Distribution Network Power Quality Insights With Optimally Placed Micro-PMUs Incorporating Synthetic and Real Field Data

Widespread deployments of optimally placed real-time power quality (PQ) monitoring tools such as distribution level micro-phasor measurement units (D-PMUs or
$\mu $
PMU), digital fault recorders, and PQ analyzers are expected to play a critical role in improving the stability and reliability of the smart grid. In this paper, an improved PQ disturbance (PQD) classification method using discrete wavelet transform (DWT) with a cubic multi-class support vector machine (CMSVM) classifier is proposed, which incorporates a decade’s worth of high-quality continuous waveform PQ data from the Australian power network. This research also introduces misclassification cost (MC) and cost-sensitive classification theory into the area of PQD classifiers to build improved and more robust network models for the future. The method is evaluated using four case studies of synthetic and real-world PQD field data combinations and five application case studies using optimally placed
$\mu $
PMUs. The results indicate similar classification performance for standard PQDs than previous literature, alongside improved MC for complex PQD classes. Comparative analysis with previous literature highlights the importance of using high-quality real PQD field data to improve the fidelity of classifiers to provide better PQ insights as more complex components are added to the distribution network.