الگويابي داده هاي لرزه‌يي با استفاده از خوشه‌بندي به منظور پيش‌بيني زلزله

Pattern Recognition of Seismic Data using a Clustering Method for Earthquake Prediction

زلزله ها همواره به عنوان يكي از مخرب ترين بلاياي طبيعي شناخته مي شوند. به دليل خسارت‌هاي اقتصادي و تلفات جاني بسيار بالا، پيش‌بيني زلزله امري ضروري به نظر مي رسد. در اين نوشتار، تغييرات زماني پارامتر b از رابطه‌ي گوتنبرگ ريشتر قبل از زلزله هايي با بزرگاي 0/6Mw = و يا بالاتر از آن در ناحيه‌ي جنوبي ايران، منطقه‌ي قشم و اطراف آن مورد بررسي قرار گرفته است. از دو روش خوشه‌بندي k-means و نقشه‌ي خود سازمان‌ده SOM، براي يافتن الگو از اين نوع زلزله ها استفاده شده است. براساس دو سنجه‌ي سيلوييت و ديويس بولدين، تعداد 3 خوشه به عنوان تعداد بهينه‌ي خوشه براي هر دو روش مذكور به‌دست آمده است. قبل از تمامي زلزله هاي مورد بررسي، خوشه‏‌يي كه معرف كاهش در مقدار b است، مشاهده شده است. به‌عنوان نتيجه‌ي نهايي، كاهش مقدار b در بازه‌ي زماني مشخص به عنوان يك الگوي مشخص براي رخداد اين زلزله هاي مخرب معرفي شده است.

Iran is known as one of the high risk seismic regions of the world. Over the past 50 years, many destructive earthquakes have occurred in this area, causing much human loss and financial damage. So, from the perspective of emergency-management and hazard preparedness, it is essential to make an effort to predict earthquake occurrence. Earthquake prediction is an instance of interdisciplinary research, which is a concern of many scientists in various fields, such as geology, seismology, engineering, mathematics, computer science and even social sciences, who study different aspects of the matter to find new solutions. Efforts in this field are divided into long-term and short-term predictions. The short-term predictions are based on precursors such as foreshock, seismic quiescence, decrease in radon concentrations and other geochemical phenomenon. Due to numerous complexities and unknown factors inside the earth, exact prediction of earthquakes is difficult and practically impossible. During the last two decades, many techniques have been developed to discover the pattern of seismic data and predict three earthquake parameters, namely; time of occurrence, location and magnitude of future earthquakes. Soft computing and data mining techniques, such as neural networks, fuzzy logic and clustering methods are appropriate tools for problems, such as earthquake prediction, that suffer from inherent complexities. Many researchers have used these approaches to reduce uncertainty in results. In this paper, the b-value of the Gutenberg Richter law has been considered as a precursor to earthquake prediction. Prior to earthquakes equal to or greater than = 6.0, temporal variation of the b-value has been examined in Qeshm island and neighboring areas in the south of Iran, from 1995 to 2012. The clustering method, by the k-means algorithm, and a self-organizing map (SOM) have been undertaken to find a pattern of variation of the b-value. Three clusters are obtained as an optimum number of clusters by the Silhouette Index and the Davis-Bouldin index. Prior to all the mentioned earthquakes ( , a cluster, known as a decrease in b-value, has been seen; so, a decrease in the b-value before main shocks has been considered as a distinctive pattern. Also, an approximate time of decrease has been determined.