The distribution of measured operating data of hydraulic turbines is non-structural and uneven density. The moving least squares method was used to fit the measured operating parameters to obtain the spatial characteristic surface of the water turbine, which can reflect the local characteristics of data surface and avoid the distortion caused by the differences of data distribution density. The Riyadh criteria was used to perform preliminary data filtering. The support radius was adaptively determined by local fitting across different ranges of discrete sample points. The robust moving least square surface reconstruction model was established combined with the Mahalanobis distance denoising method. Compared with the least squares method, the example calculation results show that the moving least squares method fits the operating characteristics of hydraulic turbines with higher accuracy and supplements the local data effectively. The result can reflect the operating characteristics of hydraulic turbines authenticity, which provides a reference for correcting the operating characteristic curve given by the turbine manufacturer.