Explainability of a Deep Neural Network Model for Prediction of Solar Panels Generation: Comparative Study

Abstract

The explainability methods LIME, SHAP, Integrated Gradients and Sailency Maps were compared to explain the predictions of a deep learning model trained to predict the electricity generation of a photovoltaic park.

These methods allow analyzing the relative importance of the input features for each prediction.

The quality of the explanations generated was evaluated using the fidelity and continuity metrics, before and after applying perturbations to the data.