Lightweight Real-Time Energy Anomaly Detection and Causal Analysis Using Multi-Tier Edge Computing

This study uses a multi-tier edge computing architecture to present a lightweight, real-time anomaly detection and Rule-Based Causal reasoning framework for energy systems. The system combines ultra-low-power microcontrollers (ESP32/STM32) at Tier 1 for sensing and on-device TinyML-based anomaly inference, a local RISC-V-based embedded host (Lichee RV Dock) at Tier 2 for Rule-Based Causal analysis and dashboard visualization, and optional cloud platforms (Firebase, ThingsBoard, AWS IoT) at Tier 3 for extended services. Real-time voltage, current, and temperature data are collected, processed locally, and interpreted using rule-based Causal logic. Experimental results demonstrate the system’s low-latency performance (~0.2s), high anomaly detection precision (95.2%), and effective interpretability in edge deployments. The proposed TinyML-based model achieved a validation accuracy of 94.8% and a precision of 95.2%, demonstrating robust performance in anomaly classification under real-time edge-deployment conditions.