Explaining Deep Learning-based Anomaly Detection in Energy Consumption Data by Focusing on Contextually Relevant Data

Black-box AI models used for energy management lack transparency: Deep learning approaches are effective at detecting anomalies in energy consumption data (which helps identify waste and equipment failures), but they don't explain why they flag something as anomalous, making it difficult for energy managers to trust or act on their findings.

Existing explanation methods have practical limitations: Current techniques like SHAP can be computationally expensive or produce unstable results that vary unpredictably, which undermines their usefulness for real-world energy management applications where consistency matters.

Context-focused explanations improve reliability: By tailoring explanations to focus on contextually relevant information rather than treating all data equally, this approach reduces explanation variability by approximately 38% on average, making AI decisions more trustworthy and consistent across different scenarios.

Robustness was rigorously tested: The findings were validated across 10 different machine learning models, 5 datasets, and 5 explanation techniques with statistical analysis, suggesting the approach is genuinely more stable rather than working by chance in limited cases.

Better explanations enable responsible energy management: When energy managers can understand why an AI system flagged an anomaly, they can make more informed decisions about equipment maintenance and energy efficiency investments, reducing the risk of acting on unreliable AI recommendations.