施工技术
施工技術
시공기술
CONSTRUCTION TECHNOLOGY
2015年
7期
87-89
,共3页
地下工程%深基坑%变形%预测%BP-ANN模型
地下工程%深基坑%變形%預測%BP-ANN模型
지하공정%심기갱%변형%예측%BP-ANN모형
underground%deep foundation excavation%deformation%prediction%BP-ANN model
利用LM算法和自适应学习速率的动量梯度下降法对BP?ANN模型进行改进，发现LM算法在建立深基坑变形预报模型时具有更高的学习效率，且输入量间的相关程度对网络效率无显著影响，隐含层节点数可通过仿真误差分布图确定和优化。仿真与预测结果表明，在平稳观测时间序列内该模型具有更高的预测精度，在实际深基坑监测预报中值得推广应用。
利用LM算法和自適應學習速率的動量梯度下降法對BP?ANN模型進行改進，髮現LM算法在建立深基坑變形預報模型時具有更高的學習效率，且輸入量間的相關程度對網絡效率無顯著影響，隱含層節點數可通過倣真誤差分佈圖確定和優化。倣真與預測結果錶明，在平穩觀測時間序列內該模型具有更高的預測精度，在實際深基坑鑑測預報中值得推廣應用。
이용LM산법화자괄응학습속솔적동량제도하강법대BP?ANN모형진행개진，발현LM산법재건립심기갱변형예보모형시구유경고적학습효솔，차수입량간적상관정도대망락효솔무현저영향，은함층절점수가통과방진오차분포도학정화우화。방진여예측결과표명，재평은관측시간서렬내해모형구유경고적예측정도，재실제심기갱감측예보중치득추엄응용。
This paper improves the BP?ANN model using LM arithmetic and adaptive leaning rate gradient descent method, and discovered that the former has higher efficiency in establishing deep excavation deformation prediction model. In addition, the degree of correlation of the input has no obviously impact on the net ’ s efficiency, and the hidden layer ’ s number can be established and optimized through simulation error distribution diagram. The result of simulation and predication indicated that the model has high prediction accuracy in smooth and steady observation time series, and it is worthy of popularization in the practice of deep foundation excavation deformation prediction and analysis.
