Time Series Machine Learning: Implementing ARIMA and Hybrid ARIMA-ANN for Electricity Forecasting Modeling

The aims of this paper are to develop a linear and nonlinear model in time series to forecast electricity consumption of the lowest household category in East Java, Indonesia. The installed capacity in the lowest household customer category has various power, i.e. 450 VA, 900 VA, 1300 VA, and 2200 VA. ARIMA models are family of linear model for time series analysis and forecasting for both stationary and non-stationary, seasonal and non-seasonal time series data. A nonlinear time series model is proposed by hybrid ARIMA-ANN, a Radial Basis Function using orthogonal least squares. The criteria used to choose the best forecasting model are the Mean Absolute Percentage Error and the Root Mean Square Error. The ARIMA best model are ARIMA ([1, 2], 1, 0) (0, 1, 0)12, ARIMA (0, 1, 1) (0, 1, 0)12, ARIMA (0, 1, 1) (0, 1, 0)12, ARIMA (1, 0, 0) (0, 1, 0)12 respectively. The ANN architecture optimum are ANN (2, 12, 1), ANN (1, 12, 1), ANN (1, 12, 1), and ANN (1, 12, 1). The best models are ARIMA ([1, 2], 1, 0) (0, 1, 0)12, ARIMA (0, 1, 1) (0, 1, 0)12, ANN (1, 12, 1), and ANN (1, 12, 1) in each category respectively. Hence, the result shows that a complex model is not always better than a simpler model. Additionally, a better hybrid ANN model is relied on the choice of a weighted input constant of RBF.