Anomaly detection in public street lighting data using unsupervised clustering

The digitization of the energy industry enables the collection and analysis of vast amounts of consumption data for improved decision-making and efficient energy resource utilization. This big data represents an opportunity to address data quality issues and identify anomalies. The detection of anomalies in data that significantly deviates from normal behavior is critical, as they can lead to economic losses. To extract valuable insights and mitigate such challenges, big data analytics platforms and machine learning techniques are increasingly being adopted. Their practical use is crucial in defining consumer-centric, sustainable, and resilient technologies for Industry 5.0. In this article, we address the problem of anomaly detection in electricity consumption for public street lighting. We present our approach and experience in analyzing anomalous behavior of public street lighting plants using electric energy data collected and provided by the ENEA PELL smart city platform. To this end, we adopt three well-known unsupervised clustering techniques: k-means, DBSCAN, and OPTICS. We conduct a comparative analysis of these methods using a real benchmark data set and additional data with known anomalies, used as ground truth data, to verify the correctness of the methods. We synthetically generate additional data from the real data set by automatically injecting anomalies based on specific patterns of anomalous behavior identified for the public street lighting domain. We examine the clustering methods’ performance by calculating their accuracy in detecting synthetic anomalies, as well as their computation cost. The experiments show that while K-means runs faster, DBSCAN is more accurate than K-means and OPTICS in detecting synthetic anomalies.