Research And Implementation Of Fluid Pipeline Leakage Identification Based On Deep Migration Learning

Pipeline is an efficient and convenient transportation tool for fluid media such as natural gas,oil and tap water,and it is an indispensable infrastructure for social life and economic production.Due to the influence of factors such as aging,corrosion,natural disasters and urban and rural construction damage,fluid pipelines leak frequently,resulting in waste of resources and economic losses.In addition,due to the large differences in the distribution of the pipe network structure of the fluid transportation pipelines,it is difficult to detect and obtain data,resulting in the lack of standard data sets for different working conditions.Therefore,it is conducive to the normal and safe operation of the pipeline network to study the intelligent diagnosis technology of the leakage state of different medium fluid pipelines,timely and accurately detect the leakage and take effective countermeasures.The main research contents of this paper are as follows:1.Aiming at the problem of low signal-to-noise ratio of pipeline leakage vibration signal caused by complex noise around fluid pipeline,a signal denoising method based on improved variational mode decomposition(VMD)and singular value decomposition(SVD)is proposed.Firstly,genetic algorithm is used to optimize the number of modal components and quadratic penalty factor of VMD,so that it can be adaptively determined according to the characteristics of input signals.Secondly,VMD decomposition is performed to obtain several IMF components and the reconstructed signal is obtained.Finally,one-dimensional signal after noise reduction is obtained by SVD decomposition and reconstruction.Experiments show that this method has better noise reduction performance than the generalized cross-correlation method and the VMD method optimized by genetic algorithm,and the correlation between the collected two leakage signals is increased by 6.92 times.2.Aiming at the problem that the pipeline leakage state identification model of a single medium has poor test accuracy under other medium pipeline data samples,and the data samples of different working conditions are insufficient,we propose a fluid pipeline leakage state identification method based on Convolution Neural network(CNN)and transfer learning.A large number of experiments are carried out on the denoised data set and the collected original data set.The results show that the model is trained in two data sets of water supply pipeline domain and gas transmission pipeline domain,and the classification accuracy of leakage state is the highest,which is basically maintained at more than 99.5%.At the same time,the convergence speed of the model is also faster than that of Alex Net,VGG16,Inception V3 and Res Net50.It is proved that the comprehensive performance of the model proposed in this thesis is outstanding,which verifies the effectiveness of the deep migration learning model in the identification of fluid pipeline leakage state.At the same time,it is also proved that the data set processed by the signal noise reduction method based on VMD and SVD is more conducive to the identification of fluid pipeline leakage state.3.Aiming at the problem of missing data in leakage data samples under different working conditions,this paper designs a generative adversarial networks(GAN)model.The depth migration learning model is used as the discriminator network and the generator network based on convolution neural network for confrontation training.After generating a new data set of fluid pipeline domain with different media,the classification training between different fluid pipeline domains and the classification training of leakage state of gas transmission pipeline are carried out.Experiments show that when the classification model based on CNN performs identification training between different domains under the generated new data set,the accuracy,recall and convergence effect are better than Alex Net model and Res Net50 model.It is proved that the data set generated by GAN network can retain the characteristics of the effective original data set and reduce the impact of data loss on the model training results;As well as the identification model between different domains of fluid pipeline and the identification model of gas pipeline leakage state designed in this paper,the identification accuracy trained under the data set generated by GAN network has reached 100%,and has good stability.4.Use web front-end technology to interact with deep learning and migration learning programs based on Python platform.A web page for cross domain communication is built through the combination of HTML,CSS,Java Script and Websocket,and the console output results after Python execution are displayed through reasonable web page layout.At the same time,it is compatible with a variety of browsers.