Abstract: Being aimed at environmental meteorological monitoring and the history data of photovoltaic plant,we put forward an on-line fault diagnosis of photovoltaic components based on cascaded random forest. The characteristic description insists of three aspect:characteristic variable analysis,real data set preprocessing,model training and application. Experiment results show the effectiveness and accuracy of our method. This sample is of good reference value for on-line fault diagnosis of intelligent photovoltaic power station.