Study on Ultra-short Term Distributed Photovoltaic Power Forecasting in the Haze Day Based on All-sky Images Combined with Support Vector Regression

Abstract

Due to the influence of many local random mutation factors, the ultra-short-term prediction of distributed photovoltaic power is faced with great challenges. This study is a modeling study for accurate prediction of photovoltaic power in hazy days based on all-sky image observation data. Firstly, based on the all-sky images, digital image processing technology is used to extract radiation-related image features of haze days, including image brightness, power spectrum, smoothness and mean of singular value. Secondly, the support vector regression model is established by combining image features and radiation attenuation coefficient. Finally, via determining the suitable forecasting time horizon, the surface irradiance can be forecasted based on the support vector regression model, and the photovoltaic power ultra-short-term prediction is realized using the irradiance-power convert model. The experimental results show that the ultra-short-term prediction method of photovoltaic power based on the whole sky image features and support vector regression has a good prediction effect when the optimal prediction time is determined, which provides an important practical basis for the accurate prediction of ultrashort-term power of distributed photovoltaic power stations.