Forecasting Day-Ahead PV Generation and Load Demand for an Individual Residential Consumer

Accurate day-ahead forecasting of PV generation and load demand for an individual consumer gets more attention as the deployment of PV-battery systems increases. The uncertainty in PV generation and load demand for an individual consumer requires optimal operation of batteries to increase flexibility. In this paper, a data-driven, long-short-term memory (LSTM)-based recurrent neural network (RNN) method is used for day-ahead prediction of load demand and PV generation. The proposed method uses historical data of a single consumer collected from a smart meter with a 5-minute time resolution to forecast the next-day PV generation and load demand. The performance of the proposed method is evaluated in terms of RMSE, MSE, and MAE values. The smallest value of RMSE is observed for PV generation prediction in the summer season, whereas for the load profile, it is observed in the winter. Based on the performance evaluation metrics, the data-driven LSTM model improves the accuracy of day-ahead forecasting of solar power output and load demand as compared to the persistence model (PM).