Sensor fault detection in photovoltaic systems using ensemble learning-based statistical monitoring chart

Photovoltaic systems have become increasingly popular as a source of renewable energy due to their environmental benefits and cost-effectiveness. However, sensor faults can significantly impact the performance of photovoltaic systems, resulting in reduced energy output and increased maintenance costs. This paper presents an effective approach for detecting sensor faults in photovoltaic (PV) systems using ensemble learning and the exponentially weighted moving average (EWMA) chart with nonparametric threshold estimation. The proposed approach trains the ensemble models using data collected during normal operating conditions of the PV system and detects any sensor faults by analyzing the residuals generated from the ensemble models. The EWMA chart is then applied to track changes in the residuals over time and detect any abnormalities. The flexibility of the chart is enhanced by computing the detection threshold using kernel density estimation (KDE). This approach improves the accuracy and reliability of the fault detection process. The proposed approach is assessed based on simulated data from a PV system using PVGIS. The results of the study demonstrate that the proposed method effectively detects sensor faults in photovoltaic systems, and the baggeed trees-based EWMA scheme outperforms the Boosted trees-based scheme in detecting faults in the pyranometer.