Currently the risk assessment methods of wind turbines are mostly based on a key component to analyze the overall operation of the wind turbine. Due to the strong coupling effect of the key components, the influence of each component to the wind turbine must be considered. In order to maintain and repair the wind turbine better, this paper takes the historical data of the wind turbine as the basis, carries on the risk assessment, through the construction of the centralized control center of the wind farm, and establishes the data model to collect the operating data of the wind turbine, and extract the key parts of the wind turbine. The characteristic quantities of the state are described by using the theory of generator set, the wind speed power curve and the actual output power vacancy to describe its running state. With the input of the characteristic quantity of the key components, the risk degree of the wind turbine is the output, the probabilistic neural network model is set up. The prediction classification effect of the model can be seen through the example simulation. This method can be used to evaluate the risk state of the wind turbine well and provide reference for the maintenance and maintenance of the operation and maintenance.