Multi-Objective Optimum Design of a Neuro-Fuzzy Network Using a Combined PSO and DE Algorithm Based on Fuzzy Logic

Because of the widespread application in complex modeling based on experimental data, neuro-fuzzy networks have attracted the attention of researchers. In the neuro-fuzzy inference system, the objective is to reduce the system's prediction error relative to the actual data. The regulation of parameters of neuro-fuzzy network is very important and affects its performance. So, a new optimization algorithm based on Particle Swarm Optimization (PSO) and Differential Evolution (DE) has been proposed. In this algorithm, the coefficients of the operator speed are calculated dynamically, using fuzzy logic. These coefficients are set according to the generation number and variance of the particles. Proposed operator leads the particles to explore and exploit the search domain more precisely. Next, the performance of the proposed algorithm is checked by optimizing three benchmarks and comparing it with the results, which are obtained by conventional PSO and DE. The results show that the proposed algorithm obtained better solution in comparison with DE and PSO and proved its performance and efficiency. Finally, a neuro-fuzzy system has been employed to forecast the time series of Mackey-Glass. The parameters of this neuro-fuzzy network are optimized by the new algorithm and the PSO and DE method multi-objectively and the Pareto charts obtained by each method of optimization are compared with each other, indicating the better performance of the new algorithm.