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

2024-12-01

Energy and Buildings

AI Summary

Existing explanation methods for time-series anomaly detection in energy systems—such as SHAP—suffer from high computational complexity and unstable outputs. To address this, we propose a context-aware relevance focusing mechanism that shifts interpretability from global features to semantically coherent local subsequences, enabling domain-knowledge-guided explainability. Our method integrates an LSTM/Transformer-based anomaly detector, gradient-driven temporal saliency analysis, and context-similarity-weighted masking. Evaluated on multiple real-world energy datasets, it achieves substantial improvements in explanation fidelity (+23.6%) and expert trust (+41%), without compromising detection performance. The core contribution is the first incorporation of context-aware relevance modeling into time-series anomaly explanation—uniquely balancing interpretability, computational efficiency, and domain adaptability.

Technology Category

Machine Learning: Transparent, Interpretable, Explainable MLHumans and AI: Explainable AI (XAI) for Human UnderstandingComputer Vision: Interpretability, Explainability, and Transparency

Application Category

User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalizationSearch and Retrieval-Augmented AI: Web query analysis, representation and understandingGraph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs

Problem

• Energy Efficiency
• Anomaly Detection
• Explainable AI

Innovation

• Deep Learning
• Enhanced SHAP Interpretation
• Stable Anomaly Detection