With the increase of wind power penetration rate in the power grid, power dispatching including wind curtailment often happens to guarantee the safety of the power system. Considering the fatigue load of wind turbines during de-loading operation, how to optimally dispatch them along with the required power instructions becomes a hot issue. Under this background, this paper mainly studies the active power dispatching based on fatigue load optimization. In order to reduce the calculation complexity of the fatigue load, this paper proposes a simplified active power dispatching model for wind turbines via a look-up table. Subsequently, the ultra-short-term wind speed prediction is performed using the least squares support vector machine for the dynamic's prediction of wind turbines, which will be applied in fatigue load pre-calculation. Moreover, in order to fast solve the optimal dispatching problem, the accelerated particle swarm optimization algorithm is adopted. Under different de-loading scenes, the simulation experiments show that the fatigue load of the wind turbines has certain regularity while different de-loading strategies yield different effects on the fatigue load. Finally, based on the evaluation of damage equivalent load (DEL), the proposed optimal dispatching strategy in this paper is proved to be effective to meet the external dispatching from the power grid while reducing the total fatigue load level of wind farm via optimizing the power instruction to each wind turbine. INDEX TERMS Active power dispatching, de-loading operation, fatigue load, wind farm, wind speed prediction.