Abstract: Wind energy is safe and clean, but, the maintenance of wind turbines which are main device to produce wind energy is a challenge and expensive. Exact values of the performance rates (levels), state transition probability and maintenance cost is unable to get when the wind turbine is modeled as an multi-state system (MSS). Therefore, a maintenance optimization method is proposed for the wind turbine under this epistemic uncertainty. In the proposed method, a selective maintenance model is formulated for the wind turbine generator system, which can be viewed as a multi-state system. The Markov decision process is proposed to resolve the selective maintenance problem. By combining the Markov decision process with the fuzzy theory, the reward and the state distribution of the system are represented as triangular fuzzy numbers to quantify the epistemic uncertainty. By the proposed method, the maintenance decision of the wind turbine generator system can be well-optimized.