Very Short-Term Photovoltaic Power Generation Forecasting with Convolutional Neural Networks

Abstract

Photovoltaic (PV) power generation forecasting is an active research topic for the efficient operation of microgrid system. Although the estimation of the direction of change in hour-to-hour power generation is also important factor, there exist few studies for hour-to-hour PV generation forecasting tasks compared with longer-terms. In this paper, we compare the characteristics of hour-to-hour PV generation forecast tasks with longer-term tasks, and we also examine the limitations of applying the LSTM/RNN-based model to this task, which has been generally considered as powerful predictor for daily ones. To overcome these limitations, we propose a pre-predicted weather value-concatenated CNN-based approach.