The reversible pump turbine (RPT) is likely to enter the S-characteristic zone, thereby inducing pressure fluctuations and oscillations to the grid connection. An innovative optimization framework of RPT runner is presented to mitigate the detrimental flow conditions associated with the S-curve under turbine mode. Comprehensive runner geometric parameters were considered with the optimal Latin hypercube (OLH) sampling technique to generate different designs. Computational fluid dynamics (CFD) was adopted to characterize the RPT hydraulic efficiency and unstable S-characteristics curve, in which the position of the second inflection point was innovatively selected as the objective function. The CFD-driven surrogate-based design methodology was achieved by artificial neural network (ANN). The multi-objective optimization evolutionary algorithm guided the search for the optimal runner configuration with high efficiency and improved S-characteristics. The vortices in the runner channels and high-speed water ring in the vanless area both blocking the flow passage are alleviated in the two selected optimized RPT runners. The total pressure head associated with the intensity of vortices is deceased in the optimized runner, resulting in the improved S-shape characteristic. Runner with higher arches and negative blade lean angle of leading edge is conducive to the smooth streamline and avoidance of the flow separation.