Characterization of Matrix Profile Technique for Enhanced Detection of Events in Sampled Values Data Streams

In the architecture described by IEC 61850 standards, merging units (MUs) and stand-alone MUs (SAMUs) are the initial elements in the measurement chain, with the main function of acquiring analog voltage and current signals, digitizing the samples and transmitting them to intelligent electronic devices (IEDs) for the evaluation of metrics of interest. Since they produce digital data, it is therefore possible to design extensions of their functionality that include data processing and the ability to identify events in the power grids. In this context, this article characterizes an event identification function to be implemented in MUs, based on the matrix profile (MP) technique and considering the Python library Stumpy. A comprehensive characterization of the MP technique is presented to evaluate its effectiveness for identifying discords in signals such as rapid events or voltage dips. The behavior of the MP technique is analyzed under different signal conditions and configurations, including variations in the number of samples per cycle and different window lengths. Noisy signals and challenging conditions such as overlapping events are considered. The MP performance is compared with other techniques used for discord detection in time series. The results obtained prove its robustness to noise and confirm its ease of use, needing only one parameter to operate.