ارزيابي حساسيت زمين‌لغزش با استفاده از روش‌هاي شبكه عصبي مصنوعي و الگوريتم ماشين بردار پشتيبان (مطالعه موردي: جاده‌ي حيران- استان اردبيل)

Assessment of landslide sensitivity by neural network method and Vector machine algorithm (Case study: Heyran Road -Ardebil province(

زمين ­لغزش­ها همواره موجب خسارات جاني و مالي، از دست رفتن منابع طبيعي و زيرساخت‌هاي زيربنايي از قبيل جاده­ها، پل‌ها و خطوط ارتباطي مي­شوند. جاده ارتباطي حيران در حال حاضر تحت تأثير فرايند لغزش و گسيختگي دامنه‌اي دستخوش تغيير مي‌باشد. در اين پژوهش حساسيت زمين‌لغزش محور ارتباطي حيران با استفاده مدل‌هاي شبكه عصبي مصنوعي و توابع خطي، چندجمله‌اي، شعاعي و حلقوي الگوريتم ماشين بردار پشتيبان موردبررسي قرار گرفت. معيارهاي مؤثر در شناسايي حساسيت زمين‌لغزش در سطح منطقه موردمطالعه شامل لايه‌هاي استخراج‌شده از سطوح ارتفاعي، زمين‌شناسي، كاربري اراضي، فاصله از گسل، شيب، جهت شيب و فاصله از جاده مي‌باشد. لايه‌هاي اطلاعاتي بعد از آماده‌سازي در محيط نرم‌افزار SPSS Modeler اجرا شد و نقش و ارزش هركدام از پارامترها بر اساس روش‌هاي مختلف به دست آمد. بر اساس نتايج ارزيابي مدل به ترتيب عامل زمين‌شناسي، ارتفاع، جهت شيب و كاربري اراضي، بيشترين ارزش را در ناپايداري دامنه‌ها در اين محدوده داشته‌اند. همچنين نتايج نشان داد، كاربري‌هاي كه در طبقه حساسيت زياد قرارگرفته‌اند عمدتاً مربوط به اراضي مرتع، زمين كشاورزي و جاده‌هاي ارتباطي ( بالاتر از 1400 متر) مي‌باشند كه در قسمت‌هاي غربي گردنه حيران واقع‌شده‌اند. سازندهاي تحت تأثير لغزش در محدوده موردمطالعه عمدتاً تناوب توف، ماسه‌سنگ توفي به همراه گدازه‌هاي برشي و گدازه‌هاي پيروكسن آندزيت مي‌باشد. در قالب مقايسه بين مدل‌ها جهت ارزيابي مطابقت آن با واقعيت منطقه به نظر مي‌رسد مدل ماشين بردار پشتيبان نسبت به شبكه عصبي مصنوعي كارايي بهتري جهت ارزيابي حساسيت زمين‌لغزش در محور ارتباطي گردنه حيران دارد. شايان‌ذكر است، لزوم توجه به نقشه‌هاي حساسيت زمين‌لغزش طي عمليات جاده‌سازي و تعريض آن مي‌تواند سبب كاهش ريسك مخاطره‌ي زمين‌لغزش در محدوده مسير جاده حيران-آستارا گردد.

The occurrence of landslides is the result of the interaction of complex and diverse environmental factors. These factors are divided into the trigger and the primary cause. Landslide occurrence triggers include weathering, earthquakes, rainfall and snow melting. Human activity like construction of roads and buildings on steep slopes and dispersal of water from supply systems and sewers could also trigger the occurrence of the phenomena (Cubito et al., 2005). In this Investigation will be using data layers Region, Identifying of the most important factors landslide prone areas in Heyran road and in this Investigation effectiveness of methods used to be examined. Introduction The occurrence of landslides is the result of the interaction of complex and diverse environmental factors. These factors are divided into the trigger and the primary cause. Landslide occurrence triggers include weathering, earthquakes, rainfall and snow melting. Human activity like construction of roads and buildings on steep slopes and dispersal of water from supply systems and sewers could also trigger the occurrence of the phenomena (Cubito et al., 2005). In this Investigation will be using data layers Region, Identifying of the most important factors landslide prone areas in Heyran road and in this Investigation effectiveness of methods used to be examined. Materials & Methods In this study landslide sensitivity preparation mapping, position had landslide occurred in the study area bye GPS it got recorded and using geology Astara map Layer was extracted fault. Besides these layers used were layers of elevation, slope and aspect of the DEM with a resolution of 30 m. To do this research using the data above according to SVM algorithm landslide susceptibility maps sensitive communication axis was determined Heyran. SVM algorithm is based on statistical learning theory. According to this theory can be bound for to data error rate unclassified machine learning, to be considered as a generalized error rate. In this study, a better estimate of landslide susceptibility neural network method was used. After preparation of the layer of elevation, slope, aspect, geology, faults, land use and other factors data inputted in software SPSS Modeler. Software output for each one of the factors was just small amounts in continue investigation process landslide susceptibility map used software GIS was prepared. Results & Discussion In the zoning of landslide sensitivity, the most important part of the work is the preparation of a layer of dispersion of landslides in the region. The accuracy and precision of zoning is main due to this part of the work. In order to assess the accuracy and correctness of the zoning of field works, the study of landslides is an integral part and field works were carried out to identify landslides in the region. GPS is most important tool at this stage. A total of 42 cases of major landslides were recorded during Heyran. In this research, the Heiran road was investigated using neural network methods and vector machine algorithm. The layers used is elevation, slope, aspect, geological formations, faults, land use. Most affected by slope instability, mainly related to land use pasture, farmland and roads linking within the elevation is 1,400 meters high. In the overall evaluation of the performance of the models, sigmoid kernel model have better performance due to the layers used and the conditions of the axis of communication. So the results of these two models can be the basis of zoning. Conclusion Slope instability as one of the most important geomorphological hazards in some areas has made significant and has created serious problems for residents. This research has been carried out to identify areas of potential Landslide in the Heyran-Astara. This communication road is very important for landslide occurrence. In addition to road hazards, there are multiple slides along the road. It was necessary to study and compare accurate zoning methods for proper evaluation in this range. In this study, neural network models and four models vector machine algorithm has been evaluated and compared. According to the results of the sigmoid kernel and sigmoid kernel models, more than 80 percent of the 42 landslides are recorded in a large and very high class. In the overall assessment of the performance of the models, the sigmoid kernel and Neural network models are more consistent with the layers used and the conditions of the Heyran-Astara communication road and the registered position of the slides. So the results of these two models can be the basis of zoning. Keywords: Landslide susceptibility, artificial neural network, support-vector machines, Heyran Road. Keywords: Landslide susceptibility, artificial neural network, support-vector machines, Heyran Road. كليدواژه‌ها [English]