东南大学学报(自然科学版)
東南大學學報(自然科學版)
동남대학학보(자연과학판)
JOURNAL OF SOUTHEAST UNIVERSITY
2013年
z2期
291-295
,共5页
模型误差%误差补偿%残差%神经网络%路基沉降预测
模型誤差%誤差補償%殘差%神經網絡%路基沉降預測
모형오차%오차보상%잔차%신경망락%로기침강예측
model error%error compensation%residual error%neural network%prediction of roadbed settlement
为了解决由路基沉降观测数据建立的沉降预测模型存在的模型误差,用基于残差的神经网络方法对预测模型进行补偿。由预测模型计算模型残差,借助神经网络根据残差对预测模型进行补偿,将预测模型与补偿结果叠加获得补偿后的实用模型。对同一地质情况不同软基处理方式进行工程实例验证,结果表明:基于残差的神经网络方法能有效补偿模型误差,神经网络方法补偿后的路基沉降预测模型,其预测精度平均提高了56%,优于原预测模型。
為瞭解決由路基沉降觀測數據建立的沉降預測模型存在的模型誤差,用基于殘差的神經網絡方法對預測模型進行補償。由預測模型計算模型殘差,藉助神經網絡根據殘差對預測模型進行補償,將預測模型與補償結果疊加穫得補償後的實用模型。對同一地質情況不同軟基處理方式進行工程實例驗證,結果錶明:基于殘差的神經網絡方法能有效補償模型誤差,神經網絡方法補償後的路基沉降預測模型,其預測精度平均提高瞭56%,優于原預測模型。
위료해결유로기침강관측수거건립적침강예측모형존재적모형오차,용기우잔차적신경망락방법대예측모형진행보상。유예측모형계산모형잔차,차조신경망락근거잔차대예측모형진행보상,장예측모형여보상결과첩가획득보상후적실용모형。대동일지질정황불동연기처리방식진행공정실례험증,결과표명:기우잔차적신경망락방법능유효보상모형오차,신경망락방법보상후적로기침강예측모형,기예측정도평균제고료56%,우우원예측모형。
Due to the fact that the settlement prediction model established with the data of settlement observation has model error, the neural network method based on residual is adopted to compensate for model error.The residual is calculated through the prediction model.On the basis of residual er-rors, the compensation to the prediction model is implemented by the neural network, and the im-proved practical model which consists of the two parts is obtained.Some engineering examples on different methods of soft foundation treatment with the same condition of geology are studied by the practical model.The results show that the practical model based on the neural network can compen-sate for model errors effectively.The prediction model of the roadbed settlement compensated for by the neural network can improve the forecast accuracy by 56% in average, which is superior to the prediction model.