Differential Evolution Algorithms Using Constraint Handling Techniques

Constrained optimization is a common optimization model in engineering application field.Since the model often involves non-convex,non-differentiable or even discontinuous functions,and the search area is always limited by constrained functions,it is often difficult for gradient-based single-point search methods to obtain global optimal solutions.Differential evolution algorithm is a kind of global search algorithms based on swarm intelligence search,which updates individuals through mutation,crossover and selection operations.The algorithm is simple in structure and easy to combine with other optimization techniques to improve the search efficiency.In this thesis we focus on the most common single-objective and multi-objective constrained optimization problems,two differential evolution algorithms are proposed to solve these two kinds of problems by designing efficient constraint processing techniques and improving evolutionary operators.The specific work is as follows:For the single objective constrained optimization problem,an improved differential evolution algorithm is proposed by improving the operator and designing constraint processing method.Firstly,in the early stage of the algorithm iteration,in order to enhance the exploration ability of the algorithm,an adaptive heuristic mutation operator with embedded individual information is constructed.Secondly,based on the idea of pattern search,a constraint handling technique is given,which can effectively reduce the constraint violation degree of infeasible points.Finally,the simulation results on 18 standard test functions show that the proposed algorithm is effective and robust.For the multi-objective constrained optimization problem,a multi-objective differential evolution algorithm is developed by presenting weight adjustment,constraint handling as well as local optimization.Firstly,based on hierarchical clustering method,the population is divided into several disjoint sub-populations,which are searched in parallel to keep the diversity of individuals.Secondly,a new constraint handling technique is given by using the uniform design method,which can make the infeasible solutions approach the feasible region.Finally,based on the golden section method of unary function,a local optimization method is designed to improve individual quality.Simulation experiments and comparison results show that these hybrid techniques improve effectively the performance of the algorithm for solving constrained multi-objective optimization problems.