SEISMIC HAZARD ANALYSIS USING ARTIFICIAL NEURAL NETWORK

Abstract

Earthquakes are among the most highly devastating natural disasters, which have been playing havoc on effected regions. Seismic hazard analysis deals with the determination of probability of occurrence of earthquake (i.e. magnitude, intensity, acceleration etc.). This leads to increased public awareness, eismic sensitivity land use planning and technically sound building construction codes of practice. Seismic hazard is typically determined using a combination of seismological, morphological, geological, geo-technical investigations combined with the history of earthquakes in the region. Variety of statistical methods are used for seismic hazard analysis, which reveal some statistical characteristics, but these are under performed due to huge computations, complex parameter consideration etc. New methods are therefore to be introduced to overcome the limitations of conventional methods. Recent trends have seen an enormous resurgence of interest in Artificial Neural Networks (ANN), and its successful adaptation to complex problems of signal and image processing. ANN is referred to as parallel distributed processing system as a paradigm of computation and knowledge representation. Seismic hazard analysis of the Himalayan region has been carried out with =in-layer feed forward error back propagation network. The whole Himalayan region is divided into six seismotectonic segments based on seismotectonics, seismic event distribution, geology etc. In each zone, the earthquake cycles are observed from the time- iii magnitude plot of seismic events. Each earthquake cycle is characterized by four energy stages of 1.accumulation, 2.increasing release, 3.intense release, 4.remnant release of energy stage. Characteristic values for each energy stage in each input segment (50 years window) are assigned as, time proportion of that stage in the segment. The network is trained with various architectures of input and output segments, hidden layers and nodes chosen. From the results of satisfactorily trained network, the conclusions drawn are as follows. In zone1, probability of occurrence of moderate to great earthquake is very low. In zone 2, probability of moderate to great earthquakes around 2030 to 2055 is very high. In zone 3, probability of occurrence of moderate to great earthquakes around 2005 to 2025 is very high. In zone 4, due to lack of data and no systematic cyclic behavior, no interpretation could be made by the selected trails. In zone 5, from year 2005 to 2030 great earthquakes probability is very high. In zone 6, probability of moderate to great earthquakes is very low in the coming 50 years.