Protection Analysis of a Traveling-Wave, Machine-Learning Protection Scheme for Distributions Systems With Variable Penetration of Solar PV

This work provides a detailed protection analysis of a fast, Traveling-Wave (TW), Machine-Learning (ML), local, non-directional, economic, and setting-less protection scheme. The goal is to provide an objective evaluation of how a data-driven TW protection scheme would perform when deployed on a power distribution system as a faster alternative to over-current (OC) protection. Fault simulations consider scenarios from none to high-penetration of renewable energy resources to show its suitability for future applications on Distributed Energy Resources (DER)-dominated distribution systems. A modified IEEE 34 node system with solar Photovoltaic (PV) in multiple locations is modeled in PSCAD/EMTDC. The TW, ML method's protection scheme is built upon an efficient signal-processing stage, using the Discrete Wavelet Transform, and scaled-down Random Forest models that classify the fault location into several protection zones. The analysis is focused on the quantification of the sensitivity and selectivity of the proposed protection scheme. Furthermore, estimations of the false trip probability and average unnecessary load loss are included, as these events may occur due to the misoperation and miscoordination of the proposed data-driven, signal-based relays. Results show high TW detection rates and fault location accuracies, which cause a small and bounded percentage of imprecise fault locations and unnecessary load loss. Similarly, the same IEEE 34 node system is modeled in OpenDSS and a custom OC protection scheme is designed. The comparison between both methods leads to the conclusion that TW, ML methods can be a significantly faster aid to traditional protection.