A novel evolutionary ensemble model to forecast hourly global horizontal irradiance under various climatic zones

The diurnal and volatile nature of global horizontal irradiance has a significant effect on the power production from solar photovoltaic plants. Hence, it leads to gird instability. In order to alleviate the grid instability by managing the demand and supply efficiently, the accurate forecasting of global horizontal irradiance is essential. Advanced deep learning and ensemble techniques are not widely applied and tested in solar radiation forecasting. In order to fill the research gaps, this study proposes hybrid models like Bidirectional Long Short-Term Memory (BiLSTM) with CNN and U-net architecture. The novel ensemble algorithm consists of Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), and Extreme Gradient Boosting (XG Boost). The other models such as smart persistence, XG Boost, Gaussian Process Regression (GPR), LSTM, CNN, Gated Recurrent Unit (GRU), BiLSTM, CNN-LSTM are also used to forecast the hour-ahead GHI. The ensemble model has the lowest Root Mean Square Error (RMSE) of 35.02 W/m2, 47.41 W/m2, 56.59 W/m2, Mean Absolute Error (MAE) of 18.84 W/m2, 25.46 W/m2, 36.10 W/m2, Coefficient of Determination (R2) of 0.9745, 0.9745, 0.9625, and Forecast Skill (FS) of 0.5342, 0.4525, and 0.1961 for the hot and dry climatic zone, composite climatic zone, and warm and humid climatic zone, respectively.