A two-step optimization strategy for S-wave velocity structure inversion

Various combinations of genetic operators are examined by the virtual inversion, as the result it is pointed out that the offline capability is best in dispersion curves inversion by combining float coding with roulette selection. One two-step inversion strategy is suggested by integrating the above two with arithmetical uniform crossover and mutation into a basic genetic inversion frame, and adding the immune heuristic method and the no-repeating forward calculation method to the above basic frame, then after several runs conducting Simulated Annealing Algorithm to the result of each individual run,finally the best of the local optimum is chosen as the final result. The immune heuristic method speeds up the evolution by adopting the best individual's information, enhances the local search in its neighboring space by adding one random number with normal distribution, while giving attention to the search in other areas by rectifying the distribution deviation. The local search and global search are linked effectively; the disturbance to evolution of the genetic algorithm is minimized. By the no-repeating forward calculation, the times of forward calculations are reduced greatly, the calculating cost is saved and the evolution efficiency is increased. The shortcoming of averaging the results of all individual runs is avoided. The stability and precision of inversion is increased and the non-uniqueness is reduced.