Abstract
Machine learning algorithms offer the promising opportunity to monitor a photovoltaic system’s performance, particularly its power output, using alternative data sources. In this work we consider performance data collected from nearby sites, rather than environmental data acquired with costly sensors or purchased from external sources. This study compares five different machine learning algorithms and their respective prediction accuracy of the output power using as input either power data from nearby systems or environmental data collected from an onsite weather station or one nearby. We show that power data from nearby sites can be used for predictions that are as precise as those based on onsite measured environmental data.