The energy situation in Japan is rapidly changing due to the deregulation of electric power and efforts to achieve a zero CO2 emissions society. While efficient energy management is required especially in large-scale factories, energy demand is affected by various factors such as equipment operation and weather conditions, and is constantly fluctuating. Consequently, accurately forecasting energy demand based on these conditions is needed. In this report, a deep learning time-series forecasting model to determine electricity demand in a factory is considered. In addition, among explainable AI, which can provide an explanation for the variables considered important by the model and the basis for its forecast, XAI which can also explain features in the time-series manner is introduced here. Specifically, the global method TIME, which shows the overall model behavior, and the local method TimeSHAP, which shows the basis of forecasting by the model, can complement the explanatory properties of the model. The global method TIME will also be shown to improve the performance of the model by extracting explanatory variables considered important by the model and then restructuring the model with those variables.
