To ensure the integrity and accuracy of the meteorological data, combined with automatic weather station's daily average temperature data which contained discontinuous noise, three types of membership functions were submitted. A compensation algorithm of Fuzzy Support Vector Machine (FSVM) based on root-mean-square membership function was designed and the compensation model was established too. Finally, the FSVM method was compared with the traditional Support Vector Machine (SVM) method. The experimental results show that the proposed algorithm has good recognition capability for noise points. After interpolation, the data precision was 1.4℃, better than 1.6℃ of the traditional SVM method. Moreover, the whole data precision was 1.13℃, superior to 1.42℃ of the traditional SVM method.