Parameter Identification in Stochastic Dynamic Systems based on API Approach (with applications in Biology)

In this talk, we discuss the following two problems. First, we present the Auxiliary Performance Functional (API) developed by Prof. I.V. Semushin and study its role in the state and parameter estimation of linear discrete-time stochastic systems. The minimization procedure of the API with respect to parameters of the Data Model is then considered. Our approach differs from what has been done earlier in the adaptive filtering theory. We may mention that the minimization with respect to the parameters of the Adaptive Filter instead of the Data Model was considered previously. Second, we concern with robust array adaptive filters grounded in SR and UD covariance matrix factorizations used for the API gradient evaluation in identification algorithms. As a reallife example, we consider an application of the API approach to linear time-invariant statespace stochastic MIMO filter systems arising in human body temperature daily variation adaptive stochastic modeling. Simulation results and conclusions are also provided. Key words: linear stochastic system, parameter estimation, model identification, Auxiliary Performance Functional (API) approach, state sensitivity evaluation methods, stochastic modeling, homeostasis, thermoregulation.