The localization accuracy of RFID tag is limited by phase noise and unsuccessful phase unwrapping. In this paper, a preprocessed method of phase data is proposed to improve the localization accuracy of the nonlinear optimization based RFID localization method. Specifically, the principle of phase change prediction is proposed and designed, by which simultaneous phase unwrapping and phase noise reduction filtering are carried out. Compared with the existing localization method, the proposed method is that it can solve the problem of unwrapping failure and reduce the influence of phase noise. The experimental tests indicate that the proposed method can improve the localization accuracy by about 30% compared with the SOTA methods.