Since the vibration signal of the reciprocating compressor has the characteristics of non-linear and non-stationary,in order to further improve the fault recognition rate,a reciprocating compressor gas valve fault diagnosis method based on improved complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and composite hierarchical dispersion entropy was proposed.Orthogonality was used as an index to select the best modal function,by which CEEMDAN’ s decomposition accuracy of non-stationary signals was effectively improved and noise residuals was reduced; kurtosis was used as an evaluation index to screen the decomposed IMF components and reconstruct the signal.The composite hierarchical dispersion entropy of the reconstructd signal was solved,the fault feature vector was extracted; the support vector machine was used for classification and recognition.The effectiveness and superiority of the method was verified by the test results.