Machine Learning Algorithms for Real-Time Fault Detection and Performance Enhancement in Solar Energy Systems

Abstract

This thesis presents machine learning algorithms and models for enabling real-time fault detection and performance enhancement of Solar Energy Systems. Modern Electroluminescence imaging technology and high-performance parallelizable ML models are used to scrutinize the condition of Electronic Focus Solar Energy Systems during power generation operations without shutdown. Utilizing the ML models trained on historical Solar Energy System generations and Electroluminescence information, we determine temporal performance degradation characteristics at otherwise latently generating Solar Energy Systems. Through temporal analysis of the actual degradation characteristics, (1) detection of a fault during generation operations, (2) prediction of cell failure time and degradation characteristic, and (3) estimation of the actual degradation characteristic, are enabled.

It is demonstrated that widening the training data window for ML model training enhances the temporal performance of the ML model. Furthermore, a higher risk of faults is identified for Solar Energy Systems located in dustier desert conditions, and these systems should be specially monitored or have preventive maintenance carried out. Experimental results indicate that the sizes and distributions of degradation zones may differ among some Solar Energy Systems, likely caused by effects like spatial temperature non-uniformity and undesired metal bridge pollution. For Solar Energy Systems attached with back contact solar cells, ML models indicate that there is a higher risk of cells with the back contacts being damaged and losing focus. To enhance the promulgation potential of the ML models proposed, we propose a new method of enfolding the temporal characteristics of degradation of Electronic Focus Solar Energy Systems in the period of risk detection based on temporal ML model enhancement. If any risk of fault or fault of an Electronic Focus Solar Energy System is detected, proper maintenance and preventive actions should be conducted to keep the systems from going through latently generating periods.