A genetic-fuzzy-neuro model encodes FNNs using SWRM and BRM

Abstract

Genetic algorithms (GAs), fuzzy logic (FL), and neural networks (NNs) are frequently used artificial intelligence (AI) techniques. Since these three methods are complementary rather than competitive, many researchers have hybridized GAs, FL, and NNs to develop a better performance model. However, most hybrid models use a multistage combination or identify partial parameters required in the model resulting in sub-optimal solutions. This research fuses GAs, FL, and NNs to develop an evolutionary fuzzy neural inference model (EFNIM) that uses GAs to simultaneously search for all parameters required in fuzzy neural networks (FNNs). Two approaches, summit and width representation method (SWRM) and block-representation method (BRM), are proposed to encode variables in FL and NNs. Simulations are conducted to evaluate the performance of EFNIM. For different problems, membership functions (MFs) with the minimum FNN structure and optimal parameters of FNN are automatically and concurrently acquired using EFNIM. The research overcomes the difficulties faced in applying FL and NNs as well as saves efforts in trial-and-error experiments, questionnaire survey, interviews with experts, etc. Both prediction accuracy and time requirement for cost estimating are much improved by the proposed method.