Application of Stacked LSTM Networks in Forecasting Day Ahead PV Power Generation from Onsite Data

Photovoltaic (PV) power generation, though a clean source poses considerable difficulties for its stable incorporation into existing electrical systems due to its weather-dependent dynamics and intermittence. For the management of PV power generation in the on-grid system, it is crucial to conduct research that allows for exact and reliable forecasting of PV power. This study involves day-ahead forecasting of PV power and weather-related characteristics. This forecasting is achieved using three variants of stacked long short-term memory (LSTM) networks: LSTM-LSTM, bidirectional long short-term memory (BiLSTM)-LSTM, and BiLSTM-BiLSTM. Finally, a comparative analysis is conducted using an actual dataset derived from one year of onsite data of SNM Solar, Cuttack, Odisha, India. A range of performance metrics—mean square error (MSE), root mean square error (RMSE), and coefficient of determination (R2) were used to evaluate the effectiveness of this method. The results indicate that the suggested model BiLSTM-LSTM is a dependable and optimum method for forecasting solar photovoltaic PV power as per the mean and standard deviation of R2 for test data.