Set-valued system identification approach with stratified sampling

This paper considers the set-valued system identification with stratified sampling process. With the sampling probability known, the traditional prospective likelihood function is updated. To make it clear why traditional one bias the estimation, an explanation about system noises is given. After comparisons, an iterative algorithm with form of quasi-Newton is constructed to achieve the maximum likelihood (ML) estimate. Extensive numerical simulations illustrate the efficiency of the updated method. In the end, a brief summary is given.