Drag coefficient sequence Cd is obtained by solving Tiangong 1 continuous 55 days (d) GPS orbit data with different arc length. The same period solar radiation flux f 10.7 and geomagnetic index ap, Ap sequences are high and low frequency multi-wavelet decomposed. Statistical analysis results of the multilayer sliding correlation between space environmental parameters and decomposition of Cd show that lag correlation between the satellite drag coefficient sequence after wavelet decomposition and the corresponding level of f 10.7 , Ap sequence is good. It is also verified that the Cd prediction is feasible. Prediction residuals of Cd with different regression models and different sample lengths are analyzed. The results show that the case setting sample length of 20 days and selecting f 10.7 regression model is best. It is also show that NRLMSIS-00 model's response in the region of 350 km (Tiangong's altitude) and low-middle latitude (Tiangong's inclination) is excessive during the ascent phase of geomagnetic activity Ap and is inadequate during decline phase. Additionally, for the low-frequency decomposition sequence NRLMSIS-00 model's response is appropriate during the ascent rising phase of f 10.7 . For the high frequency decomposition sequence, NRLMSIS-00 model's response is small-scale inadequate during the ascent phase of f 10.7 and is excessive during the decline phase of f 10.7 . Finally, the summary of potential use and outlook are listed; this method has an important reference value for improving the spacecraft orbit prediction accuracy.

%U http://www.yhxb.org.cn/EN/10.3873/j.issn.1000-1328.2015.02.003