Multi-site solar power forecasting using gradient boosted regression trees

• Published in: Solar energy 2017-07, Vol.150, p.423-436
• Main Authors: Persson, Caroline, Bacher, Peder, Shiga, Takahiro, Madsen, Henrik
• Format: Article
• Language: English
• Subjects: Climatology; Electric power generation; Energy consumption; Forecasting; Forecasting techniques; Gradient boosting; Learning algorithms; Machine learning; Mathematical models; Multi-site forecasting; Performance prediction; Photovoltaic cells; Power; Regression analysis; Regression trees; Renewable energy sources; Solar cells; Solar energy; Solar power; Solar power forecasting; Spatio-temporal forecasting; Weather forecasting
• ISSN: 0038-092X; 1471-1257
• DOI: 10.1016/j.solener.2017.04.066

•Gradient boosted regression trees applied for prediction of solar power generation.•A non-parametric approach for multi-site prediction of solar power generation 1-6 hours ahead.•Gradient Boosted Regression Trees outperforms simpler autoregressive models.•Combining and ranking the importance of explanatory variables for various forecast horizons. The challenges to optimally utilize weather dependent renewable energy sources call for powerful tools for forecasting. This paper presents a non-parametric machine learning approach used for multi-site prediction of solar power generation on a forecast horizon of one to six hours. Historical power generation and relevant meteorological variables related to 42 individual PV rooftop installations are used to train a gradient boosted regression tree (GBRT) model. When compared to single-site linear autoregressive and variations of GBRT models the multi-site model shows competitive results in terms of root mean squared error on all forecast horizons. The predictive performance and the simplicity of the model setup make the boosted tree model a simple and attractive compliment to conventional forecasting techniques.