Short-Term Load Forecasting by Machine Learning

Abstract

In the global energy transition, Taiwan government has legislated the law to require large-scale power consumers with the obligation to partially use renewable energy. Many companies choose to follow the regulation by purchasing green energy. To purchase the energy effectively, it is necessary to understand its own electricity consumption. In this paper, electricity load forecasting models are studied and compared. The impact of the holiday adjustment policy of Taiwan on the forecasting is investigated. Experimental results demonstrated that the recent, deep-learning technique LSTM achieved the best performance. On the 9-month test data, MAPE of the LSTM was 1.85%.