Time series prediction based on ensemble fuzzy extreme learning machine

Abstract

Fuzzy time series has been utilized to solve the problem of time series prediction in diverse fields. In the fuzzy time series prediction, the fuzzy logical relations have great impacts on the prediction performance. To obtain the exact and complex fuzzy logical relations between the fuzzy variables for the time series prediction, we propose to extract them from the historical data and model them with the ensemble extreme learning machines. In addition, we focus on the variations of the real data instead of the actual one, which can describe the patterns of the time series more precisely. Our prediction method is tested on two actual time series comparing with four prediction methods. The results demonstrate the better performance of our method and the rationality to model the fuzzy logical relationships with the ensemble extreme learning machines.