Intrusion Detection System for Smart Grids: Machine Learning-based Anomaly Detection in Energy Consumption Data

This repository implements an Intrusion Detection System (IDS) for smart grid security using machine learning. The system establishes a behavioral baseline of normal energy consumption patterns to detect anomalies that could indicate cyber attacks such as False Data Injection (FDI) or meter tampering. KEY FEATURES:• Behavioral Baseline Establishment: Random Forest model with 99.3% accuracy (R²=0.993)• Security Monitoring: Statistical thresholding for anomaly detection• Temporal Feature Engineering: Captures daily, weekly, seasonal patterns• Theoretical Attack Detection Analysis: Statistical probability calculations SECURITY APPLICATIONS:1. False Data Injection (FDI) Attack Detection2. Meter Tampering Identification3. Anomaly Detection in Cyber-Physical Systems4. Behavioral-based Intrusion Detection DATASET:The American Electric Power (AEP) hourly energy consumption dataset (2004-2018) is used. The dataset is publicly available on Kaggle: https://www.kaggle.com/datasets/robikscube/hourly-energy-consumption METHODOLOGY:1. Data preprocessing and temporal feature engineering2. Random Forest regression for normal pattern learning3. Statistical security threshold establishment (τ = 3σ)4. Theoretical detection probability analysis RESULTS:• Baseline Accuracy: R² = 0.993• Prediction Uncertainty: σ = 204.7 MW• Security Threshold: τ = 614.1 MW• Detection Probabilities: 100% for scale attacks >10%, 28.9% for 500MW offset attacks This work demonstrates how high-accuracy forecasting models can serve as foundational components for smart grid security systems.