Abstract: With the construction and operation of cascade reservoirs in large basins, the research on short-term optimal operation of cascade reservoirs has received more and more attention. However, the optimization problem contains many variables and complex constraints, and its model modeling and solution are more and more difficult with the expansion of the reservoir group. Although traditional optimization methods or intelligent optimization methods can obtain optimal solutions that meet relevant objectives and constraints, the results under the optimization algorithm are often too biased towards the target value, resulting in large volatility and randomness, which is difficult to meet the actual scheduling needs. In view of this, this paper proposes the construction technology of power station output scenario based on the historical dispatching results, cluster analysis and other data mining methods, considering the scheduling period, business type, scheduling period length, scheduling objects and other scenario elements, and forms the output scenario set by mining typical output scenarios based on a long series of historical operation data. On this basis, a short-term optimal operation method of cascade reservoirs based on the output scenario set is proposed. Through the case study of Shaxi cascade reservoirs, it is found that compared with the traditional optimal operation method, the scenario method proposed in this paper can not only obtain the output process that meets the actual operation requirements, but also shorten the calculation time relatively, which verifies the feasibility and practicality of this method. The research results have a good reference significance for short-term optimization scheduling of cascade reservoirs in other basins.