ارزيابي پارامترهاي موثر در تفرق امواج زلزله‌هاي دور از گسل به كمك شبكه‌هاي عصبي مصنوعي

Evaluation of Effective Parameters on Wave Diffraction of Far-Fault Ground Motions using Artificial Neural Networks

سازه‏هاي زيرزميني به عنوان شريان‌هاي حياتي، نقش به‌سزايي در مديريت بحران پس از رخداد زلزله ايفا مي‌كنند. زلزله‌ي ايجادشده بر اثر برخورد با تونل، امواج تفرقي ايجاد مي‌كند كه اين امواج در بسياري از موارد تخريب‌ها را در سازه‌هاي مجاور تونل افزايش مي‌دهند. براي بررسي اثر تفرق زلزله در حوزه‌ي دور، تونل دو قلوي متروي شهر شيراز و سازه‌هاي مجاور آن در نظر گرفته شده است. در اين نوشتار، يك شبكه‌ي عصبي مناسب جهت برآورد ميزان تفرق امواج زلزله در حوزه‌ي دور ايجاد و مورد آموزش قرار گرفته است. نتايج نشان مي‌دهد كه شبكه‌ي عصبي دقت قابل قبولي نسبت به مدل‌سازي‌هاي ايجادشده با نرم‌افزارهاي اجزا محدود دارد. همچنين يك آناليز حساسيت بر روي پارامترهاي ورودي صورت گرفته و ميزان تاثير هر يك از پارامترها تعيين شده‌ است.

Local site conditions have a great effect on ground motion and, consequently, underground structures, as important lifelines, have a vital influence on disaster management after earthquake occurrence. Among the local site topographies of interest are tunnels, which are located underground. A tunnel is an underground passageway, completely enclosed except for openings for the entrance and exit, commonly at each end. A tunnel may be for pedestrian or vehicular road traffic, for rail traffic, or for canals. Some tunnels are aqueducts to supply water, for consumption or for hydroelectric stations, or are sewers. Upon impact with a tunnel, induced earthquake motion would generate diffraction waves, which increase the damage in adjacent structures. In order to investigate the effect of wave diffraction on near fault ground motions, the twin tunnels of the Shiraz subway and their adjacent structures have been studied in the present article. Artificial neural networks (ANNs) are a field of science aimed at mimicking natural learning using mathematically based approximation. A single biological neuron is composed of three major parts: the cell body, the axon, and the dendrite. With known combinations of input and output data, a neural network can be trained to extract the underlying characteristics and relationships from the data. Then, when a separate set of input data is fed to the trained network, it will produce an approximate but reasonable output. Neural networks are highly nonlinear and can capture complex interactions among input/output variables in a system without any prior knowledge about the nature of these interactions. In this study, an appropriate artificial neural network has been generated in order to estimate the amount of diffraction of near fault earthquake waves. The results show that an idealized neural network has a high level of precision in comparison with results derived from finite element analysis. Finally, a sensitivity analysis was performed on input parameters and their percent of importance was evaluated.