Abstract: Accurate assessment of wind turbine operation health is very beneficial to reduce the failure rate and operation and maintenance cost. Output power is one of the most basic parameters to characterize the operation performance of wind turbines. The fluctuation of output power can directly reflect the change of the operation state of wind turbines. When the actual output power obviously deviates from the expected value under the normal operation state, it indicates that the health state of wind turbines may be abnormal. Therefore, a risk assessment method is proposed for wind turbine operation based on output power prediction. Firstly, a short-term prediction model of wind turbine active power is constructed by using random forest algorithm. In the prediction process, a variety of meteorological factors are considered to improve the prediction accuracy, and then the risk state severity of wind turbine is quantified by the active power prediction error. Next, the fuzzy c-means algorithm is used to build the outlier model of wind turbine operation risk severity, which realizes the clear division of wind turbine operation risk level. Taking the measured data of a wind farm in Nantong, Jiangsu Province as the sample, the rationality and effectiveness of the proposed method are verified.