A new fuzzy-neural system for time series forecasting

Abstract

This paper proposes a new time series forecasting system, whose learning algorithm is a hybrid of the fuzzy c-means algorithm, the genetic algorithm, and the backpropagation algorithm. The proposed fuzzy-neural system consists of 5 layers: the input layer, the fuzzification layer, the rule layer, the hidden layer, and the output layer. The fuzzy cmeans algorithm is used to determine the center and width of the fuzzy membership functions. The artificial neural network is used as the fuzzy inference engine, while the genetic algorithm is used to optimize the fuzzy rule-base. This proposed system is tested with five time series data. The results obtained are very encouraging.