Method Research Of Pipeline Failure Mode Diagnosis Based On Neural Networks

Pipeline transportation is the most economical and reasonable transport mode of oil and natural gas. With the large number of oil and gas pipeline laying and the growth of service time, the failure incidents of pipeline happened frequently. There are many factors influencing pipeline failure. Some of them are random, fuzzy, incomplete and other characteristics, traditional diagnostic methods are often not adaptive for pipeline failure mode analysis. Artificial neural networks(ANN) is an artificial intelligence model which is researched and applied extensively in the field of pattern recognition in recent years. Because it has highly nonlinear mapping capability, large-scale parallel processing and good adaptive learning mechanism, ANN is very suitable for solving the problems that the methods of traditional pattern recognition are difficult to model. Therefore, the methods and technologies of ANN are applied to research the problem of pipeline failure mode diagnosis, and they have a good adaptability.In view of some issues of pipeline failure mode diagnosis, the paper mainly researches the ANN model, learning algorithm and its application technology suited for problem solving, and combines ANN with diagnosis theory, pattern recognition, fuzzy logic and system simulation, etc. On the basis of analyzing gas pipeline failure modes and fault diagnosis modeling technology, we summarize three types of pipeline failure mode diagnosis problems, they are numerical mode, fuzzy information mode and dynamic mode.The paper constructs different neural networks model in order to achieve the solution of above-mentioned different problems. To the numerical model, an adaptive method used to identify BP networks structure could recognise the failure modes of pressure pipes with defects. To the fuzzy information mode, considering the unclear relationship between conditions and results and the importance degree of conditions impacting on results, a weighted fuzzy reasoning networks is constructed based on traditional fuzzy neural networks. It solves the fuzzy information of corrosion data impacting on pipeline corrosion degree better. To the dynamic model, process neural networks is combined with RBF neural networks, and the concept and model of RBF process neural networks are introduced. The model integrates the advantages of both, and it has a good adaptability to the prediction problem of pipeline corrosion rate changing nonlinearly with time. At the same time, in view of the problem of process variables trend prediction, the structural ideas and methods of traditional support vector regression machines are extended to time-varying function space. We establish a process support vector regression machines. The model can solve the time prediction problem of dynamic system better.The paper uses neural networks technology to research pipeline failure mode diagnosis, it evaluates the safety of oil and gas pipelines. It can provide a scientific basis for risk assessment and management decision-making of pipelines, and it has important practical significance and application prospects.