Data-Driven Approach Utilising Random Forest Regression for PV Performance Monitoring
Abstract
The performance monitoring of PV plants is essential to ensure their correct operation, detect faults, and maximize their output. To be able to decide whether a plant is operating within normal parameters, a reference is needed, which indicates the PV performance that the plant should have at certain weather conditions. Weather data and historical monitoring data from times, where the PV plant was operating correctly, can be used to train machine learning models, which can provide the reference PV performance. In this research a random forest regression machine learning algorithm is used to train models which predict the electrical power that is measured by 19 PV inverters. Several random forest models utilizing different parts of the available weather data were trained. Their prediction performance was evaluated for five different time resolutions and three different PV module orientations. The results indicate that random forest regression is a suitable tool to predict the performance of PV plants.