基于NAR邻近基坑开挖建筑位移预测模型

Prediction Model of Building Deformation in Adjacent Foundation Pit Excavation Based on NAR

准确预测邻近建筑位移变化,设置预警机制进行警度判断是保证建筑安全的关键。现有的预测方法因缺乏数据和参数获取难而影响预测结果的可靠性。针对此问题,借助NAR神经网络进行邻近基坑建筑位移预测模型研究,建立地表-建筑位移预测模型。经过多轮超参优化,确定最优超参及预测范围。建立双重保障预警流程,依据预测结果进行警度判断,并与仅以建筑历史时刻的竖向位移为输入值的预测效果进行比较分析。结果表明,NAR网络模型对基坑开挖邻近建筑变形具有良好预测性能,在建筑变形发生趋势变化时,地表-位移法展现更佳拟合效果,预测误差值小于0.5 mm。NAR神经网络能够较好解决具有动态特性和时序关系的非线性问题,对于解决基坑开挖邻近建筑变形预测问题具有高精度和低误差的优势。

Accurate prediction of displacement changes of adjacent buildings and setting early warning mechanisms are the key to ensuring building safety.The reliability of prediction results is affected by the lack of data and the difficulty of obtaining parameters.In response to this problem,the NAR neural network is used to study the displacement prediction model of adjacent foundation pits,and the surface-building displacement prediction model is established.After several rounds of parameter optimization,the optimal parameter and prediction range is determined.Establish a dual security early warning process,and judge the warning degree according to the prediction results.The prediction effect is compared with that only using the vertical displacement at the historical moment of the building as the input value.The results show that the NAR network model has a good prediction performance for the deformation of adjacent buildings during foundation pit excavation.When the trend of building deformation changes,the surface-displacement method shows a better fitting effect,and the prediction error is less than 0.5 mm.NAR neural network can better solve the nonlinear problems with dynamic characteristics and time sequence relationship,and has the advantages of high precision and low error in solving the deformation prediction problem of adjacent buildings in foundation pit excavation.