Abstract: In gas pipeline leak detection，small leaks are most difficult to detect． And the main difficulties are noise reduction and feature extraction． VMD ( Variational Mode Decomposition ) is an adaptive signal decomposition algorithm with good anti-noise performance and the ability to extract the center frequency of the signal． A VMD improved algorithm is used to detect small leaks． The error energy algorithm is used to screen the IMF of the signal after the variational mode decomposition to obtain the final reconstructed signal． In order to extract the leakage components in the acoustic signal of gas pipeline，a feature extraction method combining VMD and error energy algorithm is proposed． Firstly use this method to decompose the original signal into a series of bandwidth-limited natural modes; then use the error energy algorithm to select the useful mode; finally，use the selected mode to perform signal reconstruction． The experimental results show that the method can effectively filter out the noise in the small leakage signal and effectively reconstruct the original signal．

Key words: pipeline leakage, variational mode decomposition ( VMD) , error energy algorithm, modes selection