Deep Learning-Based Short-Term Power Output Prediction using Hybrid CNN-LSTM Model for Calatagan Solar Farm, Philippines

Le résumé fourni par la source

Solar energy is heavily reliant on solar radiation, which complicates the optimization of the system resulting in power output variability. To address this issue, forecasting techniques were implemented to predict power output for efficient energy planning. This paper presents a hybrid CNN-LSTM forecasting model using an estimated year-long worth of hourly solar PV power output of Calatagan Solar Farm, generated from the National Renewable Energy Laboratory (NREL) PVWatts® Calculator. Dataset was split into 75-15-10 for training, validation, and testing, respectively. In the training and testing phase, the trends for both the forecasted values and the NREL data during 3-day and 7-day forecasts were compared. The model’s performance was then evaluated utilizing RSME and MAE in the validation phase. Results showed that the 3-day forecast demonstrated a tolerable deviation with 0.2684 MAE and 0.3707 RMSE, which was marginally better than the 7-day forecast having the values of 0.2956 and 0.3964, respectively. The model demonstrated the capability of forecasting the solar PV power output that can help market participants make informed decisions regarding energy trading and market operations. For future research, atmospheric conditions should be included as one of the features to minimize the discrepancies between the values.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.