An Overview of Forecasting Methods for Monthly Electricity Consumption

Abstract

Mid-term electricity consumption forecasting is analysed in this paper. Forecasting of electricity consumption is regression problem that can be defined as using previous consumption of an individual or a group with the goal of calculation of future consumption using some mathematical or statistical approach. The purpose of this prediction is multi beneficial to the stakeholders in the energy community, since this information can affect production, sales and supply. The Different methods are considered with the main goal to determine the best forecasting model. Considered methods include Box-Jenkins autoregressive integrated moving average models, state-space models and exponential smoothing, and machine learning methods including neural networks. An additional objective of the conducted research was to determine if modern methods like machine learning are equally precise in forecasting mid-term electricity consumption when compared to traditional time series methods. The performances of forecasting models are evaluated on the monthly electricity consumption data obtained using real billing software owned by the Distribution System Operator in Bosnia and Herzegovina. Mean absolute percentage error is selected as a measure of prediction accuracy of forecasting methods. Every forecasting method is implemented and tested using the R language, while data is collected from Data Warehouse in the form of total monthly consumption. The efficiency of presented solution will also be discussed after presentation of the results.