Performance monitoring of photovoltaic modules using machine learning based solutions: A survey of current trends

Performance monitoring is a major issue for solar energy systems in the development of a reliable, efficient, and durable solar energy facility. According to the response characteristics to the different environmental conditions of radiation, temperature, and humidity, optimum performance of PV modules can never be tracked and maintained without continuous evaluation of operational condition. Traditional performance forecasting methods, which are mostly statistical, such as multiple regression analysis, fail to provide solutions to many of the complexities and high dimensionality of data that modern PV systems generate. Such difficulties make it computationally expensive and sometimes impossible to derive an accurate predictive model using conventional techniques. Machine Learning (ML) is an emerging data-driven approach that has, so far, contributed effectively to overcoming some of the limitations of conventional methods. Enabling real-time monitoring, fault detection, and optimization of performance, ML accomplishes this through the use of more advanced algorithms that can detect patterns and relationships in unstructured data. This review opens the door for deeper broader coverage in future surveys concerning the application of ML trends in PV performance monitoring. The paper explores the full array of machine learning (ML) methods, supervised techniques for predictive analytics, including support vector machines (SVM) and random forests. Among the various unsupervised techniques, clustering helps detect anomalies, while the deep learning frameworks process large-scale and multi-modal data. The survey reveals areas where ML is being exploited in the most critical areas of PV system performance monitoring, such as energy yield prediction, degradation analysis, and fault detection. The paper also extends to ML involvement with IoT devices by providing automated data collection for enhanced decision-making. Several case studies demonstrate ML applications in different PV systems and results exceeding those observed with conventional means in terms of operational efficiency and cost savings in maintenance. Besides reviewing new advancements on ML-based solutions, this paper also looks at key challenges facing the field such as quality and availability of data, computation costs, and robust and interpretable ML models. The review also suggests strategies for addressing these challenges, such as transfer learning, hybrid models, and the production of open-access datasets. The ultimate aim of this work is to bring to the fore, current trends and the potential impact of this rapidly changing domain of artificial intelligence, in the field of (particularly low scale) PV systems monitoring, for the benefit of researchers and industry practitioners.