Machine Learning for Photovoltaic Systems Condition Monitoring: A Review

Condition Monitoring of photovoltaic systems plays an important role in maintenance interventions due to its ability to solve problems of loss of energy production revenue. Nowadays, machine learning-based failure diagnosis is becoming increasingly growing as an alternative to various difficult physical-based interpretations and the main pile foundation for condition monitoring. As a result, several methods with different learning paradigms (e.g. deep learning, transfer learning, reinforcement learning, ensemble learning, etc.) have been used to address different condition monitoring issues. Therefore, the aim of this paper is at least, to shed light on the most relevant work that has been done so far in the field of photovoltaic systems machine learning-based condition monitoring.

I. Introduction

Renewable energies have important results for energy consumption in many sectors, including public and industry. Their climate-friendly energy transformation translates into the attraction of power generation companies from around the world [1]. Statistics from "World Energy Data" [2] explain that energy consumption with renewable energy resources reaches 23.6% of world energy consumption. Among the many renewable energy resources, solar energy consumption occupies 11.44% of the total amount of renewable energy expenditure with future potential increase. However, satisfying energy consumption needs and ensuring high quality distribution totally depends on a reliable condition monitoring system capable of real-time assessment of health status while providing necessary information on maintenance planning [3]. Accuracy of a condition monitoring system is itself based on a well-constructed virtual model capable of simulating the behavior of the actual studied system [4]. Most of the literature works indicate that a set of rational physical interpretations (thermal modeling, electrical modeling, etc.) will certainly lead to a very powerful model in case of a lower depth level of the treated problem [5]. The depth of the problem lies in the complexity of the system that may include multiple parameters such as number of components, nature of interactions, and external effects [5]. As the complexity reasonably increased, data-driven solution will be the only available modeling paradigm. Among data-driven methods, while taking into account new varieties of advanced sensor technologies, especially in the Industry 4.0 era, machine learning (ML) theories have become one of the main way of large problems treatment [1]. As a result, in the field of health condition monitoring of photovoltaic (PV) systems, several ML approaches have been studied in-depth. Several training paradigms ranging from hybrid to deep learning has been discussed.