Swarm To Deal With The Study Of Constrained Optimization

In reality, constrained optimization problems(COPs) are widely exists in the areas of science, engineering, economy and national defence. But the optimization problems become more and more complex, traditional deterministic methods have represented its own limitation. So exploring more effective methods has been a hot spot. In recent years, intelligent optimization algorithm, especially evolutionary algorithm has made great process, and presented much better ability on constrained optimization in some documents. While particle swarm algorithm, as a new intelligent optimization algorithm, hasn't achieved the best ideal effect on constrained optimization problems. It is necessary to do further research on particle swarm algorithm for soving COPs.The aim of this paper is to improve particle swarm algorithm so that it can more effectively solve the COPs. This paper pays attention to the following three aspects:1. To balance the relationship between constraint satisfaction and objective optimization, ULPSO algorithm is proposed by incorporating infeasible personal best for each particle,the searching of particles is reinforcing on or near the boundary of the feasible region. Meanwhile, this paper analyzed the guidance effects of better-infeasible solutions in different processes. Finally, it has tested the validity of this method by simulation experiments.2. To avoid premature convergence which exists in the above algorithm., SLPSO algorithm is proposed by combining the probabilistic jumping property of simulated annealing to renew global best. The simulation experimental results show that the improved algorithm has strengthened the global searching ability. And it has overcome the shortage of premature convergence.3.ALPSO algorithm is proposed aim at solving the problem of hard to generate initial feasible solution in above algorithms on strong COPs, the updating of history best solution based on simple feasibility rules, meanwhile the new algorithm incorporated arithmetic crossover and dynamic social learning rate(c2). The simulation experimental results show that the new algorithm has improved the quality and efficiency of search when the feasible region is very small .