基于网络基坑沉降预测的马尔可夫修正模型

分析灰色GM(1,1)预测模型存在的理论缺陷,指出灰色GM(1,1)预测模型虽可用于小样本基础数据预测,但对基础沉降一类随机性强、波动性较大的数据拟合质量较差,预测精度降低。因而,提出利用马尔可夫链修正神经网络模型,其计算过程为:首先建立神经网络动态拟合模型作为基础沉降变化的基准线,在此基础上应用马尔可夫链确定系统状态转移概率矩阵,最后通过系统状态划分样本值与模型拟合值之间的残差及中误差等指标分析计算,最终完成基础沉降的准确计算,该模型应用于基础沉降工程实例运算,取得较好效果。

Neural Network Based Markov Chain for Foundation Settlement Forecast

This article analyzes the theoretical defect of gray predication formula.Gray forecast model is able to establish models for less data based forecast.But the foundation settlement data which are great randomness and fluctuation will make the data fitting difficult and reduce the forecast precision.For the first time,Markov chain is used to improve neural network on foundation settlement forecast to get a neural network the dynamic baseline for foundation settlement development.Secondly Markov chain is applied to achieve state transition probability matrix.At last the foundation settlement interval is forecasted and analyzed in the form of probability by the system state classification,the calculation of the residue between true value and model fitting value.This model is applied to analyze the foundation settlement data.The result shows that it is practicable to predict the settlement deformation through this method.