A Sky Image Analysis System for Sub-minute PV Prediction

Advanced energy management systems are increasingly gaining importance. These systems will allow the further introduction of Photovoltaic (PV) power generation, but for them to be really effective, sub-minute (1–60 sec.) PV generation prediction is required. In this context we propose a sub-minute PV prediction system based on the analysis of sky images. This is done by analyzing cloud movement, and thus the system does not rely on i) historical PV data, ii) dynamic model of the local weather, nor iii) location dependent information. The proposed system works as follows: from multiple image exposures, high dynamic range images are obtained (one per second), cloud movement is estimated, sky images are predicted, and finally PV generation is estimated using the predicted sky images. The proposed system achieves low error under various weather conditions.