The Application Of BP Neural Networks In Deformation Prediction Based On Wavelet Denoising Theroy

Nowadays the development of engineering construction is closely related to our country economy and the speeding up of urbanization, people put forward higher requirements to the size, shape and design of buildings. Under this background, the meaning of the deformation monitoring work becomes more significant. As is known to all, during the construction of engineering buildings, deformation usually appears due to the influence of various subjective or objective factors. If the deformation value is beyond the prescribed limit, it will affect the normal use of buildings. Even more, the security of people’s lives and property will suffer severe threat.The analysis and predication of engineer deformation monitoring, which bases on deformation monitoring and involves multi-disciplines such as systems theory and nonlinear science is a fast development technique. With the continuous updating of monitoring instruments, variability of monitoring method, and diversification of monitoring contents,it has become one of the most important issues in the researching of deformation monitoring to deeply analyze the nonlinear and complexity of engineer deformation,explore the extraction of deformation trend information, and forecast and determine the stability of the deformed body.Based on the complexity of deformation of deformed body, the scholars and experts mainly suggest various forecasting methods which are used to get the deformation values both theoretically and practically. In practical application, the method of measured data analysis has been used widely. For the time being, the models established in domestic are of different features and advantages but also have respective limitations. Research shows that the prediction effect of preprocessing data model is superior to the single nonlinear method. At this point, subsidence prediction model is established by using wavelet denoising data based on BP neural in this paper.The paper introduces the basic principle and key process of wavelet denoising and BP neural network. It discusses and analyses the parameter of wavelet denoising and BP neural network model based on the data from several actual projects which are referred in this paper. It also optimizes properties of traditional BP algorithm, such as slower convergence speed and falling easily into local optimal solution by improving the algorithms of additional momentum method and Levenberg-Marquartdt method. Predictions show that using data after wavelet denosing and optimizing algorithm will improves the predict accuracy and falling speed of BP neural network.