This paper proposes a Multi-task Logistic Low-Ranked Dirty Model (MT-LLRDM) for fault detection in power distribution networks by using the distribution Phasor Measurement Unit (PMU) data. The MT-LLRDM improves the fault detection accuracy by utilizing the similarities in the fault data streams among multiple locations across a power distribution network. The captured similarities supplement the information to the task of fault detection at a location of interest, creating a multi-task learning framework and thereby improving the learning accuracy. The algorithm is validated with real-time PMU streams from a hardware-in-the-loop testbed that emulates real field communication and monitoring conditions in distribution networks. The results showed that the MT-LLRDM outperforms other state-of-the-art classification methods using actual synchrophasor data achieved from a power hardware-in-the-loop testbed.

This paper presents a shape preserving incremental learning algorithm that employs a novel shape-based metric called the Fisher-Rao Amplitude-Phase Distance (FRAPD) metric. The combined amplitude and phase distance metric is achieved on a function space from the Fisher-Rao elastic registration. We utilize an exhaustive search method for selecting the optimal parameter that captures the amplitude and phase distance contribution in FRAPD when performing a clustering process. The proposed incremental learning structure based on the shape preserving FRAPD distance metric utilizes continuously updated fault shape templates with the Karcher mean. The seamless updating of abnormal events enhances the clustering performance for power systems fault detection. The algorithm is validated using the actual data from real-time hardware-in-the-loop testbed.

With the advent of Distribution Phasor Measurement Units (D-PMUs) and Micro-Synchrophasors (Micro-PMUs), the situational awareness in power distribution systems is going to the next level using time-synchronization. However, designing, analyzing, and testing of such accurate measurement devices are still challenging. Due to the lack of available knowledge and sufficient history for synchrophasors’ applications at the power distribution level, the realistic simulation, and validation environments are essential for D-PMU development and deployment. This paper presents a vendor agnostic PMU real-time simulation and hardware-in-the-Loop (PMU-RTS-HIL) testbed, which helps in multiple PMUs validation and studies. The network of real and virtual PMUs was built in a full time-synchronized environment for PMU applications’ validation. The proposed testbed also includes an emulated communication network (CNS) layer to replicate bandwidth, packet loss and collisions conditions inherent to the PMUs data streams’ issues. Experimental results demonstrate the flexibility and scalability of the developed PMU-RTS-HIL testbed by producing large amounts of measurements under typical normal and abnormal distribution grid operation conditions.

This paper presents a novel method for online power quality data analysis in transmission networks using a machine learning-based classifier. The proposed classifier has a bundle structure based on the enhanced version of the Extreme Learning Machine (ELM). Due to its fast response and easy-to-build architecture, the ELM is an appropriate machine learning model for power quality analysis. The sparse Bayesian ELM and weighted ELM have been embedded into the proposed bundle learning machine. The case study includes real field signals obtained from the Turkish electricity transmission system. Most actual events like voltage sag, voltage swell, interruption, and harmonics have been detected using the proposed algorithm. For validation purposes, the ELM algorithm is compared with state-of-the-art methods such as artificial neural network and least squares support vector machine.