Title :
Application of improved neural network algorithm in automated deformation monitoring
Author :
Bao, Huan ; Zhao, Dongming ; Fu, Ziao ; Zhu, Jiang ; Gao, Zhan
Author_Institution :
Service Centre of Meas. Instrum., Zhengzhou Inst. of Surveying & Mapping, Zhengzhou, China
Abstract :
Deformation that happens in the real world is a nonlinear process, and so are the outliers in deformation observations. With the requirements on automation, real-time and accuracy becoming stronger and stronger, it is also more and more important to fast detect and remove the outliers in monitoring observations. In the paper the approximation of nonlinear function mapping relation using artificial neural network (ANN) was introduced, and issues about BP neural network were analyzed. Aiming at the automated deformation monitoring, the method that preprocess the original observations using nonlinear regularization function and memorize original weights or threshold values was proposed, which greatly raised the converging speed of ANN and preventing the model from reaching local minimum and thus improved the accuracy of model fitting. The results of some examples show that the method is easy for programming, real-time and highly efficient, which is suited for automated deformation monitoring.
Keywords :
backpropagation; condition monitoring; convergence; data analysis; deformation; neural nets; structural engineering; BP neural network; artificial neural network; automated deformation monitoring; convergence; deformation observation; model fitting; monitoring observation; nonlinear function mapping relation; nonlinear process; nonlinear regularization function; threshold value; Accuracy; Artificial neural networks; Fitting; Monitoring; Presses; Real time systems; Training; automation; deformation monitoring; neural network; nonlinearity; outlier detection;
Conference_Titel :
Natural Computation (ICNC), 2010 Sixth International Conference on
Print_ISBN :
978-1-4244-5958-2
DOI :
10.1109/ICNC.2010.5583719