An End-to-End Explainable AI Framework with Automated LLM-Based Natural Language Explanation Generation for Energy Systems

Black box AI models need explanation layers for high-stakes decisions: The paper demonstrates that while deep neural networks perform efficiently, their decisions are difficult to interpret—making explainability essential in critical domains like energy systems where transparency and accountability matter most.

Multiple explanation methods provide different insights: The framework combines LIME (for local, prediction-specific explanations) and SHAP (for global, feature-importance explanations), showing that different XAI algorithms capture different aspects of model behavior and together provide more complete understanding.

Explanations must be evaluated for quality: The paper uses fidelity and stability metrics to verify that generated explanations actually reflect what the model is doing—a crucial step often overlooked, ensuring explanations are trustworthy rather than just plausible-sounding.

Natural language makes AI explainability accessible to non-experts: By converting technical XAI outputs into plain language through large language models, the framework enables non-technical stakeholders (like energy system operators or policymakers) to understand and trust AI decisions without requiring data science expertise.

Practical framework tested on real energy applications: The approach was validated on actual use cases—power system fault detection and building energy labeling—demonstrating that end-to-end explainable AI is feasible for real-world energy infrastructure decisions where failures could have significant consequences.