Short-term solar irradiance forecasting model based on hyper-parameter tuned LSTM via chaotic particle swarm optimization algorithm

Accurate prediction of solar irradiance is important for the solar industry for efficient power generation and reliable integration into the micro-grids. With this incitement, this paper suggests a hybrid deep learning (DL) model via Long Short-Term Memory (LSTM) networks and Chaotic Particle Swarm Optimization (CPSO). The proposed DL architecture leverages the power of LSTM to learn complex temporal patterns in short-wave solar irradiance data. The main objective of the CPSO is to minimize the prediction error through optimizing the LSTM’s hyper-parameters such as neurons in hidden layers, learning rate, batch size, dropout rate and activation function. During evaluation of the proposed model, several performance matrices such as MAE, MAPE, RMSE and coefficient of determination are calculated and analyzed. The lowest error is observed for the 60-minute prediction in the Rainy session, with MAE = 0.04327, MSE = 0.00419, and RMSE = 0.07913, indicating the best forecasting performance among all cases. The outcome of this experimental study is ensured by comparing its results with standard LSTM architecture, other optimization algorithms such as FA and PSO, and conventional approaches. The output of the comparative study demonstrates that the proposed CPSO-LSTM model outperforms benchmark models, attaining a significant improvement in forecasting accuracy. Therefore, the proposed model can be used as a valuable tool for solar power plant operators, utilities, and policymakers to optimize solar energy production and ensure a reliable supply of renewable energy.