Research On Analysis And Back-analysis Of Deformation Monitoring For Dam Based On Aritificial Neural Network

Recently, artificial neural network is spreading and extending in the field of Dam safety monitoring. Artificial neural network is a kind of implicit expression without explicit expression as statistical regression model, and it is its disadvantage. But because of high nonlinear function and global convergence, good capacity of error tolerance and associative memory, strong self-adapting and self-learning ability of artificial neural network, it is able to get better accuracy of fitting and prediction than traditional regression analysis methods. If the network is trained when the results of finite element analysis is as its input and corresponding mechanical parameters is as its output, we will gain actual mechanical parameters when the monitoring data is as input of the trained network. So this paper will discuss about research of analysis and back-analysis of deformation monitoring for dam based on artificial neural network.After exposing model structure and learning rules of BP network, and analyzing the deficiency of conventional BP algorithm that it only separately calculates back and forth with every specimen and leads to slower rate and precision of convergence, this paper puts forward L-M BP neural network based on traditional numeric optimization algorithm-L-M algorithm and applies it on analysis and back-analysis of dam deformation monitoring. Against classical BP network, L-M BP network can get higher prediction precision and needs much fewer training times with the same goal. The merits of parameters back analysis with neural network is that there needn't create explicit identification format and may use in the multi-parameter system in the contrast of traditional parameters back-analysis methods.In this paper, the statistical model and mixed model of dam deformation monitoring is as the basis of the research that artificial neural network is applied to analysis and back-analysis of deformation monitoring for dam. In the paper, the deformation monitoring model of Jinshuitan dam is established with L-M BP network, and the analysis result is that L-M BP network model is effected less by monitoring data serial's length in the course of dam deformation prediction when comparing to the statistical model. And in the paper, the dam body and foundation average elastics of Jinshuitan dam is get by back-analysis with L-M BPnetwork, and the average elastics is proved better by mixed model, so we can conclude that the method is feasible and the results is credible when the L-M BP network is applied to back-analysis of dam deformation parameter.