Constrained particle swarm optimization using the filter approach

Penalty function approach is the common method for constrained optimization in engineering design. However, its results are sensitive to the penalty factor, and it is necessary to acquire the suitable penalty factor for specifically practical problems with many trials. To avoid the repeated parameter selection process, the filter constraint-handling mechanism and particle swarm optimization (PSO) are integrated for solving constrained optimization. The filter approach is based on the concept of domination from multi-objective optimization and the filter is constructed by a list of objective value and constraint violation pairs in which no pair is dominated by any other. Consequently, the constraint could be considered without using penalty function. In the evolution process of the constrained PSO using filter, the filters are constructed separately for historical optimum of each particle and historical optimum of the whole swarm, and the particle is updated using the optimum chosen from the corresponding filter with the feasible solution preferred comparison strategy. At last, the performance of the filter PSO algorithm is compared with the penalty PSO and genetic algorithm on some engineering design benchmarks. The simulation results show the filter PSO could find the more feasible, optimal and robust solution, and it is a effective method for constrained optimization.