Multiple Realisation of Real-coded Genetic Algorithm: A Tool For 2D Traveltime Tomographic Inversion

SUMMARY Seismic traveltime tomography provides method for direct estimation of velocity distribution in the subsurface from the P or S-wave first arrival traveltime data. Some of the seismic tomographic algorithms like - SIRT, ART and LSQR are widely used for their low compute intensiveness. But they may not yield the better image of the complex subsurface and especially with poor initial model. Also, these methods belong to the category of local optimization techniques, which are likely to get stuck in local minima and are derivative based. Genetic Algorithms (GA), in turn, one of the most popular global optimization method which works even with poor or no initial guess and without derivatives too. It can work well in various operator and coding modifications, e.g., binary and real number coding etc, with any global optimization problem. Any inversion algorithm consists of two important modules; one is the optimization technique and other is the forward modeler scheme. We have to start with certain model, which is nothing but the set of parameter values to be inverted, do a forward modeling, compare the results with the experimental data, update the guess using the optimization method and repeat the process until the match is satisfactory. In current investigation, we are dealing with a problem of inverting the first arrival traveltime data to reconstruct the subsurface velocity model which is nothing but a tomographic inversion problem. Our aim of the present study, is to develop and demonstrate an efficient and robust tool for traveltime tomographic inversion. To achieve this, we have developed an inversion algorithm taking real-coded genetic algorithm (RCGA) as optimization technique and used its multiple realization (run), in order to make sure that we are really near to the global minima. Here, the forward modeler is an efficient finite difference based scheme for 2D first arrival traveltime calculation of a given velocity model. The developed algorithm is highly compute intensive, hence parallelized using Hybrid Island model. The results and the performance of the algorithm are presented and discussed here for a complex synthetic velocity model.