An Integration Method Using KPCA And SVDD For Pipeline Leak Detection With Multiple Operating Modes

As one of the effective and economical transportation methods for fluid media,pipelines are often affected by factors such as corrosion,aging,weld defects,and third-party damage,resulting in frequent leakage accidents.Therefore,an effective pipeline leak detection method is a powerful means to ensure pipeline integrity and safe operation.Due to the fact that there are few samples of leakage signals and the complicated site conditions during the actual pipeline operation,the change of various working conditions increases the difficulty of leak detection.How to detect various working conditions in the absence of leaking samples is the focus of current research.The main contribution of this paper is to propose a multi-operating pipeline leak detection method based on kernel principal component analysis and cascade support vector data description.Numerical simulation experiments verify the effectiveness of the proposed method in pipeline leak detection.The main work contents are as follows:(1)By consulting and collecting relevant domestic and foreign literatures,the development trend of pipeline transportation,its research background in leak detection and pipeline leak detection methods are introduced,and the research status of pipeline leak detection methods based on support vector data description is analyzed.The basic principles and contents of local mean decomposition algorithm,kernel principal component analysis method and support vector data description are summarized.(2)Aiming at the problem of difficulty in obtaining samples of leaking signals during pipeline leak detection and the diversity of operating conditions,this paper proposes a Kernel Principal Component Analysis(KPCA)based on non-leakage sample data during pipeline operation and Support Vector Data Description(Support Vector Data Description,SVDD)multi-case pipeline leak detection method.First,local mean decomposition is performed noise reduction on the measurement signal,and the time domain and waveform eigenvalues is solved of the noise reduction signal.KPCA is used to reduce the dimension and extract the non-linear kernel principal components to obtain better signal description eigenvalues.Second,K mean(K-means)clustering algorithm is used to identify multiple operating conditions and trains for different operating conditions data respectively,and multiple support vector data description models are established to obtain the decision boundary of the corresponding hypersphere;finally,according to the operating condition occurrence probability the multiple SVDD models are sorted by probability occurrence,and the Cas-SVDD model's multi-operating pipeline leak detection is realized.(3)In view of the fact that a small number of abnormal points in the actual training normal samples have a negative effect on the established Cas-SVDD model,which leads to the decrease in the accuracy of the pipeline leakage detection by the KPCA-Cas-SVDD method,this paper proposes a cascade fuzzy support vector data description Fuzzy Support Vector Data Description,(Cas-FSVDD)pipeline leak detection method.First this method is distinguished various working conditions through the fuzzy K-means algorithm;then the fuzzy membership function is introduced into the support vector data description,the relative density of the data samples are used to calculate the fuzzy membership value,and the training model is different according to different samples The contribution rate of each sample is given an appropriate fuzzy membership value,and the samples with abnormal points are assigned smaller fuzzy membership values,and the other samples are assigned larger fuzzy membership values,thereby using the fuzzy membership of data samples from different operating conditions Multiple FSVDD models are established with degree values to obtain a more compact hypersphere description boundary and the accuracy of pipeline leak detection is improved.