1.School of Safety Science and Engineering, Xi’an University of Science and Technology, Xi’an , China;2.Shaanxi Provincial Cuijiagou Coal Mine,Tongchuan , China;3.School of Science, Xi’an University of Science and Technology, Xi’an , China

BP神经网络可以解决地表沉陷等非线性关系问题，为了更精确地进行地表沉陷变形预测，引入Adaboost算法对BP神经网络进行改进，并运用Matlab R2014a建立基于Adaboost的BP神经网络地表沉陷预测模型。首先通过BP神经网络进行训练、测试，经过多次迭代，将每个BP神经网络作为一个弱预测器加权组合，形成强预测器，并首次对青岛地铁3号线保河区间隧道进行地表下沉值预测。预测结果表明：Adaboost的BP神经网络预测下沉值的平均绝对误差为0.585 3 mm，平均相对误差为5.82%，与BP神经网络预测相比，绝对误差降低了2.594 7 mm，相对误差降低了27.46%，由此表明Adaboost的BP神经网络适用于地表沉陷预测，且预测精度更高。

A BP neural network could solve a non-linear relationship problem of the surface subsidence. In order to more accurately predict the surface subsidence deformation, an Adaboost algorithm was introduced to improve the BP neural network. A Matlab R2014a was applied to establish a surface subsidence prediction model with BP neural network based on the Adaboost. Firstly, a training and test were conducted with the BP neural network. After several iterations, each BP neural network as a weak predictor would be weighted and combined and then would form a strong predictor. A surface subsidence prediction was conducted in Baohe section tunnel of Qingdao Metro No.3 Line. The prediction results showed that the predicted average absolute error predicted with the BP neural network of the Adaboost was 0.585 3 mm, the average relative error was 5.82 %. In a comparison with the prediction of BP neural network, the absolute error was reduced by 2.594 7 mm and the relative error was reduced by 27.46 %. Therefore the BP neural network with the Adaboost could be suitably applied to the prediction of the surface subsidence prediction and the prediction accuracy would be higher.