Multivariate statistical monitoring of photovoltaic plant operation

Detecting anomalies in a photovoltaic system play a core role in keeping the desired performance and meeting requirements and specification. For this propose, a simple and efficient monitoring methodology using principal component analysis model and multivariate monitoring schemes is designed to monitor PV systems. The principal component analysis model is used to generate residuals for anomaly detection. Then, the residuals are examined by computing the monitoring schemes (T2 and square predicted error) for the purpose of fault detection. However, these conventional schemes are usually derived under the hypothesis of Gaussian distribution. Thus, the major aim of this paper is to bridge this gap by designing assumption-free principal component analysis-based schemes. Specifically, a nonparametric approach using kernel density estimation is proposed to set thresholds for decision statistics and compared with the parametric counterparts. Real measurements from an actual 9.54 kWp grid-connected PV system are used to illustrate the performance of the studied methods. To evaluate the fault detection capabilities of the proposed approach, six case studies are investigated, one concerning a string fault, one involving a partial shading, and one concerning the loss of energy caused by inverter disconnections. Results testify the efficient performance of the proposed method in monitoring a PV system and its greater flexibility when using nonparametric detection thresholds.