Linking short-term electricity demand forecasting and explainable AI: A review for building energy applications

The energy sector necessitates strategies to minimise energy losses and enhance energy efficiency. Artificial intelligence has emerged as the most viable solution for forecasting tasks owing to its capability to predict energy consumption and generation patterns. It enables decision-makers to make more informed decisions, with regard to demand response programs (which rely on forecasting algorithms to align demand with supply or system operator requests). These applications often involve short-term forecasts (from 1 min to one day anticipation). This results in forecasting performances that rely on the accuracy of the training and target datasets and the parameterisation of the forecasting algorithms. This concept is presently related to explainable artificial intelligence (XAI). Numerous studies have evaluated this subject and presented various applications and challenges. However, these studies identify several deficiencies, particularly in short-term forecasting. These include the absence of focus on short-term scenarios and insufficiency of explainable algorithms for forecasting applications, particularly in buildings. Therefore, this research conducts a literature review on applications that generate explanations regarding the forecasting performance of energy demand for short periods in buildings with the support of XAI algorithms.