Weekly Power Generation Forecasting using Deep Learning Techniques: Case Study of a 1.5 MWp Floating PV Power Plant

Power generation forecasting is essential for power system planning. Deep learning (DL) is an artificial intelligence technology applied to power generation forecasting. This paper aims to compare the effectiveness of various deep learning techniques to forecast the weekly power generation of a 1.5 MWp floating PV power plant at Suranaree university of technology's hospital, Nakhon Ratchasima, Thailand. This study uses eleven deep learning techniques to examine forecasting effectiveness, including CNN, LSTM, BiLSTM, GRU, BiGRU, CNN-LSTM, CNN-BiLSTM, CNN-GRU, CNN-BiGRU, BiLSTM-GRU, and BiLSTM-BiGRU. The actual three months measurement data trained each deep learning technique and then was used to forecast a weekly power generation. Three error indexes, MAE, MAPE, and RMSE, were used to confirm the effectiveness of each deep learning technique. The forecasting results showed that LSTM, CNN-BiLSTM, and CNN-BiGRU techniques gave the slightest inaccuracies in the forecasting results of the interesting deep learning techniques.