Nonparametric Event Detection in Multiple Time Series for Power Distribution Networks

Abstract

With the unprecedented advancement of sensing technology, smart city applications are now enriched with massive measurement data related to system states, patterns, and the behavior of its users. However, classic data analysis or machine learning tools ignore some unique characteristics of the multistream measurement data, in particular, the coexistence of strong temporal correlation and interstream relatedness. To this end, in this paper we discuss the problem of novelty detection with multiple coevolving time series data. To capture both the temporal dependence and the interseries relatedness, a multitask nonparametric model is proposed, which can be extended to family of data distributions by adopting the notion of Bregman divergence. Albeit convex, the learning problem can be hard as the time series accumulate. In this regard, an efficient randomized block coordinate descent algorithm is proposed. The model and the algorithm is tested with a real-world application, involving novelty detection and event analysis in smart city power distribution networks with high-resolution multistream measurements. It is shown that the incorporation of interseries relatedness enables the detection of system-level events, which would otherwise be unobservable with traditional methods. The experimental results not only justify the benefits of incorporating information from different sources, but also demonstrate the potential of the proposed multistream analysis tool as one of the core computational components to improve smart city observability, security, and reliability.