关键词: 泄漏孔径识别, 总体局域均值分解（ELMD）, KL散度, 稀疏表示分类器, 过完备字典

Abstract: Natural gas pipeline leakage was influenced by the aperture, the sensor distance, the pressures in the pipeline and many factors, so the feature extraction and recognition algorithm is relatively complicated. A novel leak aperture identification method which combined feature extraction based on ELMD-KL model with SRC was proposed. ELMD was applied to adaptively decompose leak signals, to obtain characteristic informations of different aperture leak signals, and to extract the principal product function(PF) components based on KL divergence which contained the main leakage informations. The method extracted multiple characteristic parameters in time domain and frequency domain as the feature vectors. For the classification of small sample complex signals, a SRC was put forward to realize the accurate classification of leak apertures. The classifier obtained the most sparse solutions of the test signals with overcomplete dictionary. The solutions were used as the sparse coefficient to reconstruct the test signals and obtain reconstruction signals in different classes of the test signals. Finally, classification of leak apertures was accomplished by judging the residual values between test signals and reconstruction signals. The experimental results show that the proposed algorithm has higher recognition accuracy compared with the traditional classification algorithm of SVM and BP.

Key words: leakage aperture identification, ensemble local mean decomposition (ELMD), KL divergence, sparse representation classifier(SRC), overcomplete dictionary