Machine Learning-Based Statistical Hypothesis Testing for Fault Detection

This paper elaborates the development of machine learning approach merged with statistical hypothesis testing aimed at enhancing the operation of photovoltaic (PV) systems by developing intelligent PV fault detection framework. Fault detection in PV systems is important to ensure optimal energy harvesting and reliable power production because PV systems usually operate in a harsh outdoor environment and tend to suffer various faults. In this paper, therefore, special attention is paid to detection of various faults during different modes of operation. The proposed approach merges the benefits of machine learning technique (MLT) with statistical hypothesis testing to enhance the fault detection and monitoring of PV systems. The proposed technique will be effective in monitoring the PV faults under both normal and abnormal conditions. For the framework developed, the modeling phase is addressed using MLT and the faults are detected using the generalized likelihood ratio test (GLRT) chart. The MLT is used to compute the residuals monitored and the GLRT chart is applied to the monitored residuals evaluated for fault detection purposes. The developed MLT-based GLRT algorithm is implemented and validated using both simulated and real PV data. The results are evaluated in terms of false alarm rates (FAR), missed detection rates (MDR) and computation time.