| Pipeline transportation is crucial in Chinese industrial development.If the pipeline leaks during its service,it will lead to energy waste and even safety accidents.Different types of leakage sources require corresponding measures to be taken for remediation,so accurately identifying the characteristics of the leakage source can help maintenance personnel take timely and correct measures to repair pipelines.Based on the curse of dimensionality problem of massive data collected when detecting pipeline leakage by artificial intelligence method,and the fact that there will be a lot of noise interference in the actual signal,this paper aims at the above problems,in order to further improve the diagnostic efficiency and accuracy of pipeline leakage detection,combines the feature dimensionality reduction method of fusion of Relief F algorithm and factor analysis with wavelet packet decomposition,and applies it to the optimization of pipeline leakage identification process.The main tasks are as follows:(1)This article proposes a feature dimensionality reduction method that combines Relief F and factor analysis.Based on the correlation between signal features and the shape and size of pipeline leakage sources,the Relief F algorithm is used to assign feature weights.The higher weighted features are inputted as sensitive features into factor analysis.The sensitive features are decomposed and their common factors are extracted using factor analysis,and then the common factors are constructed as new features.The distribution pattern of new features in the feature distribution map can reflect the differences between different categories of pipeline leakage sources.(2)Noise reduction of pipeline leakage acoustic emission signals based on wavelet packet features.In response to the high noise content in the collected signals in practical applications,wavelet packet decomposition is first used to decompose the acoustic emission signal of the pipeline leakage source with added noise into multiple sub bands according to frequency.Based on the energy proportion,sub bands are selected as feature vectors,and then time-domain and frequency-domain features are extracted to form wavelet packet features.After calculation,it can be concluded that the new feature composed of common factors extracted from the wavelet packet features has a high correlation with the pipeline leakage source,which is suitable for pipeline leakage detection.(3)Pipeline leakage source feature recognition based on support vector machine.Input the feature sets before and after dimensionality reduction into the support vector machine to train the recognition model for pipeline leakage,and judge the quality of the feature set based on the recognition accuracy and running time of the model.The calculation results show that the dimensionality reduction method proposed in this article can minimize the dimensionality of features while ensuring that the useful information of the feature set is not excessively lost.Compared with the feature sets obtained by other dimensionality reduction methods,the feature set constructed in this paper has a higher accuracy and lower variance compared to the model trained in the support vector machine,making it more stable.This article proposes a machine learning recognition method for pipeline leakage sources that combines feature reduction and wavelet packet features.The experimental results show that this method have high recognition accuracy and certain theoretical value. |
