Nonlinear crosshole traveltime tomography using hybridizing particle swarm optimization with linearized least squares algorithm

Crosshole tomography problem usually uses linearized (local) techniques. Applying a linearized scheme, the cost function will converge to the global minimum only if a starting model close enough to the global minimum is available. In contrast, global optimization approaches have the ability to fi nd global minimum: even in functions with several local minima regardless of the starting model. However, they are expensive especially for most large dimensional seismic problems. We overcome the limitations of each individual approach by introducing a sequential hybrid algorithm combining particle swarm optimization (PSO) with linearized least squares (LLS). PSO simulates the social behaviour of birds fl ocking or fi shes schooling. It is implemented to obtain the appropriate initial guess for the LLS scheme. To compare the performance of the LLS, PSO, and hybrid algorithms, the methods are tested on the noise-free and noisy synthetic data sets. The resulting tomograms show that the proposed hybrid approach is a more effective inversion tool rather than using the LLS and the PSO. Moreover, a comparison between PSO and simulated annealing approach reveals that PSO provides superior results. Finally, real data from the Seinsfeld site was inverted by the hybrid method and a low-velocity anomaly was detected in the site.