Research On Pipeline Leakage Detection Method Based On Deep Belief Network

As one of the main energy sources,oil and gas play an important role in the development of society.As the main transportation mode of oil and gas,it is very important to ensure the safety of pipeline transportation.Due to human destruction and natural factors,pipeline leakage accidents occur,resulting in serious economic losses and ecological problems.Therefore,it is of great significance to detect the leakage of pipeline network.The change of pressure in the normal condition of the station is similar to the change of the pressure caused by the leakage,which leads to the higher false alarm rate of the existing leakage detection system.According to the pipeline information collection more interference factors and the condition adjustment of complexity,this paper presents pipeline leakage detection method based on deep belief networks,mainly in the following three aspects:data filtering,pipeline leak detection algorithm and algorithm efficiency optimization.Firstly,aiming at the situation that the signal collected by the pipeline is more interference and the effect of denoising is poor,the Variational Mode Decomposition(VMD)is proposed.This method is better than the traditional method in anti-jamming and low frequency signal separation.Aiming at the problem that VMD method is difficult to select K with predetermined size and difficult to determine effective IMF components after decomposition,a pipeline leak detection method combining VMD algorithm and correlation coefficient is proposed.The experiment shows that the method can effectively improve the effect of signal filtering.Secondly,the pipeline leakage signal parameters and adjustment of working conditions is complex,cause leakage detection system of high false alarm rate,this paper introduces the theory of deep learning depth based on belief network(Deep Belief Network,DBN),the information of the monitoring station equipment condition and pipe characteristics data,constructs the model of pipeline leak detection a deep belief network based on.According to the DBN network model of the detection performance of the network structure and the parameter sensitive problem,through the analysis of the situation of abnormal leakage of pipelines with different characteristics,the parameter optimization of network oriented engineering;test in the actual pipeline data on the results show effectiveness of the proposed.Thirdly,in order to solve the problem of slow convergence in the training process of DBN network,an improved method for RBM training and global adjustment(fine-tune)is proposed.First of all,in the training process of RBM,avoid the Bureau advantages due to the vicinity of gradient changes slowly and bring training time-consuming by impulse factor;the reconstruction error of the traditional impulse factor in the early and late in the training set for different constant due to abnormal fluctuations,the adaptive momentum factor,to achieve a continuous change in training in the iterative process of impulse factor;in the global adjustment,using the parallel processing method to replace the traditional K fold cross validation in serial mode,effectively improve the training speed.