Efficient differential evolution algorithm-based optimisation of fuzzy prediction model for time series forecasting

Abstract

This paper proposes a differential evolution algorithm with efficient mutation strategy DEEMS for fuzzy prediction model FPM optimisation. The proposed DEEMS uses a modified mutation operation which considers local information nearby each individual to trade-off between the exploration ability and the exploitation ability. In the FPM design, we adopt an entropy measure method to determine the number of rules. Initially, there is no rule in the FPM. Fuzzy rules are automatically generated by entropy measure. Subsequently, the DEEMS algorithm is performed to optimise all the free parameters. During evolution process, the scale factor and crossover rate in the DEEMS algorithm are adjusted by adaptive parameter tuning strategy for each generation. It is thus helpful to enhance the robustness of the DEEMS algorithm. In the simulation, the proposed FPM with DEEMS model FPM-DEEMS is applied to two real world problems. Results show that the proposed FPM-DEEMS model obtains better performance than other algorithms.