Research On Identification Technology Of Oil Pipeline Condition Based On GIF Elman Neural Network

In recent years, with rapidly development of pipeline transportation industry in our country, the position of pipeline transportation is growing importance in national economic system, so the safety of pipeline transportation is attracted widely attention. Pipeline leak detection is an important part of pipeline safety transportation. Pipeline leak belong to failure condition, so the technology of pipeline condition is an important means of pipeline leak detection. Whether the incidence of oil pipeline leakage could be detected and determined accurately and timely through this technology, it is of great practical significance to reduce economic losses, protect the natural environment and ensure the property of resident safe around the pipeline.View of the characteristics of pipeline transportation system are strong time delay, strong nonlinearity and strong interference, it is difficult to establish accurate mathematical models through the mechanism analysis, and the pipeline transportation is dynamic system which is complex, so the model which should be met requirement that is dynamic. Thus, the model of condition analysis for pipeline transportation system based on GIF (Global Information Feedback) Elman neural network which belong to dynamic recursive type for safety problem of pipeline transportation is proposed. The model is used to identify pipeline conditions accurately and fast, in order to reduce false positives rate and missing report rate, or eliminate the existence of them, and improve the accuracy of pipeline condition information. The main contents in the following aspects:Firstly, the improve direction of Elman neural network is proposed that is GIF Elman for Elman lack of dynamic performance. And then, feasibility of the modeling plan is conducted after analyze characteristics of dynamic recurrent neural network base on GIF Elman neural network and the pipeline system. Secondly, a data pre-processing method is proposed for the characteristics of pipeline pressure data is strong nonlinearity and with strong noise. In order to achieve the effect that reduces data dimension and eliminates noise, the feature of pipeline pressure data is extracted, and then analysis these features using kernel principal component analysis technology. Thirdly, the model of GIF Elman neural network is derived theoretically according to Elman. And the model of condition analysis for pipeline transportation system based on GIF Elman neural network is established. Finally, GIF Elman is verified not only the stability of results of training and testing, but also the generalization of trained network are all have a good performance, through comparing the BP feed forward neural network, the standard Elman neural network and GIF Elman neural network use of field data of pipeline.In short, GIF Elman neural network is ideally suited for modeling pipeline leak detection system. In this thesis, feasibility of the plan is verified both from theoretical and experimental results. Both stability and generalization ability of condition analysis model for pipeline transportation system established by GIF Elman neural network are achieved requirements and results which are desired. The difficulties in identification of conditions are well solved by this model, and the accuracy of identification of conditions has been improved greatly.