كليدواژه :
زمين‌لغزش , شبكه عصبي مصنوعي , حوضه سيمره چنار , سيستم اطلاعات جغرافيايي , مدل منطق فازي
چكيده فارسي :
زمين‌لغزش به عنوان يكي از مخاطرات طبيعي در مناطق كوهستاني هر ساله منجر به خسارات زيادي مي‌شود. حوضه آبريز سيمره چنار، با داشتن ويژگي هاي كوهستاني و شرايط طبيعي مختلف داراي استعداد بالقوه زمين‌لغزش است. هدف از اين پژوهش، مقايسه مدل شبكه عصبي مصنوعي با مدل منطق فازي، جهت ارزيابي خطر زمين لغزش در حوضه سيمره چنار است. بدين جهت ابتدا پارامترهاي مؤثر در وقوع زمين‌لغزش استخراج و سپس لايه‌هاي مربوطه تهيه شده است. سپس نقشه پراكنش زمين‌لغزش‌هاي رخ‌داده شده حوضه تهيه و با تلفيق نقشه عوامل مؤثر بر لغزش با نقشه پراكنش زمين‌لغزش‌ها، تأثير هر يك از عوامل شيب، جهت شيب، سنگ‌شناسي، بارش، فاصله از گسل، كاربري اراضي، خاك، فاصله از آبراهه در محيط نرم‌افزار ArcGIS محاسبه گرديد. در اين مطالعه به منظور مقايسه مدل‌ها، در پهنه‌بندي خطر زمين‌لغزش حوضه سيمره چنار، از مدل‌هاي شبكه عصبي مصنوعي و منطق فازي استفاده گرديد. در مدل شبكه عصبي مصنوعي الگوريتم پس انتشار خطا و تابع فعال‌سازي سيگموئيد بكار گرفته شد. ساختار نهايي شبكه داراي 8 نرون در لايه ورودي، 14 نرون در لايه پنهان و 1 نرون در لايه خروجي گرديد. پس از بهينه شدن ساختمان شبكه، كل اطلاعات منطقه در اختيار شبكه قرار گرفت و در نهايت با توجه به وزن خروجي، نقشه پهنه‌بندي زمين‌لغزش تهيه شد. در مدل منطق فازي از اپراتورهاي عملگراجتماع فازي، عملگراشتراك فازي، عملگرضرب جبري فازي، عملگرجمع جبري فازي، عملگرگاما فازي مدل منطق فازي استفاده شد. براي ارزيابي نتايج خروجي مدل‌هاي مورد استفاده در برآورد خطر لغزش منطقه از ضريب آماري كاپا استفاده شد. نتايج بدست آمده نشان مي‌دهد كه مدل شبكه عصبي مصنوعي با ضريب كاپاي 91/0 مدل كارآمدتري نسبت به مدل منطق فازي در تهيه نقشه خطر لغزش‌هاي حوضه سيمره چنار است. از ميان عوامل تاثيرگذار بر زمين لغزش در منطقه مورد مطالعه عامل شيب به عنوان مهمترين عامل و عوامل سنگ‌شناسي و خاك در مراتب بعدي قرار گرفتند. بر اساس پهنه‌بندي صورت گرفته با استفاده از مدل شبكه عصبي مصنوعي، به ترتيب 12/10، 92/22، 04/31، 76/20، 16/15 درصد از مساحت منطقه در كلاس‌هاي خطر خيلي‌كم، كم، متوسط، زياد و خيلي‌زياد قرار گرفته است.
چكيده لاتين :
Introduction Landslide is considered as one of the natural hazards ever occurring throughout the world and is of great importance. This phenomenon is one of the major geomorphic processes affecting the evolutionary landscape in mountainous regions, which has caused catastrophic accidents. Due to the special climate conditions, physiography and change of the country, Iran has always faced with the problem of mass movements, and it is necessary to pay attention to this natural limitation. Lorestan province also has diverse geological features such as petrography, land management, seismicity and special climate conditions, including areas with slip potential. The topographic and geological conditions of the study area are such that the slip of rock and soil fragments from small to large scale has been provided.
Methodology
Fuzzy logic is a logic of several values, that is, its parameters and variables, in addition to the number of 0 or 1, can take all the values between these two numbers. The basis of the difference between fuzzy methods and other methods is to define the membership function. The membership function can be used to determine the degree of attribution of the elements of the reference set to its subset. The operator of the fuzzy society is the collection community.In this way, it extracts the maximum membership membership. The fuzzy subscription operator is the collection subscription. In that way, it extracts the minimum degree of membership. Fuzzy algebraic multiplication multiplies all the information layers together. Because of the nature of the numbers between zero and one, which is the same as the membership membership in a fuzzy set, the operator makes the number of the numbers smaller and goes down to zero. The complementary fuzzy algebra operator is obtained by the algebraic multiplication. Therefore, in the outbound map, unlike the fuzzy algebraic operator, the value of pixels goes toward one. The fuzzy gamma operator is the product of multiplication, fuzzy coherent multiplication in the fuzzy algebraic summation. The results obtained from this operator are more accurate than other operators.An artificial neural network is a computational mechanism that can provide a series of new information by gathering information and calculating them. In the artificial neural network, the structure of the human brain and the body's neural network is similar to that of a brain that has the power to learn, make and decide. In the neural network model for preparing the network from the layers, along with a number of real samples, they entered the network as inputs, and with this method a pattern was obtained between the input parameters and the areas where the landslide was located. A probability ratio was used to determine the land slide sensitivity index. In order to facilitate artificial neural network convergence, the values of the input neurons were normalized. To estimate the accuracy of artificial neural network, the mean squared error error was used.
Results and discussion
An artificial neural network with multilayer perceptron structure with error propagation algorithm and non-linear sigmoid function as an activation function was used. The simple learning factor was ignored because of convergence and failure to make a valid error. Also, the error rate of the variable learning coefficient was higher than the Levenberg-Marquard method, which is why the Levenberg-Marquard method was used. For training and network testing, 80% of the data was used for training and 20% for testing. The final structure of the 1-14-8 grid was considered appropriate and based on this structure, the final zoning was performed.The operators of the fuzzy logic model were used. The result of the fuzzy community operator generated the maximum membership membership membership. The fuzzy share operator extracts the minimum membership membership. The result of the operator of the fuzzy algebraic multiplication is reduced to zero numbers. The output map of the operator of the fuzzy algebra sum of the value of the pixels is close to the maximum. In order to modulate the very high sensitivity of the fuzzy algebraic operator and the very low accuracy of the fuzzy algebraic operator, a 0.9% gamma-gamma operator was used. Kappa coefficient for artificial neural network model was 0.83 and for fuzzy logic model 0.66.
Conclusion
The evaluation of the results obtained from the fuzzy logic model and the artificial neural network using kappa statistical coefficient shows that the artificial neural network with Kappa statistical coefficient is 0.91 compared to the fuzzy logic model with a kappa coefficient of 0.88 more than the prediction of the risk of landslide In the Seymareh Chenar Basin. Based on the zoning, the artificial neural network model was 10.12, 22.92, 31.44, 20.76, 15.16 percent of the area in the low, medium, high and very high risk classes has it.
