Fig 1.
The architecture of the proposed SolarTrans for solar power prediction LLM-based generated explanations.
Table 1.
Layer-wise architecture detail of the proposed SolarTrans model.
Table 2.
Transformer encoder-decoder block details in the SolarTrans.
Table 3.
Summary of solar power generation dataset.
Table 4.
Hyperparameters of the proposed model.
Fig 2.
Training and validation loss curves over 60 epochs on combined datasets.
Fig 3.
Training and validation loss curves using Plant 1 dataset.
Fig 4.
Training and validation loss curves using Plant 2 dataset.
Fig 5.
Error distribution of SolarTrans model across Plant 1, Plant 2, and Combined datasets.
Table 5.
Performance of SolarTrans.
Fig 6.
Prediction and error analysis of SolarTrans on the Plant 1 dataset at forecasting steps (a) 1, (b) 4, and (c) 8.
Fig 7.
Prediction and error analysis of SolarTrans on the Plant 2 dataset at forecasting step 1.
Fig 8.
Prediction and error analysis of SolarTrans on the Combined dataset at forecasting step 1.
Fig 9.
Irradiation and Ambient Temperature Distribution at Plant 1 and Plant 2.
Fig 10.
Prediction vs. ground truth DC power (8 steps ahead) on combined datasets.
Fig 11.
Test sample predictions and ground truth DC power across two solar plants.
Table 6.
Performance comparison of SolarTrans model on combined dataset.
Table 7.
Evaluation metrics for FLAN-T5-Solar.
Fig 12.
Samples of generated AI explanation for the predicted DC power.
Fig 13.
Screenshot of the interface for prediction and AI-generated explanation.
Fig 14.
SHAP summary plots.
Fig 15.
Temporal distribution of features using SHAP heatmaps.