Research Of The The Magnetic Leakage Signal Anomaly Detection Based On Machine Learning

With the rapid development of the world economy,the demand of oil and gas continues to rise.The pipeline of oil and gas plays a very important role in the national economy,which is known as "energy blood".The magnetic flux leakage(MFL)detection the is recognized as the most effective means in the pipeline industry.The anomaly detection and identification of the MFL signal is the key step in the processing of the analysis of the MFL data,which is of great significance.In this paper,the prime task is the analysis of the MFL data which is collected from the metal pipelines,and the following four aspects are completed.The feature extraction of different types of MFL signals is completed.An adaptive threshold MFL anomaly detection algorithm is designed.A MFL anomaly detection algorithm based on Boosting algorithm is designed.A MFL anomaly detection algorithm combined with KPCA and Boosting is designed.First of all,to understand the characteristics of different types of MFL signals,different types of MFL signals is analyzed.After the analysis of the characteristics of the MFL signals is completed,a feature extraction algorithm of the MFL signals is designed based on the wavelet transform to realize the feature extraction of different types of MFL signals.Then an adaptive threshold MFL anomaly detection algorithm is designed.Based on the features extracted,the adaptive threshold of the feature values is set and the MFL signal anomaly detection algorithm is designed according to the threshold of the feature quantity.Complete the exploitation of the algorithm based on the Matlab software platform.Validate the effectiveness and the accuracy of the algorithm using the experimental field data,and analyze the shortage of the algorithm.Aiming at the shortages of the adaptive threshold MFL anomaly detection algorithm,a new algorithm of MFL anomaly detection based on Boosting is proposed and designed.Based on the Matlab software platform,a Boosting classifier is trained.Comparing the accuracy of the Boosting classifier in different situations and a optimal classifier is selected.The experimental data is tested by trained classifiers.Finally,in order to improve the detection accuracy furtherly,A MFL anomaly detection algorithm combined with KPCA and Boosting is designed.The KPCA dimensionality reduction method is used to select the features,and the data after the dimension reduction is input to the Boosting classifier to train,which not only reduces the training time of the classifier,but also improves the detection accuracy of the algorithm.