The invention discloses a high-speed train rail side movement parameter self-adaptive recognizing method. The method includes the steps of firstly, conducting downsampling and filtering on signals X(t) collected by a rail side microphone to obtain x(t); secondly, conducting short-time Fourier transform (STFT) on the x(t) to obtain time-frequency distribution STFTx(t,f); thirdly, initiating a railside model parameter group gamma{v,r,f0}; fourthly, establishing a Doppler window wgamma(t,f) meeting a Doppler time-frequency variation rule on the basis of the frequency shift formula fk(t) of parameters and the rail side signals in the third step; fifthly, meeting the equation: x0(t)=wgamma(t,f)*STFTx(t,f); sixthly, sequentially calculating corresponding signal energy values E of the area of the established Doppler window in different frequency bands in sequence from low frequency to high frequency in the whole time-frequency domain; seventhly, executing the third step, the fourth step, thefifth step and the sixth step again until the obtained maximum energy value Emax is obtained, and making the gamma{v,r,f0} corresponding to the maximum energy value Emax serve as the rain movement parameter recognizing result. The anti-noise capacity and parameter estimation self-adaptive degree are improved, and the method can be used for train bearing acoustic signal failure detection.