Enhancement and Validation of Cloud Detection Methods - Comparison between a Threshold Based and a Clear Sky Library Technique for Cloud Detection

Kurzfassung

Short-term forecasting (0-30 minutes) of solar irradiance is of great signi�cance for the optimal operation and power prediction of grid-connected photovoltaic (PV) and concentrating solar power (CSP) power plants. One method to provide such nowcasting, uses all sky cameras. The resulting images are of higher resolution than what can be obtained from satellites and the upwards pointing nature of the camera makes it easy to capture low-lying clouds. An accurate cloud detection will improve the accuracy of short-term forecasting of solar irradiance. Cloud detection algorithms based on ground based imagery have been developed and employed in the last years. The Clear Sky Library (CSL) method uses the red to blue ratio (RBR) from clear sky days throughout the year, as a reference to detect clouds in the image. However, it has shown, that changes in the atmospheric aerosol concentration, causes variations in the clear sky RBR. As a consequence, the �xed threshold techniques, as well as the CSL method are frequently unable to detect thin clouds in an atmosphere with a high aerosol content. In this thesis an enhanced CSL method is presented, taken account of the atmospheric condition by introducing a turbidity factor (Linke Turbidity (TL)). In literature there exists several cloud detection methods. To date, there is no satisfactory single parameter for the performance of cloud detection algorithms, which makes it difficult to compare di�erent methods. To be able to measure their performance and to see how di�erent methods compare, a validation method is developed. The validation method is applied to the enhanced CSL algorithm and the previous algorithm (based on a xed threshold technique). Comparing both validation results showed, that the CSL method outperforms the previous algorithm by yielding an improvement in the overall accuracy of at least 20 %. The best improvement could be achieved for overcast scenarios. Reasons for shortcomings in accuracy and the performance are discussed, and ideas to further improve the cloud detection algorithm, especially in problematic cases, are investigated.