Using Simulated Annealing Embedded Modified Gauss-Newton Algorithm to identify parameters of nonlinear degradation model

High accuracy parameter identification is important to the life prediction by the degradation model. In this paper, the Simulated Annealing Embed Modified Gauss-Newton (SAEMGN) Algorithm is developed and has been applied in the degradation model parameters eliminating for Dielectric Resonator Oscillator (DRO). By comparing the local search and global search methods, we use the modified Gauss-Newton method as the local search embedded in the Simulated Annealing. Then, we established simulation model of a DRO in Step-Stress Accelerated Degradation Test to study the convergence properties of the algorithm. Numerical comparisons with MGN, SA, and Very Fast Simulated Annealing (VFSA) shows that the new algorithm could offer a higher accuracy solution with the error values is no more than 10^-20. This algorithm will help to further improve the life prediction accuracy and credibility.