To address the high-dimensional and nonlinear complex optimization problems in the optimal operation of cascade reservoirs, a two-stage multi-objective improved artificial fish swarm-particle swarm optimization (TMIAFS-PSO) algorithm was proposed. This algorithm employs segmented mapping to expand the search space of the initial population, and enhances local and global search capabilities by adjusting the adaptive step size and introducing a diversified movement strategy. Additionally, the algorithm adopts a two-stage filtering strategy to retain particles that meet the constraint conditions and incorporates an improved artificial fish swarm optimization strategy to further expand the particle search range. A case study was conducted on the cascade reservoir group consisting of Wudongde, Baihetan, Xiluodu, and Xiangjiaba in the lower reaches of the Jinsha River. The results indicate that, compared to other algorithms, the Pareto solution set of the TMIAFS-PSO algorithm exhibits better convergence and uniformity, demonstrating the superiority of this algorithm. By analyzing the water level variations of the operation schemes generated by the TMIAFS-PSO algorithm, a relatively stable optimal operation scheme for this cascade reservoir group is summarized.