基于粗集理论算法-脊波神经网络的深基坑变形预测与应用

基坑开挖变形具有非线性特性,在脊波神经网络的基础上,采用粗集理论算法优化初始权值和阈值,建立了基于粗集理论算法-脊波神经网络的深基坑变形预测模型,应用该模型对西南地区某市火车站综合交通换乘中心南广场的基坑开挖过程进行了变形预测。结果表明:粗集理论算法能够对脊波神经网络进行优化,提高了脊波神经网络基坑变形预测结果的收敛速度和泛化能力;脊波神经网络能逼近基坑变形的非线性部分,避免了模型误差影响基坑开挖变形预测精度,提高了系统整体抗干扰性能。模型的预测值与实测值之间的误差在5%以内,满足实际工程的要求。

Deformation forecast of the deep foundation pit based on the rough set theory and the algorithm ridgelet neural network

The present paper is aimed to propose a model that can meet the demands for increasing the prediction or forecast accuracy to reflect the characteristic features of the deformation and training flexibility in the engineering. practice by setting the network structure,initializing the weights and entering the data measured in the early stage of the deep foundation pit. The neural network used to be faced with such problems,as the local minimum value and slow proficiency when it has to forecast and predict such deep foundation pit deformation. In such a situation,it would be necessary to optimize the initial weights and the thresholds for the prediction model of the deep foundation pit deformation to be applied based on the ridgelet neural network with the rough set theory and algorithm to be used. And,for the purpose of practical application,it is necessary to work out the algorithm error before the application is done with the error corrected so as to meet the demands. And,for the said purpose,enough training samples have to be prepared and a deep foundation pit prediction model to be generated to be able to predict the impending deformation based on the measured data. And,such a kind of model has to be made up of 8 input items and 1 output item,with the hidden layers containing 11 nodes and 20 subgroups. To reduce the cost of the prediction,it would be necessary to predict the horizontal displacement and the settlement displacement. Furthermore,it is also necessary to do the deformation prediction analysis of the foundation pit excavation process of the railway station in the southwestern region of the mine to validate the effectiveness of the model. The results of our examination and forecast show that it would be possible to optimize the ridge neural network and improve the convergence speed and generalization ability of the forecast results the rough set theory and the algorithm can help to produce. On the other hand,the ridge neural network can deal with the nonlinear part by using the foundation pit deformation theory and avoid the model error,unless they may influence the accuracy of the foundation pit excavation deformation by improving the anti-interference of the system.Thus,comparing the predicted and measured data, the root mean square error of the horizontal displacement prediction can be reduced by 2. 96%,whereas the average percentage error can be reduced by 4. 13%,with the root mean square error being left over 4. 98%. Besides,the error with the surface settlement prediction and the error with the average percentage can also be reduced by 2. 97%. Thus,in so doing,the model can be made to meet the demands of the practical engineering project so as to turn to be an effective tool for the deformation prediction analysis of the deep foundation pits.