The 500kW wind generator was developed as one of the projects of NEDO (New Energy and Industrial Technology Development Organisation in Japan), and several tests on its performance have been made. Based on the measurement data, we modelled the NEDO 500kW wind generator, and have investigated a method to reduce power variations under fluctuating wind conditions. This paper presents a control strategy based on a probabilistic optimal control scheme, and makes comparisons between power variations with the conventional PID controller and with the proposed controller. The result showed that the proposed controller could reduce the covariance of power variations to approximately one third.