Machine learning applications for urban photovoltaic potential estimation: A survey

Abstract

A growing worldwide consensus agrees that a global energy transition to renewable energy sources is urgent to avoid the direst consequences of rapid climate change. This transition is a substantial challenge facing humanity, which requires cooperation and innovation across disciplines and nations. In this context, the precise estimation of the renewable potential of a given area is valuable for decision-makers. Cities will play an essential role in this transition through distributed photovoltaic generation as evidenced by the UN 11th sustainable development goal. However, the complex nature of cities makes this estimation a difficult problem. Recently, several machine learning approaches have successfully contributed to different aspects of the urban photovoltaic potential estimation problem. In the present manuscript, these proposals are summarized, following a hierarchical framework usually described in the literature, including the latest available research. Input and target variables involved in the discussed works are reclassified using a novel categorization. This categorization highlights interesting trends in the field, for each sub-problem in the hierarchical approach, which allows the identification of knowledge gaps and possible future lines of research. The present work presents other unexplored avenues and lists concisely the techniques and variables used for each estimation problem, facilitating improvements on already explored techniques or innovation on not yet explored avenues.