This paper investigates the causes associated to the low performance of an ILU preconditioner and proposes a methodology to improve its quality and the GMRES iterative method convergence rate as well. The goal is solve the load flow sublinear problem, emphasizing heavy load scenarios of difficult convergence. The methodology is based on two steps. On the first one, a preconditioner is constructed based on the Jacobian matrix calculated and reordered at the first Newton-Raphson iteration. The second step is taking into account only in case of GMRES failure due to low quality preconditioning. The process is restarted an a better quality preconditioner is constructed based on the preprocessed Jacobian matrix through scaling and nonsymmetric and symmetric (reordering) permutations. In order to reduce the computational cost associated to the construction of the preconditioner, it is proposed a strategy for keeping it fixed as long as possible. Numerical experiments corroborate the numerical robustness and the computational efficiency of the proposed methodology.
Krylov subspace iterative methods; Preconditioners; Algebraic Linear Equations; Load-Flow
