Artificial Neural Network Analysis for the Evaluation of Slope Stability
The knowledge data-base which consist of 80 practical case problems for predicting slope stability is developed based on the Statistical Artificial Neural Network method. The forecasting results for the slope stability related to permanent shiplock rock s
- PDF / 965,256 Bytes
- 8 Pages / 482 x 692 pts Page_size
- 39 Downloads / 266 Views
W.Xu Unniversity of Hydraulic and Electric Engineering, Yichang, Hubei, China J.-F. Shao Universite des Sciences et Technologies de LiUe, Villeneuve d' Ascq, France
ABSTRACT The knowledge data-base which consist of 80 practical case problems for predicting slope stability is developed based on the Statistical Artificial Neural Network method. The forecasting results for the slope stability related to permanent shiplock rock slope engineering in Three Gorges Project have illustracted in details. It is shown that application of Statistical ANNs methods for the prediction of stability of slope engineering is realiable and practical. KEYWORDS slope engineering; artificial neural network; stability; Three Gorges Project
1 Statistical Artificial Neural Networks and its Graphics
Artificial Neural Networks(ANN), neurocomputing or brainlike computation is based on the wishful hope taht we can reproduce at least some of flexibility and power of human brain by artificial means. Artificial Neural Networks consist of many simple computing element-generally simple nonlinear summing junctions - connected together by connections of varying strength, a gross abstraction of brain, which consist of very large numbers of far m6re complex neurons connected together with far more complex and far more structured couplings.
A. Cividini (ed.), Application of Numerical Methods to Geotechnical Problems © Springer-Verlag Wien 1998
666
W. Xu and J.-F. Shao
The statistical artificial neural networks for prediction some geotechnical problem is developed by authors these years. As a self-organized forwardfeed artificial neural networks it is based on the statistical priciple familiar to self-study nector networks. The major advance of this networks is to solve problem rapidly and to simplify large problems. The graphics of the statistical artificial neural networks is shown as Figure I.
Hidde:1 layer l
Predictioa
Fig. I Graphics of Statistical Artificial Neural Networks
Notation in Figurel have the following meaning: p: prediction vector, m: model vector, i: indicates the neuron, belonging to the input variable, o: indicates the neuron, belonging to the output variable, N: number of model vectors. M: number of input varieable of the phenomenon. Different statistical methods demand the.selection of shape of the function which best suits the discriptiort of the geotechnical problem. The coefficients of empirical/regression equations are then determined with the least squares method. In such a way we try to describe the prblem with some · advanced presumed empirical law. While the data are usually incomplete the selected law is fitted very good to the available data and usually fails when more data is acquired. The available database are not representative in most of the practical cases and therefore the automatic modeling of the phenomenon is very appropriate when new data are obtained. Compared to the parametric methods the algoritmic model prepared by our software does not need any advanced presumed law. The training phase here is ver
Data Loading...
