Deep learning based short-term load forecasting incorporating calendar and weather information

Short-term load forecasting has been an important approach for economical and sustainable power systems. Various methods have been proposed for obtaining an accurate forecasting result, among which deep learning models achieve state-of-the-art performance. While external factors have been considered in the modeling of the load forecasting process, there is a lack of comparison between the effect of calendar and weather information. In this letter, a TCN-based load forecasting model incorporating calendar and weather information is proposed and outperforms three deep learning and four machine learning baselines on an open real-world load dataset, with and without leveraging the calendar or weather information. It is found that weather information is more helpful for improving the load forecasting performance than calendar information through numerical experiments.