An iterative algorithm using a neural network for nonlinear traveltime tomography

Traveltime tomography based on bending ray paths is a complex nonlinear problem. This paper proposes a new iterative approach for this problem. Firstly, a trial slowness model is assumed. Secondly, find the ray paths with the least traveltime by searching the shortest path in a directed graph. Thirdly, use a neural network proposed by Tank and Hopfield (called the TH net) to solve a damped least-squares problem with some constraints. The solution can give the best least-squares fit to the measured traveltime data. This paper describes how to determine the parameters for the net in detail. The fourth step is to determine whether or not the iteration should stop. Experimental results produced by the algorithm are given, and some conclusions are drawn