應用類神經網路由現場試驗結果預測與分析地工設計參數

Predicting and Analyzing Geotechnical Design Parameters from In-Situ Tests Using Artificial Neural Networks

盧炳志(Ping-Chih Lu) ； 倪勝火(Sheng-Huoo Ni)

《地工技術》 91期 (2002/06) Pp. 47-54

https://doi.org/10.30140/SG.200206.0005

摘要

對大地工程師而言，從現場試驗資料求取或預測工程所需設計參數是一件非常重要且具挑戰性的工作。本文討論應用類神經網路模式由現場試驗資料預測地工設計參數的流程與方法。除介紹倒傳遞網路的架構建立及其訓練方法之外，本文以一個利用電子錐貫入試驗（CPT）資料建立用來預測砂土相對密度Dr值的實際網路，探討以類神經網路模式預測地工設計參數的成效及其與統計迴歸方法之比較。此外，本文亦討論應用於分析現地試驗資料所具空間變異性之類神經網路，並以電子貫入錐之錐頭阻抗為例，比較不同型態類神經網路在分析空間變異性特性上之適用性。

關鍵字

電子錐貫入試驗 ； 相對密度 ； 倒傳遞網路 ； 半徑式函數網路 ； 通用迴歸型網路

並列摘要

Predicting and analyzing engineering parameters from in-situ tests is an important and challenging task for geotechnical engineers. In this study, an artificial neural network approach is proposed to predict these design parameters. In this paper, a brief introduction of back-propagation networks is provided and then a network for predicting relative density Dr from CPT measurements is established. Discussions on the established network are presented, along with comparisons of the results by this network with existing methods. Meanwhile, the application of artificial neural network approach on the spatial variability characteristic of in-situ tests is also discussed in this study. Some details of the development of various network models for analyzing CPT measurements and the comparisons of the network predictions are presented.

並列關鍵字

CPT ； relative density ； back-propagation network ； radial basis function network ； generalized regression neural network