The Application of Adaptive BP Neural Network on Numerical Prediction on Tunnel Vault Subsidence Based on the Moving Average

Tunnel vault subsidence is a very complex nonlinear dynamic system whose feature is difficult to describe accurately by traditional methods. In this paper, the moving average model of time series and the adaptive BP neural network are adopted to build the model, which use smoothing method of time series to curb and weaken the error in field surveying data, in an effort to reduce the numerical fluctuations of time series and use the adaptive learning rate and the momentum method to improve weaknesses of easily trapped into local minima, slow convergence and other shortcomings in the BP neural network. The prediction examples show that the model based on the moving average and the adaptive BP neural network can effectively restrain and weaken the measurement error and the method has features of simple, fast convergence and high accuracy prediction. Therefore, this method can be widely used.