A. Hossein Morshedy; H. Memarian
Abstract
Zoning is an important practice in earth sciences. In zonation, the study area is divided into separate parts and by compiling the results of these parts, a unique model is obtained. In this study, clustering methods are applied for zoning of Semilan dam site. Optimal number of clusters are measured ...
Read More
Zoning is an important practice in earth sciences. In zonation, the study area is divided into separate parts and by compiling the results of these parts, a unique model is obtained. In this study, clustering methods are applied for zoning of Semilan dam site. Optimal number of clusters are measured based on geotechnical parameters (lugeon, RQD), the importance of various dam structures and lithology indicators. By ranking of 7 clustering validity indexes, the optimum number of clusters found to be 4. In this paper, clustering was performed by faults locations and self-organizing neural network. In the former case, the study area was divided into four zones based on faults. This two dimensional zoning is independed of the third dimension (depth) and each sample belonged to a cluster. In the later case, a self-organizing map (SOM), which is a kind of neural network capable of clustering, was used. The SOM input data consists of, three dimensional parameters (X,Y,Z), geotechnical parameters (lugeon, RQD) and finally indicators of importance of various dam structures and lithology. Then, 7 input parameters were normalized between 0 to 1 and entered the network for training.The output data were allocated to four zones (clusters). For RQD spatial distribution realization, variography and anisotropy parameters for all four zones were calculated for both cases, Based on the main principal of clustering method which is maximum difference between clusters and maximum similarity between members of each cluster, performance and validation of two cases of clustering, RQD data were defined. Clustering quality index defined as sum of mean differences between two clusters divided by sum of standard deviation of clusters. Maximizing of this index is optimal solution. This study showed that clustering by SOM gives more accurate results than clustering by faults.
