Intrahour cloud tracking based on optical flow

Abstract

Drifting of the clouds may shade the sun which causes significant power output fluctuations in solar power systems. Forecasting the location of clouds could provide sufficient time to smoothen the power output. A method for identification and tracking cloud pixels from a commercial All-Sky camera is presented. Sky images taken every 5s were processed to determine the optimal threshold of different criterion to identify the six classes of clouds as well as clear-sky pixels. Cloud boundaries are tracked using Lucas-Kanade optical flow estimation method. The time when clouds will shade the sun can be forecasted 30s ahead with an acceptable accuracy. The research results demonstrate that the method presented has the potential for tracking different type of clouds with high variability.