Review of forecasting methods to support photovoltaic predictive maintenance

Predictive maintenance models are thought to be a reliable alternative to costly on-site maintenance techniques in the solar photovoltaic industry. They provide the owners with a third-party system to objectively diagnose and prevent failures in the system or any of its components. These models depend on data acquisition and analysis to compare the estimated performance with the actual energy production of the system, to conclude on the systems' health status. Ambient temperature, cell temperature, and solar irradiance forecasting are crucial parameters in predictive maintenance models' formulation. These parameters represent the input variables to the model and define its efficiency to a great extent. Nonetheless, the current approaches to predict weather-related parameters have a high dependency on data availability and, in some cases, sensibility to the location where they are being measured. Forecasting climate parameters using neighboring weather stations seems to be a promising option to tackle the cost of on-site sensing and data availability, however, the scarcity of applications of specialized forecasting methods to predict solar irradiance and temperature (cell and ambient) for such applications represents an obstacle. This document will present a review of the state-of-the-art comparing the techniques to forecast solar irradiance, and ambient and cell temperature, and their relationship with predictive maintenance models, by acquiring data from weather stations. This review evaluates the suitability of the models for solar photovoltaic applications, taking on their forecasting horizon, computational cost, and improvement opportunities to propose a framework for future work in the photovoltaic industry.