Research On Prediction Model Of Wall Thickness Of Natural Gas Storage Pipeline Based On Grey Theory And Neural Network

The underground gas storage is one of the important components of the gas transmission network due to its unique advantages in peak shaving and safe gas supply.The Hutubi underground natural gas storage is China 's second natural gas storage that has been put into production outside the Tianjin Dazhangtuo underground gas storage,and it is also the largest natural gas underground storage in China by far.The Hutubi gas storage has the advantages of large storage capacity,large adjustment working gas volume,high safety and low cost,and has the dual functions of seasonal peak shaving and emergency reserve.It has greatly improved the current situation of insufficient gas supply in Xinjiang in winter,and as the first large-scale supporting system for the West-East Gas Pipeline,it also has a place in ensuring the national energy strategic reserve.Corrosion of gas pipelines in gas storage,that is,wall thickness change is the focus of people's attention,so it is particularly important to be able to establish a highly accurate prediction model for wall thickness change.This paper focuses on the direction of the wall thickness prediction model and based on the test data of Hutubi gas storage in previous years,carried out the following work:According to the characteristics of the wall thickness of the gas storage,the gray prediction model and the neural network model are analyzed,and the combination of the gray GM(1,1)model and the BP neural network is studied to realize the traditional GM(1,1)prediction After that,the combination forecasting model No.1 and No.2 was established by combining gray GM(1,1)and BP neural network.G-B-? uses the characteristics of the BP neural network to adjust the error generated by the GM(1,1)model.Its core is to first use the GM(1,1)model to process the original sample data.According to the processing results,you can calculate the traditional The error of the model,the obtained error is used as the input layer information of the BP neural network,and regression training is performed on it.After convergence,the error of the predicted value of the traditional model can be obtained.The final predicted value is the sum of the error after the training and the predicted value.The G-B-? first uses some data(in this paper,five data,six data,and seven data respectively)to construct the GM(1,1)model group,find the restored values ??of these model groups,and then combine these restored values It is imported into the BP neural network as an input layer,and the actual value is used as a tutor to train the model.After that,the predicted values obtained by the original model group are substituted into the trained network,and the new predicted values after correction can be obtained by using the constructed mapping relationship.After verifying and analyzing the two models of Hutubi gas storage,it can be seen that the error is greatly reduced compared with traditional methods.After predicting the data using the two prediction models,several pipeline sections with large thinning were screened,and the remaining strength evaluation based on ASME B31.G was performed on these pipeline sections,which reduced the workload compared with the individual evaluation.