How Machine Learning Improves PV Predictions
The ability to accurately, reliably predict generation at a solar site is understandably crucial to maximizing output and efficiency—and revenue, of course. Machine learning techniques help us fine-tune such predictions, utilizing not only site characteristics but also important weather and climate data to achieve nuanced results. As Trimark boasts an incredible amount of SCADA data in our systems, we are able to research and develop more accurate, advanced prediction models that will bolster our performance management capabilities significantly.
May Cai, a data scientist at Trimark, has been at the forefront of these efforts. May recently collaborated with Thushara Gunda from Sandia National Laboratories on a research paper that details the merits of applying machine learning in this capacity. Titled “Foresee the Future: Using Machine Learning, Climate, and Site Characteristics to Predict PV Solar Plant Generation,” the paper outlines the weaknesses of existing prediction models (such as PVSyst and IEC 61724), the importance of incorporating site-specific weather and climate conditions, and various data limitations that must be accounted for. Moreover, Trimark and Sandia’s joint effort shows us that collaboration within the industry is the path forward, as working together will help us learn a great deal about this fascinating technology.
To download and read the full white paper, submit a request here.