Slow feature analysis (SFA) is an effective method for extracting slow-changing features from fast-changing signals. Its proposal enriches the means of reconstruction of non-stationary system's driving force signals. Two-dimensional non-stationary system model be constructed based on Henon chaotic mapping. The authors try to test the ability of reconstructing driving force signals from two-dimensional and complex non-stationary system by SFA method. The experimental results show that the SFA can successfully extract the driving force signals from the non-stationary time series with one time-varying parameter. The driving force signals were also successfully extracted from the non-stationary time series with two time-varying parameters by SFA and wavelet transform technology. In addition, The driving force of Beijing air temperature was reconstructed by using SFA method. Wavelet transformation technique is then used to analyze the scale structure of the derived driving force. These efforts will provide new ideas for the study of climate system's driving force.