The Research And Application Of Settlement Prediction Methods In Engineering

At present, the time series settlement prediction methods are very extensive, mainly divided into two big kinds, one is the theory method, because of too many influence factors, the complex physical and engineering property, it is hard to predict the settlement using these methods. The other is building mathematic model based on real measured data. The model, which is timely integrated the measured data and could reflect the law of the systems, is widely applied. The main research results are as follows:The initial value condition and input dimension of Gray Model are calculated and analyzed, and then the best initial value condition and input dimension are to be used in prediction.BP artificial neural networks select the optimal transfer function and learning function.The Multi-layer Recursive Time Series Settlement Prediction Model is based on time series model and multi-layer recursive theory; the time-dependant parameters in the new model are predicated in the second layer, and the parameters identification is studied.The case shows that the prediction models developed in this paper every method has their own advantages, but The Multi-layer Recursive Time Series Settlement Prediction Model has higher precision and better applicability.