A comparative study on electricity load forecasting using statistical and deep learning approaches

Abstrak

Load forecasting has become reproving aspect of an energy management system (EMS). It gives basic advantage to grid stability, cost effectiveness and battery storage system (BSS). For this purpose, machine learning (ML) is widely adopted to forecast the electricity load. This research paper investigates the performances of various time series estimating models applied to electricity load data for an Irish company. The research mainly adopts the autoregressive integrated moving average (ARIMA) model, long short-term memory (LSTM) networks and transformer neural network (TNN) to forecast the electricity load. A comparison evaluation is conducted encompassing various quantifying measures such as root mean square error (RMSE), mean square error (MSE) and mean absolute error (MAE). The results are then compared to get an understanding whether the TNN using attention-based mechanism is better than the two state of the art models. Hence provides a complete understanding about which of the model needs improvements in its architecture for enhancement of operational efficiency and cost effectiveness in the realm of EMS.