Solar Radiation Prediction Using Machine Learning Techniques: A Review

Solar radiation estimation determines how much energy the sun provides to a particular region. This radiation is the primary energy source of conversion in photovoltaic plants and solar thermal power plants. The incident radiation is not constant and depends on climatic data, which results in an intermittency in its behavior and changes in the production of electrical energy are observed. This justifies the development of a tool for predicting and estimating incident radiation in order to foresee changes in the performance of photovoltaic generation systems. This paper presents an analysis and review of the literature published in the Science Direct and IEEE databases since 1990, from the point of view of techniques application for the estimation of the primary solar resource. These techniques are classified according to the nature of the model from numerical and analytical approaches to Machine Learning. These approaches use different databases, inputs and mathematical relationships to establish dependencies among solar radiation, longitude, latitude and climatic parameters. In this paper, the selection criteria and behavior of the models are identified from the linearity treatment of the database to be used, the number of entries, the deviation between the value that is predicted and the test portion of the set of data that evaluate their behavior and provide decision tools for their use. Many authors apply Machine Learning to estimation both with unitary predictors and with hybrid models that profit from their potential.