Edge-Based AI Models for Detecting Nonlinear Energy Consumption Patterns in Smart Grid Environment Networks

This paper presents an innovative Edge AI framework to detect nonlinear energy usage patterns in real-time with the application to Smart Grid infrastructures. Designed for use on Edge devices where there is limited processing power but required to be highly analytic error-free, the proposed framework is lightweight. The authors developed GBOCLE - Energy, an efficient Anomaly Detector Model based on advanced Gradient Boosting methods specifically for low latency and real-time energy consumption data analysis. This model uses compressed forms of Light Gradient Boosting and One-Class SVM Algorithms to discover temporal, contextual, and relational anomalies across multiple Nodes on the Smart Grid network. Additionally, three techniques, Simplified Terrestrial Analysis, Adaptive Isolation-based Scoring, and Lightweight Graph-Based Neighbour Mechanisms, were used to further enable the detection of nonlinear relationships and interactions among the different components within the grid. The model uses Compact Nonlinear Indicators (e.g., Consumption Dev Index, Device Influence Vector and Reduced-Order Chaotic Metrics) as Analytical Features and Hidden Indicators for Anomalous Energy Usage. The experimental findings indicate that the suggested ensemble model has excellent detection accuracy (0.984), precision (0.974), sensitivity (0.975), specificity (0.965), F1-score (0.975), and AUC (0.984) while minimizing the computational and memory requirements for edge deployment. These findings support the conclusion that optimized edge-based detection is an efficient and feasible solution for detecting energy consumption anomalies and therefore offers the potential to enable predictive management and enhance operational performance in the Smart Grid environment.