Abstract: To improve the energy recovery efficiency of the axial-flow pump as turbine, research on its hydraulic performance optimization was conducted based on entropy production theory and particle swarm algorithm. The entropy production theory method was introduced to analyze the entropy production losses in various flow components of the axial-flow pump operating in reverse as a turbine. It was found that under small flow conditions, design conditions, and some large flow conditions, the impeller was the flow component accounting for the largest proportion of the total entropy production loss in the entire machine. Due to the matching relationship between the guide vanes and the impeller, synchronous optimization design of the impeller and guide vanes was required. The particle swarm algorithm was adopted to carry out the optimization design of the impeller and guide vanes. Firstly, a parameter sensitivity analysis was conducted on the input parameters of the impeller and guide vanes to screen out the parameters most sensitive to the response variables of head and efficiency. Secondly, the advanced Latin hypercube sampling method was utilized to establish the sample database data, and high-quality response surfaces and optimal prediction models were employed to construct an approximate prediction model. Finally, global optimization was performed on the approximate prediction model based on the particle swarm algorithm to obtain the Pareto solution set. The optimal solution was identified from the Pareto frontier solution set, and the optimized impeller and guide vanes were obtained through the data of the optimal solution. Through comparative analysis of the external characteristics of the axial-flow pump as turbine before and after optimization, it was found that the hydraulic loss in the flow passage of the optimized turbine was significantly reduced, and the efficiency, head, and power were all improved under all flow conditions.