Numerical methods of identification for discrete-state continuous-time Markov processes

A numerical technique for identification of discrete-state continuous-time Markov models is presented. Appropriate algorithms including gradient method, the exhaustive search method and the brute force of significant parameters based on sensitivity analysis of minimization criterion are presented. A study of performance characteristics was carried out, the procedure and conditions of computer experiments are presented including types of models in use, parameters of the algorithms, parameters of the computer system. Analysis of computer experiments showed advantages of the developed identification methods over the classical first order gradient method.