The increasing need for energy security and the development of clean energy sources to mitigate the impact of climate change necessitate the development of multisource energy generation systems comprising solar panels, generators, and grids. Four hybrid deep learning models were developed to predict the solar power generation (SPG) from historical solar panel data. These include a time convolutional network with enhanced multilayer perceptron (TCN-EMLP), a convolutional neural network with long short-term memory (CNN-LSTM), an LSTM with AutoEncoder (LSTM-AE), and a transformer model. These models were deployed for the predictions of solar power generation (SPG). All models were applied to a dataset containing 378 observations of solar energy data, allowing for a direct comparison between the hybrid deep learning methods employed. The input variables used include battery level (BL), ambient temperature (Temp), solar Irradiance (Ir). The results indicated that LSTM-AE showed superior performance relative to TCN-EMLP, LSTM-AE, and the transformer model with a strong R2 of 0.8359, a summary R2 of 0.8059, a root mean square error (RMSE) of 7.4304 W, and a mean absolute error (MAE) of 5.9322 W. CNN-LSTM achieved a significantly high performance comparable to TCN-EMLP and the transformer model. The utilization of deep learning to build intelligent automated multisource energy systems could lead to enhanced prediction accuracy, better performance, and higher sustainability by lessening reliance on non-renewable backup systems.