Monitoring Grid Stability Using Time Series Analysis and Machine Learning-Based Anomaly Detection: A Comprehensive Framework for Real-Time Power System Reliability

Modern electrical grids are becoming increasingly complex due to the integration of renewable energy sources, smart grid technologies, and fluctuating demand patterns. This study presents a comprehensive framework for monitoring grid stability using time series analysis and machine learning-based anomaly detection. A multi-layered approach was developed, combining seasonal-trend decomposition using Loess (STL), ARIMA-based residual analysis, isolation forests, and statistical outlier detection implemented in R programming language. Real-world electricity consumption data from the University of California, Irvine (UCI) Machine Learning Repository, spanning four years (2006–2010), were analyzed to validate the framework. The hybrid anomaly detection system achieved a precision of 92.3% and a recall of 87.6% in identifying consumption anomalies. Statistical decomposition methods detected 156 meaningful anomalies, while machine learning techniques identified 134 anomalies with higher confidence. The framework demonstrates strong potential for real-time grid monitoring applications and, with further operational development, could contribute to reducing unplanned outages.