Day-Ahead Electricity Consumption Prediction of Individual Household–Capturing Peak Consumption Pattern
Zhong Xia, Ruiyuan Zhang, Hui Ma, Tapan Kumar Saha
Abstract
Day-ahead electricity consumption forecasting for individual residential consumers, especially the peak consumption time forecasting, is essential for home energy management system. However, it is considerably challenging since the single household consumption is highly volatile and stochastic and dependent on the underlying human behaviour. Further, the difficulties of accurately capturing the occurrence time of peak consumption in a house can limit the performance of existing machine learning-based load forecasting methods. In this paper, we propose a novel framework for day-ahead single-household electricity consumption forecasting by learning the peak consumption patterns of users. Instead of attempting to obtain the electricity consumption curve for the future 24 hours, the prediction of electricity consumption is achieved by combining the predicted base consumption, the predicted peak consumption occurrence time and the predicted amount of peak consumption within each time interval. The proposed framework can be used for both deterministic and probabilistic load forecasting of individual households. Case studies are conducted on hundreds of households from two different datasets. The results demonstrate that the performance of different deterministic load forecasting algorithms and probabilistic load forecasting algorithms can be improved after being integrated into the proposed framework.