The wind power forecast error leads to the reduction of the operating economy of the power system and the increase of the wind farm assessment cost. An optimization strategy for wind farm power reporting based on minimization of assessment power is proposed. Firstly, the improved K-means algorithm based on the differentiation of distribution function is used to establish the probability distribution model of forecast error with obvious differences under different meteorological modes. Secondly, the support vector machine is used to realize the meteorological mode recognition, and then the Latin hypercube sampling method considering the time correlation is used to generate the day-ahead power curve scenario set. Finally, an optimization model of the day-ahead power reporting with the goal of minimizing the expected value of the assessment power is established, and the particle swarm algorithm is used to solve it. Taking the historical data of a wind farm in central China as an example, it is verified that the proposed method can effectively improve the operational economic benefit of wind farm compared with the strategy of directly reporting the forecast wind power.