Monthly energy forecasting using decomposition method with application of seasonal ARIMA

This paper presents a new forecasting approach for seasonal regressive time series which applies well-known autoregressive integrated moving average (ARIMA) method to classical decomposition techniques. The proposed technique starts with decomposing time series data into trend-cycle and seasonality components by using multiplicative decomposition. Then the seasonal autoregressive integrated moving average (SARIMA) is applied to the trend-cycle part to find the model that best describes it. The SARIMA trend-cycle is then combined with estimated seasonal component obtained separately to make a series of forecast values. The proposed forecasting approach is applied to monthly energy data of an electric distribution utility in Thailand. The results of the proposed technique are compared to those of the standard approach, which forecasts the trend-cycle component by projecting it using a mathematical function. The comparison shows that the decomposition forecasting with SARIMA trend-cycle is preferred