The hour ahead forecasting of global horizontal irradiance (GHI) is essential for integrating the electrical energy produced from solar photovoltaic (SPV) plants into electrical grids. Solar energy is highly intermittent and diurnal in nature. The electrical energy produced from the SPV plants cannot be directly integrated into the electrical grid since it causes grid instability. In order to alleviate this issue, GHI forecasting is critical. In this study, advanced deep learning techniques such as Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), Gated Recurrent Unit (GRU), Bidirectional Long Short-Term Memory (BiLSTM), U-net, Extreme Gradient Boosting (XG Boost), Gaussian Processing Regression (GPR), hybrid models, and an ensemble model are used to forecast the hour ahead Global Horizontal Irradiance. The weather parameters for this study are obtained from Solcast®. The models used were hyper-tuned to determine the optimal configuration of the architecture. Root mean square error (RMSE), mean absolute error (MAE), coefficient of determination (R2), and forecast skill (FSRMSE) are used to assess the model’s performance. A smart persistence model is used as a baseline model for comparison. The proposed ensemble model yields the FSRMSE 27.80% for temperate climatic zone, 28.47% for composite climatic zone, 51.08% for hot and dry climatic zone, 20.16% for warm and humid climatic zone, 27.99% for cold climatic zone. It is concluded from the results that the proposed ensemble models comprised of CNN, LSTM, and XG Boost performed exceptionally for hot and dry climatic zones as compared to the other climatic zones.