Research For The Structural Deformation Of Metro Rail Line Interval Based On Neural Network Technology

Nowadays, with the rapid development of China’s national economy, the city subway has become a important traffic travel way, it is also the key national urban transport infrastructure construction. In view of the importance of the construction of the subway in the city and people’s life, safety monitoring and deformation forecasting subway construction is an important issue. In safety monitoring, it is recognized that deformation monitoring is only a means, the ultimate goal is to predict the deformation monitoring. Select and build a reasonable model to predict the deformation of subway structure instability is the purpose of our security monitored.In this paper, the monitoring data of the subway near the NanJing Railway as an example.Firstly, the importance of safety monitoring and forecasting is described, and the monitoring methods of subway structure at the present stage are summarized. Secondly, in view of the application of "free-stationing" method has been widely used, and it has been proved by practical example that a suitable standard benchmark stability criterion is introduced. Finally, the neural network itself robust adaptive, self-learning and fault-tolerant ability and deal with nonlinear problems advantages, combined with wavelet transform filtering denoising method is proposed a applicable to subway monitoring in the operation period of the data analysis model, the main research contents are as follows:(1)This paper studies the theory and method of safety monitoring of subway structure, and introduces the method of deformation monitoring in Nanjing metro line two. On the basis of the above, the automatic deformation monitoring and prediction process of subway structure in operation period are summarized.The innovation point lies in the fusion of multi-source data, the real-time forecast, which shows the superiority of the method.The proposal can be used in the similar underground long and narrow space structure monitoring.(2)Furthemore, study the related theory of the analysis of the stability of monitoring datum points the subway structure deformation, put forward a kind of application on the "free-stationing" method of benchmark stability analysis method of test, as an example, and analyzes the applicability of the method, the method for free net adjustment has certain applicability.The results showed that to eliminate unstable reference point and adjustment error of unit weight is greatly improved.(3)Research on the basic principle and process modeling of BP neural network (BPNN), analysis the different wavelet de noise principle and choose the reasonable wavelet denoising model according to the characteristics of operating subway monitoring data, after the data analysis, the Symlet4 wavelet is chosen to carry on the three layer decomposition, and puts forward the small wave to noise and artificial neural network combination method is applied to deformation monitoring data analysis of ideas.Finally, combined with the specific monitoring data, constructing reasonable wavelet algorithm and neural network structure, the results of the comparison between the predicted data and the actual value are analyzed, and the results of the traditional BP neural network method and the BP neural network method after wavelet denoising are compared. And with the traditional multiple linear regression method, grey model method are compared, assessment of prediction accuracy, based on their own observations with high accuracy. According to the amount of elevation as the target value, from the aspect of mean error, The results showed that traditional BP neural network model is ± 0.9539mm, Symlets wavelet to noise based on BP neural network method is ± 0.6727mm, multiple linear regression model is±1.0567mm, grey model method is ± 0.8416 mm, it is concluded that the fusion model has a high precision of the conclusion, show that the method has strong practicability.If the settlement is the target value, wavelet de noise, BP neural network method to forecast mean error is±0.2166mm, and four kinds of model prediction accuracy were improved, come to the accumulative land subsidence to predict target is better.