Deep hybrid CNN–biLSTM model for accurate solar photovoltaic power forecasting: A comparative study with classical and neural models

Accurate short-term forecasting of solar photovoltaic (PV) power is essential for grid stability and renewable energy integration, but remains challenging due to the inherent variability and intermittency of solar generation. This paper introduces a hybrid model that combines a Convolutional Neural Network (CNN) with a Bidirectional Long Short-Term Memory (biLSTM) network to address this challenge. The proposed CNN-biLSTM model is evaluated against four benchmark models, including Multilayer Perceptron (MLP), Support Vector Regression (SVR), Random Forest (RF), and a unidirectional CNN-LSTM, using historical meteorological and PV power data. Performance is assessed through a comprehensive suite of statistical metrics (R², RMSE, MAE, MAPE, sMAPE, and normalised RMSE). The results demonstrate that the CNN-biLSTM achieves superior accuracy, with the highest coefficient of determination (R2=0.99848) and the lowest error metrics (RMSE=0.5939 W, MAE=0.398 W, and nRMSErange=1.18 %), significantly outperforming all benchmarks. The bidirectional architecture uniquely captures temporal dependencies in both forward and backward directions, enabling more effective modeling of nonlinear solar fluctuations. This work establishes the CNN-biLSTM as a robust and reliable solution for real-world solar energy management systems, enhancing forecasting precision and supporting the stable integration of renewable energy into smart grids.