A Hybrid CNN–BiLSTM deep learning framework for accurate and robust solar power generation forecasting

Accurate solar power forecasting plays a vital role in maintaining the reliability and stability of modern power grids with increasing penetration of photovoltaic (PV) systems. The stochastic and nonlinear nature of solar energy generation, influenced by rapidly changing meteorological conditions, poses significant challenges to conventional forecasting techniques. To address this issue, this study proposes a hybrid Convolutional Neural Network–Bidirectional Long Short-Term Memory (CNN–BiLSTM) deep learning framework for short-term solar power generation forecasting. The proposed model combines the convolutional layers’ ability to extract spatial and local temporal features with the BiLSTM layers’ capacity to capture bidirectional long-term dependencies in time-series data. The CNN–BiLSTM model was trained and evaluated using real PV plant data, including environmental and meteorological parameters such as solar irradiance, temperature, humidity, and wind speed. Model performance was assessed through multiple evaluation metrics, including Root Mean Square Error (RMSE) and Nash–Sutcliffe Efficiency (NSE). Experimental results demonstrate that the proposed CNN–BiLSTM architecture achieves superior accuracy and robustness compared to conventional Deep LSTM (DLSTM) models.