Research On The Leakage Diagnosis Method Of Oil Pipeline Based On Extreme Learning Machine

The safe operation of oil pipeline is one of the important factors to keep a healthy and steady development of Chinese national economy.In recent years,frequent leakage accidents of oil pipeline in China has brought a huge loss of life and property.And the improper handling of the case may also bring a severe pollution to the surrounding environment.The safety of oil pipeline is attracting more and more attention.Therefore,the research on the leakage diagnosis method of oil pipeline is becoming increasingly important.Based on the analysis of current research status,it is found that the current method of leakage diagnosis has the problem of high false alarm rate and missed alarm rate because of the complexity of model building,which can not be described by accurate mathematical forms.Therefore,combined with the nonlinear approximation ability of neural network algorithm,an optimized algorithm is proposed to establish the accurate oil pipeline leakage diagnosis model.But most of these studies are lack of consideration of the imbalanced distribution of the sample data in the pipeline.Learning from imbalanced data is a relatively new challenge,attracting growing attention from both academia and industry.As a simple and efficient feed-forward neural network learning algorithm,the extreme learning machine can effectively improve the learning performance of the imbalanced data set by the method of sample weighting.Therefore,this thesis combines the computer technology,sensor techno logy with extreme learning machine to establish the oil pipeline leakage diagnosis model.Firstly,the feature extraction of oil pipeline is carried out by the wavelet packet algorithm,then the extreme learning machine is used to establish the oil pipeline leakage diagnosis model.However,the performance of weighted ELM is greatly affected by the training sample weights set by users.Thus,this thesis proposes a boosting weighted ELM,which embeds the weighted ELM into a modified AdaBoost framework,which makes the sample weight dynamically updated.Then this thesis introduces the cost factor to the algorithm of the weight updating,and makes the proposed algorithm a cost sensitive learning algorithm,effectively improving the diagnostic performance of oil pipeline.Overall,this thesis studies the leakage diagnosis method of oil pipeline based on extreme learning machine.The weighted extreme learning machine is used for the model training,and the problem of the high missed alarm rate caused by imbalanced distribution of the training sample is solved.Then this thesis proposes weight updating method for weighted extreme learning machine training,avoiding the weight extreme learning machine to increase the false alarm rate cost to reduce the missed alarm rate,effectively reduce the pipeline leakage diagnosis of missed alarm rate and false alarm rate.Finally,this thesis introduces the cost factor to the algorithm,further reduce the pipeline leakage diagnosis in false alarm rate.