensive learning strategy is proposed.One of the subpopulations constructs the learning exemplars by DLS to guide the local search of the particles,and the other subpopulation constructs the learning exemplars by the comprehensive learning strategy to guide the global search.To verify the effectiveness of the proposed algorithm,16 benchmark functions,30 CEC2014 test functions,and 1 real-world optimization problem are used to test the proposed algorithm against with 5 typical PSO algorithms and,1 champion algorithm--differential evolution(DE)algorithm.The experimental results show that TSLPSO has not only high optimization accuracy but also fast convergence speed on 16 benchmark functions.The average success rate of 16 test functions is 100%.At the same time,TSLPSO is also effective on CEC2014 test functions,and its overall performance is better than most traditional PSO algorithms.In addition,TSLPSO has ranked the first in the PSO algorithm for the problem of radar polly phase code design.The experimental results show that the proposed dimension learning strategy and two-swarm TSLPSO algorithm can effectively solve real engineering optimization problems.Wilcoxon sign rank test and Friedman test show that DLS and TSLPSO proposed in this paper are significantly better than PSO algorithm in convergence speed and accuracy.