Dissertation > Industrial Technology > Automation technology,computer technology > Automated basic theory > Artificial intelligence theory > Artificial Neural Networks and Computing

Research on the Building Foundation Settlement Prediction by Wavelet Neural Network

Author YangWenTuo
Tutor YangFan
School Liaoning Technical University
Course Geodesy and Survey Engineering
Keywords Wavelet neural network Settlement monitoring of Buildings Wavelet packet threshold de-noising the Setting of Initial Right Value
Type Master's thesis
Year 2011
Downloads 75
Quotes 0
Download Dissertation

With the rapid development of the national economy, the amount of tall buildings is increasing. On one hand, tall buildingscan alleviate the inadequacy of the city land resources. On the other hand, if there is a serious violation of the construction in the period of operation, it will cause the infrastructure and superstructures of buildings imbalanced, and unsafe deposition will be occured to the buildings. And it will endanger people’s lives and property. So deformation monitoring is of great significance. Deformation monitoring is a kind of means to get credible datas. Its real purpose is how to use the data to forecast for decision-making. Processing the datas effectively and establishing an accurate model is the bridge to contact the two aspects. Aiming at dealing with the problems of random errors exist in the deformation datas and the problems of slow convergence peed and low precision exists in the traditional wavelet nerual network, taking the deformation datas for the object, this paper studies on the application of de-noising process of the deformation data and how to establish the model of WNN, and improve the way of selecting the network initial weights and learning algorithms, according to the theory of wavelet packet threshold de-noising algorithm and WNN. The main work as follows:(1) For the problems that these is noise in the collected deformation monitoring data, this paper introduces the algorithm of wavelet packet threshold denoising to de-noise the data. After the pretreatment, the noise ratio of signal was greatly improved, and variance was significantly reduced. It provides a more accurate sample for the follow-up forecasting.(2) Combining the advantages of wavelet packets and neural networks, this paper studies on building the wavelet neural networks and improving it. The accuracy and absolute error of improved wavelet neural network has absolutely improved.(3) In order to improve the low convergence speed and accuracy exists in the traditional BP neural network, this paper uses a method which considers wavelet type of initial parameters, wavelet time-frequency parameters and samples to select the initial parameters of the wavelet neural network.(4) For the problems exist in the traditional model, such as Easy to fall into local minimum, low accuracy, slow convergence speed, In this paper, neural network learning algorithm is proposed to improve. According to introducing a factorsγk to balance the tion of the output layer and hidden layer, and make sure the predictive value approaches the measured in a stable way.

Related Dissertations
More Dissertations