An evolutionary multi-layer perceptron neural network for solving unconstrained global optimization problems

Abstract

This study presents an evolutionary multi-layer perceptron neural network (EvoMLPNN) method, which consists of an MLPNN and an improved quantum-behaved particle swarm optimization (IQPSO) method. This study develops a network topology of an MLPNN that can be used to solve unconstrained global optimization (UGO) problems, and optimizes the weights of the MLPNN by using the IQPSO approach. To evaluate the performance of the proposed EvoMLPNN approach, a set of benchmark UGO problems was used and the numerical results obtained using the EvoMLPNN method were compared with those obtained using published algorithms. Experimental results show that the proposed EvoMLPNN method can find a global optimization solution for each test UGO problem and can solve highly dimensional UGO problems, and that the numerical results of the EvoMLPNN approach outperform to those of some published algorithms.