Abstract: Considering that it is difficult to obtain the optimal operation scheme from the conventional reservoir operation diagram and that the stochastic optimal operation solution of cascade reservoirs causes “dimension disaster”, this paper takes Baishan and Fengman cascade reservoirs hydropower stations in the second Songhua River in northeast China as examples to analyze the actual runoff characteristics of cascade reservoirs, based on the characteristics that the natural runoff of cascade reservoirs in the same basin has strong correlation, a reduceddimension Markov chain stochastic runoff description model is proposed to solve the dimension disaster problem of the stochastic optimal operation model of cascade reservoirs. At the same time, the description method of runoff forecast with and without time intervals is used to make the runoff description more practical. According to the proposed runoff description model, the maximum expected value of power generation during the operation period is used as the objective function of the optimal operation model, and the model recurrence equations under the conditions of no runoff forecast in the adjacent period, no runoff forecast in the adjacent period, no forecast in the current period and no forecast in the later period, no forecast in the current period and no forecast in the later period are studied. The optimal operation model of cascade reservoirs considering the combination of runoff forecast in the current period and no forecast in the later period is created for the first time. The method of combining the penalty coefficient and changing the guaranteed output constraint is used to solve the optimal decision scheme of the reservoir under various output guarantee rates. Under the same output guarantee rate, through a comparative analysis of the simulated optimal operation results of the historical runoff series of Baishan and Fengman cascade reservoirs, it can be seen that the optimal operation rules obtained in this paper obtain more power generation benefits than the conventional reservoir operation chart and the traditional stochastic optimal operation rules, and the model solving time is also several times more than that of the traditional method. It fully illustrates the effectiveness of the model and the superiority of the scheduling rules created in this paper.