High-resolution data-driven anomalous event detection from solar farm data using clustering large applications based upon randomized search

Abstract

Solar energy is a significant and fast-growing source of low-carbon electricity. The usual means of utilityscale solar farm condition monitoring is limited by poor measurement accuracy and low-resolution data collection rates. A micro-synchrophasor measurement unit (µPMU) has been adapted and integrated with a power quality monitor (PQM). This apparatus provides the high-resolution, high-precision, time-stamped data needed by analysts to make solar farms more costeffective and to better understand decentralize grid behaviour. The resulting big data necessitates applying machine learning (ML) for automatic event forecasting, fault detection, and site maintenance. The limited availability of data knowledge, data volume, and performance issues drives the exploration of data-driven based unsupervised ML methods on this occasion. Clustering Large Applications based upon RANdomized Search (CLARANS) algorithm has been employed owing to its suitability to categorise events from the big data. CLARANS has been performed to recognize inefficient voltage phase unbalance. The voltage and current waveforms and related issues such as, voltage dip or voltage sag and phase imbalance events have been considered among multiple data streams and various power distribution issues for this investigation. Ten consecutive days of empirical data have enabled this research. Altogether, 250.92 million power quality data points have been tested and validated.