Detection of Anomalies in Power Profiles using Data Analytics

Abstract

Deployment of high reporting rate smart metering infrastructure together with a multitude of sensors for automation and control are an increasing trend among energy communities and prosumers. These systems provide useful information for data-driven prediction and classification models for micro-loads and local power generation. Matrix Profile is a promising general purpose data mining technique for time series data, such as electrical measurements from advanced smart meters. In this work, we first describe the measurement context that provides rich data availability for current advanced energy analytics applications. We target power profiles for both generation and load to highlight salient and complementary characteristics thereof, which can be leveraged in applications involving data-driven analytics for enhancing observability in distribution grids. A sensitivity analysis investigating the chosen method under various input noise assumptions is presented using Monte Carlo simulation. The comparative results indicate the relative robustness of the Matrix Profile approach for anomaly detection tasks in energy measurements traces.