Graph-based Data Mining, Pattern Recognition and Anomaly Detection for Intelligent Energy Networks

Intelligent Energy Networks play an important role in the contemporary landscape of energy production and distribution, and the large amount of data produced by their smart sensors recently caused a great deal of interest in Data Analysis and Data Mining strategies to support network operators in optimizing operations, reliability and profitability. This work proposes for the network operator an unified data analytics framework, composed by clustering, anomaly detection and knowledge propagation algorithms, and based on a graph representation of the measurements dataset. The usage of the graph is particularly suited with multivariate and high-dimensional time series characterizing large and complex energy networks, it allows to express additional attributes on the data points and hence incorporate previous expert’s knowledge, and the leveraging of the graph structure and connectivity naturally captures dependencies among data for discovering energy patterns and anomalies in the original energy network. The effectiveness of the presented graph-based framework is validated on a number of real-world cases and private operators of District Networks. •Unified data analytics framework to capture patterns in an energy network.•Graph-based representation of the multi-variate and high-dimensional time-series data.•Clustering, anomaly detection and expert’s knowledge propagation algorithms.•Extensive numerical evaluation on a number of real-world complex district networks.