In this paper, based on the theory of wavelet transform, presents several new estimation method of time varying long memory parameters, and gives the consistency and asymptotic properties of estimators of these new methods. Finally, this paper studies the behavior of small samples of non-stationary long memory process, gives the reference solution of some factors selected which can affect the accuracy of estimate such as wavelet scale et al. At the same time, this paper compares the advantages and disadvantages of various estimation methods, and proves the effectiveness and robustness of the new methods.