Data-Driven Estimation of Soiling Loss and Optimal Cleaning Schedule for a Utility-Scale PV Plant

Abstract

Soiling of panels in solar power plants can reduce production levels. In this thesis, we estimate the effect of soiling on power production and efficiency, as well as the gains from cleaning. Power data from a plant in southwest India was recorded every 5 minutes spanning 6 months. We analyzed this data to estimate efficiency degradation rates resulting from accumulation of soil and dust. The major challenge was filtering dataset noise/anomalies due to variations in micro-weather conditions. The key contribution of the thesis is a data-driven cleaning schedule algorithm. The algorithm detects cleaning events and produces a segmentation of the timeline into cleaning and soiling intervals. From the cleaning intervals we estimate the gains from panel cleaning, and from the soiling intervals we calculate the rate of power/efficiency loss. We apply these results to solve optimization problems regarding the cleaning schedule of a solar power plant. For example, by comparing the cost of cleaning against the potential gains in power production, we answer the questions “Which panel should I clean first/on this day?” and “Which day should I clean all panels?”. We hope that the contributions of this research will provide important insights for any party working with solar power data.