Improved regression algorithm for automated well-test analysis

Automated well-test analysis is a familiar technique, but the procedure still depends on the speed and robustness of the regression algorithms used. We examine a modification of the Cholesky factorization (CF) method for the solution of nonlinear-parameter-estimation problems and show that this modification is a very beneficial adaptation of the Gauss-Newton method. The new algorithm, the modified Gauss-Cholesky (MGC) method, is more robust under unfavorable conditions than two of the most reliable existing algorithms [the Gauss-Marquardt (GM) and the Newton-Barna (NB) methods]. This robustness seems to depend on the algorithm's ability to control the rapid change of ill-defined parameters and not on computational singularity or higher-order representation.