Abstract: Under the coupling effect of offshore wind, wave, mechanical load and other factors, the state data of wind turbines fluctuates rapidly. The sensitivity of wind turbine state characteristics under time-varying conditions leads to the dynamic change of wind turbine maintenance requirements, which increases the difficulty in accurately dispatching maintenance tasks of wind farm.This paper proposes a dynamic dispatching method for maintenance tasks of wind farm considering the risk situation of wind turbine state under time-varying sea conditions. First, the fuzzy C-means clustering algorithm is used to divide the time-varying working conditions of wind turbines. The improved joint-domain adaptive convolutional neural network is used to minimize the difference of feature distribution, so as to realize the adaptive extraction of the wind turbine condition features under time-varying working conditions. Then, the Markov model is used to describe the initial state transition matrix of each component. Considering the influence of incomplete maintenance on the performance of the unit components, the state risk situation prediction model considering the adaptive state assessment of the wind turbine is established by the component performance degradation model. On this basis, with the constraints of maintenance vessels, personnel and working hours, a dynamic dispatching model of maintenance tasks of offshore wind farm is established to minimize the cost of unit power dispatching and maintenance dispatching, which realizes the dynamic dispatching for maintenance tasks of offshore wind farm under time-varying conditions. Finally, the effectiveness and economy of the proposed method is verified by taking an offshore wind farm as an example.