Adaptive Fitness Function for Evolutionary Algorithm and Its Applications

Abstract

One of the popular methods of global optimization, the evolutionary algorithm (EA) is heuristic based and converges prematurely to a local-nonglobal solution sometimes. Our adaptive fitness function method, initially proposed for improving the validity of the evolutionary algorithm by avoiding this premature convergence, allows the evolutionary algorithm to search multiple, hopefully all, solutions of the problem. Every time the evolutionary search gets stuck around a solution, the proposed method transforms (or inflates) the fitness function around it so that the searching process can avoid coming back to this explored region in future search. Numerical results for some well known test problems of global optimization and mixed complementarity problems show that the method works very well in practice.