خوشه‌بندي فازي روابط كاهندگي در تحليل خطر لرزه‌يي براي نواحي لرزه‌خيز ايران

Fuzzy Clustering of Attenuation Relationships for Seismic Hazard Analysis in Iran

با توجه به تاثير چشم‌گير روابط كاهندگي در ارزيابي خطرپذيري لرزه‌‌يي، سازمان‌دهي روابط كاهندگي به گروه‌هاي مشابه مي‌تواند به دريافت اطلاعات ضمني ناشناخته در ميان روابط مختلف منجر شود. به همين منظور در اين مطالعه رويكرد خوشه‌بندي فازي به‌منزله‌ي روشي كارآمد در زمينه‌ي گروه‌بندي طيف وسيعي از داده‌ها مورد توجه قرار گرفته است. با انتخاب روش خوشه‌بندي فازي C-mean و اصلاح الگوريتم براساس ويژگي‌هاي روابط كاهندگي، مجموعه‌‌يي شامل 45 رابطه‌ي منتخب در سراسر جهان به گروه‌هاي مشابه افراز شده‌ است. با دراختيارداشتن چنين طبقه‌بندي و با لحاظ‌كردن شرايط خاص منطقه‌ي مورد مطالعه، مي‌توان به مناسب‌ترين مجموعه‌ي روابط در منطقه‌ي موردنظر دست ‌يافت. اين امر در مورد فلات ايران از طريق انطباق مجموعه‌ي داده‌هاي لرزه‌يي اين منطقه بر زيرمجموعه‌هاي حاصل از خوشه‌بندي مورد بررسي قرار گرفته و بر اين‌ اساس روابط تاكاهاشي (2004) در رخدادهاي پوسته‌‌يي،[1] ژايو (2006) در رخدادهاي پوسته‌‌يي و داخل‌ صفحه‌يي،[2] و زعفراني (2008 و2012) [3و4] با مناسب‌ترين انطباق معرفي شده‌اند.

Attenuation relationship models are key elements within any seismic hazard analysis (SHA). As the results of SHA can be significantly changed by using different attenuation models, it is necessary to select an appropriate attenuation model for SHA. This selection is usually done without enough attention paid to the compatibility of the chosen attenuation model with the historical site database. That is, the selection of an appropriate attenuation model for regions that suffer from lack of their availability is usually a serious challenge. Therefore, the fuzzy C-Mean clustering approach is employed and modified in this study in order to classify available well-known attenuation models into different clusters. A set of 45 worldwide attenuation models were selected in this research for the purpose of classification. It worth noting, that the input definition for all the chosen models should be consistent, and so, a transformation procedure was needed. They were classified into 14 different clusters, in which, each cluster was defined to have the most dissimilarity with the other clusters, and to have the most similarity within the cluster. The 45 models were then classified into a different numbers of clusters and the corresponding border-models were eliminated. This clustering was iterated until a reasonable clustering was achieved. The final clustering consists of 34 attenuation models put into 14 different clusters. The resulted clusters were, then, tested for the Iranian ground motion database. The database consists of 28 earthquake events with 78 accelerograms. The root mean of the sum of the squares (RMSE) and the mean absolute error (MAE) indicators were employed, in order to compare the different clusters for the given database. The results show that the best fitted models consist of Takahashi et al. 2004, Zhao et al. 2006, for the crustal and inter-slab zones, and Zafarani et al. 2012. On the other hand, the worst attenuation models, which did not adequately fit the Iranian database, included Khademi et al. 2002, Gulkan and Kalkan et al. 2002 and Ghodrati et al. 2010.