Prediction of solar energy generation is crucial in providing stable electrical supply and integrating renewable energy onto the electrical grid. To address challenges associated with complexity of the data, efficient modeling for the data and achieving interpretability of the predictions is essential. This paper presents a hybrid Deep Learning (DL) model for forecasting solar energy generation. The method initiates with a complete data preparation process for solar generation data by using the IQR method for robust outlier identification, followed by a thorough analysis of the distribution and variability in the data to determine trends in the data. The forecasting model for solar generation is developed using a new architecture that combines a Neuro Sparse Attention Integrated Liquid State Machine with Cross Latent Transformer Networks (NS-LSM-CLTNet), Hawkfish Optimization (HFO) algorithm is implemented to find the optimal configuration of the deep neural networks for ensuring the best possible performance from the trained models. The proposed NS-LSM-CLTNet model's forecast accuracies are validated in Python and achieves highest R² of 0.99 and lowest error rates such as MSE of 1306.94, RMSE of 36.151 and MAE of 21.602, which indicate that the proposed model performs significantly better than the existing approaches. Additionally, SHapley additive explanations (SHAP) framework is used to enable an explainable model. As a result, this provides certainty and transparency to grid operators and energy stakeholders about the forecasts produced by the system, thereby eliminating the black-box quality associated with many forecast solutions.