COMBINATION OF GENERALIZED APPROXIMATION METHOD (ANFIS) AND GLOBAL OPTIMIZATION TECHNIQUES (GENETIC ALGORITHM) IN ESTIMATION STRONG GROUND MOTION ATTENUATION LAW

In the most conventional method of obtaining attenuation relationships, a functional form which originates from physics of attenuation waves and experts knowledge is postulated and parameters at such form are determined based on regression analysis among existing catalogs of earthquakes. Such kinds of attenuation relationships are strongly depend on the global form of presumed functional, expert knowledge and less on the behavior of data in catalogue. In this paper, knowledge based and data oriented method; called Adaptive Neuro-Fuzzy Inference Systems (ANFIS) in combination with the global optimization procedure (Genetic Algorithm (GA)) is implemented to establish an attenuation relationship which is completely according to the characteristics of the seismic catalogue. Three important advantages arise from the proposed method: 1. omits the prevalent functional form and leads to unified form of attenuation laws, 2. extracts the form and expert knowledge from the data, and 3. considers the uncertainty and dependency of the events inherently based on the data fusion concepts. To verify results of the proposed method, the catalogue used by Ambrasyes et al (2005) for horizontal and vertical peak ground acceleration for 589 earthquake records in Europe and Middle East caused by shallow crustal earthquakes with magnitude Mw=5 and distance to the surface projection of the fault less than 100 km is considered and attenuation law for horizontal and vertical peak ground acceleration is extracted and compared with Ambrasyes et al (2005) relation.