Day-ahead Load Forecasting using Explainable Artificial Intelligence
Daniel Aunan Bolstad, Ümit Cali, Murat Kuzlu, Ugur Halden
Abstract
A real-time balance between power production and consumption is needed to ensure stable power flow and frequency. Due to the fact that electricity is needed to be consumed at a very instant time, accurate energy forecasting is required. Meanwhile, the recent high integration of Renewable Energy Sources (RES) into electrical grids introduces the accuracy problem in forecasts due to their inherited volatile nature. To overcome these challenges, digitalization of the energy sector started with implementation of emerging technologies such as Artificial Intelligence (AI), Information and Communication Technology (ICT),and the Internet of Things (IoT) for a better forecasting. However, the perceived black-box nature of AI gave rise to increased demand on understanding the inner working mechanics of them. This paper presents a use case of day-ahead load forecasting using an eXplainable Artificial Intelligence (XAI) tool, i.e., SHapley Additive exPlanations (SHAP), which helps understand the opaque, complex nature of the AI system by interpreting the feature importance.