Electricity consumption forecasting using DFT decomposition based hybrid ARIMA-DLSTM model

Abstract

Forecasting electricity consumption is vital, it guides policy makers and electricity distribution companies in formulating policies to manage production and curb pilfering. Accurately forecasting electricity consumption is a challenging task. Relying on a single model to forecast electricity consumption data which comprises both linear and nonlinear components produces inaccurate results. In this paper, a hybrid model using autoregressive integrated moving average (ARIMA) and deep long short-term memory (DLSTM) model based on discrete fourier transform (DFT) decomposition is presented. Aided by its superior decomposition capability, filtering using DFT can efficiently decompose the data into linear and nonlinear components. ARIMA is employed to model the linear component, while DLSTM is applied on the nonlinear component; the two predictions are then combined to obtain the final predicted consumption. The proposed techniques are applied on the household electricity consumption data of France to obtain forecasts for one day, one week and ten days ahead consumption. The results reveal that the proposed model outperforms other benchmark models considered in this investigation as it attained lower error values. The proposed model could accurately decompose time series data without exhibiting a performance degradation, thereby enhancing prediction accuracy.