Deep learning based electricity demand forecasting to minimize the cost of energy imbalance: A real case application with some fortune 500 companies in Turkiye

Abstract

In this study, the electricity demands of some Fortune 500 companies in Turkiye have been forecasted by using deep learning techniques. This is a quite harder problem than the forecasting of the aggregated electricity demand in which the negative and positive fluctuations are absorbed on paper. Forecasting of firm-level electricity demand is an important problem since it can help automating firms' routine forecasting operations, reducing the electricity supply costs of the firms by improving the quality of the forecasts, and improving the quality of the electricity on the transmission network. Long Short-Term Memory (LSTM) and Multilayer Perceptron (MLP) techniques have been preferred concerning the successful results in the literature. As the originality of this paper, the Multiple Seasonal-Trend Decomposition using Loess (MSTL) technique is used for the electricity demand forecasting problem for the first time. The obtained results showed that although it is simple to implement, MSTL outperforms MLP and LSTM for most of the firms operating in mass production form. It is seen that the complexity of the model does not always guarantee good results and simple methods sometimes can work well. Load balancing studies are also very important for the economic sustainability of the industry since the electricity price and imbalance penalty have extremely increased (i.e., 8 times in Turkiye) during the post-pandemic period. Therefore, the energy cost reduction potential of the companies has also been assessed. This study resulted in cost savings of approximately 378 minimum wages for the pilot company.