Explainable Lightweight Deep Learning for Solar Radiation Prediction Using Mobilenet-Bilstm with Multi-Head Attention

Abstract

As the global demand for sustainable energy grows, the importance of solar energy as a clean and inexhaustible source of energy has become increasingly prominent. However, the efficiency and yield of solar power generation are directly affected by the solar radiation intensity, so accurate prediction of solar radiation is crucial to improve the performance of solar power generation systems. This paper proposes a hybrid Mobilenet-BiLSTM deep learning model combined with multi-head attention mechanism for solar radiation prediction. The lightweight model MobileNet is used to extract image features, and the BiLSTM model is combined to process time series features. Moreover, a multi-head attention mechanism is introduced to enhance the model's ability to capture the correlation of input sequences at different positions, so as to improve the prediction accuracy. In the model performance evaluation, the hybrid model performs well on multiple key indicators, with Mean Absolute Error (MAE) of 0.2355, Mean Square Error (MSE) of 0.1159, and Root Mean Square Error (RMSE) of 0.3405. In addition, Explainable Artificial Intelligence (XAI) techniques, including Grad-CAM, SHAP, and LIME, are integrated to interpret the predictions of the model and improve the transparency and reliability of the model. The experimental results show that the hybrid model has achieved remarkable results in prediction accuracy, model interpretability and practical application, which provides a new solution for the efficient utilization of solar energy resources.