Missing Data Dynamic Forecasting of Fuzzy Time Series Based on Gaussian Mixture Model

Fuzzy time series model (FTS) has been widely used to forecast various time series data. When some data are missing, the forecasting accuracy will be greatly reduced. Therefore, a new forecasting method G-FTS is proposed based on the Gaussian mixture model and FTS. First, the distribution of the samples is estimated by using a Gaussian mixture model, the missing data are filled with the estimated distribution, and then a fuzzy time series forecasting is performed. The effectiveness of this method is illustrated by forecasting the monthly inventory data of small and medium-sized industrial enterprises in China. According to the characteristics of industrial data, such as time correlation, fast arrival, and large quantity, a dynamic forecasting method Dynamic-GFTS is proposed. This method is used to forecast the data collected by a textile factory. Simulation results show that Dynamic-GFTS is more effective in actual industrial data.