Time Series Forecasting Using Optimized Rolling Grey Model

Abstract

This study attempts to improve the forecasting accuracy of rolling grey model by applying Gaussian bare-bones differential evolution (GBDE) to optimize the weight of background value and number of data points used to construct a rolling-GM(1,1). Experimental results on two real time series forecasting problems show that the proposed GBDE-based rolling-GM(l,l) outperforms the traditional rolling-GM(l,l) in terms of fitting accuracy and forecasting accuracy.