Abstract: Multi scale Permutation Entropy（MPE） is an analytical method that can describe the complexity of time series and has been widely applied in feature extraction of reciprocating compressors. To solve the problems of MPE in coarse-grained computing, the variance method is replaced by the mean method in coarse-grained computing. A variance based multi-scale permutation entropy（VMPE） method is proposed, and its stability and superiority are studied by analyzing Gaussian white noise signals and comparing them with MPE. Based on the advantages of VMPE, a fault diagnosis method for reciprocating compressors based on VMPE and ELM is proposed and applied to the fault diagnosis of reciprocating compressor valve. The experimental results show that this method can identify the type of valve fault and has a higher fault recognition rate.