The accuracy of photovoltaic(PV)power generation prediction in the microgrid has high relativity with the weather condition. Under cloudy and rainy conditions

random fluctuations of meteorological factors result in low precision of the ultra-short-term power prediction. For this reason

a modified model based on combination of Adaboost clustering and Markov chain is proposed. First

an improved K-nearest neighbor(KNN)classifier is trained with the characteristic variables extracted from solar radiation using the moving average method. To improve the prediction accuracy of cloudy and rainy days

the attenuation coefficient of solar radiation is introduced to modify the Hottel model. A weighted Markov chain model is developed to predict the microgrid PV generation subsequently. The simulation results indicate that the proposed model can appreciably improve the precision of power prediction under different weather conditions and is of great significance to real-time economical dispatch.