Explainable deep learning approach to predict residential electricity demand

The escalating demand for electricity in residential buildings is a significant concern that necessitates a comprehensive understanding of the underlying reasons and responsible parameters. The electricity forecasting problem involves integrating the time series of weather conditions and electricity consumption into an intelligent model that can explain the influence of weather parameters on electricity consumption. Artificial intelligence techniques have shown excellent electricity prediction performance, but the crucial challenge is reliable and interpretable predictions. This paper proposes an intelligent electricity prediction approach by exploiting the potential of eXplainable Artificial Intelligence (XAI). The proposed electricity demand prediction model integrates a transformer-based long short-term memory model with genetic algorithms. Further, an XAI tool has been applied to interpret the prediction results and provide a deeper understanding of the factors influencing electricity consumption. Two experiments have been conducted to evaluate predictive performance, the first on a dataset of real-time electricity consumption and the second on a benchmark dataset of residential buildings. The proposed approach outperformed the other state-of-the-art models and achieved the lowest mean absolute error.