Nonlinear time series of deformation forecasting using improved BP neural networks

Cai-yun Gao^{1, 2}, Xi-min Cui¹

¹College Of Geoscience and Surveying Engineering, China University of Mining and Technology, Beijing 100083, China

²College of Surveying and Mapping Engineering, Henan University of Urban Construction, Pingdingshan 467044, China

Although the back propagation neural network has been successfully employed in various fields and demonstrated promising results, literatures show its performance still could be improved. Therefore, we present a comprehensive comparison study on the application of different BP algorithm in time series of deformation forecasting. Four types of typical improved BP algorithm, namely, momentum, conjugate gradient, Quasi-Newton and Levenberg-marquardt algorithms, are investigated. An illustrative example of high-rise building settlement deformation is adopted for demonstration. Results show that the improved BP algorithms can increase the prediction accuracy and have faster convergence speed.