Smart Grid Anomaly Detection System

Abstract

This project presents an AI-powered smart grid anomaly detection system with a Flask backend, SQLite database, and a modern web interface. The system analyzes more than 50,000 data points collected at 15-minute intervals and detects normal operation, point anomalies, contextual anomalies, and collective anomalies (5+ consecutive anomalies). The machine learning stack combines Isolation Forest for unsupervised outlier detection, LSTM autoencoders for sequence reconstruction, and a Gradient Boosting classifier as an ensemble decision layer. The application exposes interactive parameter sliders, live computed engineered features, a results page with confidence/probabilities, a history view with exports, and an analytics dashboard. Results indicate high accuracy and low-latency inference suitable for real-world monitoring. Keywords: Smart Grid, Anomaly Detection, Isolation Forest, LSTM Autoencoder, Gradient Boosting, Flask, SQLite.