Abstract: To explore a new method for solving the optimal operation problem of a hydropower station group, this paper presents an improved firefly algorithm (IFA) that adopts the large-scale random walk, dominant reservation, and mutation mechanism. A mid-long term optimal operation model of maximal cascade power generation is developed and solved using IFA in a case study of the Heihe hydropower station group. The following conclusions are obtained: (1) When the power generation is maximized, the leading reservoir is filled up rapidly in high water season and maintains a high stage operation, and the reservoir stage goes down gradually to the dead water level in the dry season. (2) Optimal values of a disturbance factor, a fluorescence absorption factor, and stochastic search interval algebra are obtained using parameter sensitivity analysis. (3) Compared with two other algorithms, IFA is better in accuracy, stability, and optimization ability. The results are practically significant to the operation of the Heihe cascade hydropower stations.