Aiming at the poor accuracy of displacement calculated by using polished rod acceleration signal in pumping unit, a new method combining complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) adaptive denoising and (extremepoint symmetric mode decomposition) ESMD removing trend items was introduced. CEEMDAN was used to decompose acceleration signal; then according to the correlation coefficients between the autocorrelation function of the intrinsic mode functions and the autocorrelation function of the original signals, cutoff point of noise components and effective components was sought out; furthermore, useful signal in the noise component was extracted by using the wavelet threshold method, and the number of wavelet decomposition was determined adaptively based on the property of permutation entropy; all the intrinsic mode functions were reconstructed to realize denoising; displacement was obtained by using the integral transformation method for denosing acceleration; finally ESMD was used to remove trend item. The simulation analysis and measured data show that the proposed method is more accurate and adaptive than the existing methods.