Dynamic-objective particle swarm optimization for constrained optimization problems

This paper firstly presents a novel constraint-handling technique, called dynamic-objective method (DOM), based on the search mechanism of the particles of particle swarm optimization (PSO). DOM converts the constrained optimization problem into a bi-objective optimization problem, and then enables each particle to dynamically adjust its objectives according to its current position in the search space. Neither Pareto ranking nor user-defined parameters are involved in DOM. Secondly, a new PSO-based algorithm—restricted velocity PSO (RVPSO)—is proposed to specialize in solving constrained optimization problems. The performances of DOM and RVPSO are evaluated on 13 well-known benchmark functions, and comparisons with some other PSO algorithms are carried out. Experimental results show that DOM is remarkably efficient and effective, and RVPSO enhanced with DOM exhibits greater performance. In addition, besides the commonly used measures, we use histogram of the test results to evaluate the performance of the algorithms.