Embedded Estimation Sequential Bayes Parameter Inference for the Ricker Dynamical System

Abstract

The dynamical systems are comprised of two components that change over time: the state space and the observation models. This study examines parameter inference in dynamical systems from the perspective of Bayesian inference. Inference on unknown parameters in nonlinear and non-Gaussian dynamical systems is challenging because the posterior densities corresponding to the unknown parameters do not have traceable formulations. Such a system is represented by the Ricker model, which is a traditional discrete population model in ecology and epidemiology that is used in many fields. This study, which deals with parameter inference, also known as parameter learning, is the central objective of this study. A sequential embedded estimation technique is proposed to estimate the posterior density and obtain parameter inference. The resulting algorithm is called the Augmented Sequential Markov Chain Monte Carlo (ASMCMC) procedure. Experiments are performed via simulation to illustrate the performance of the ASMCMC algorithm for observations from the Ricker dynamical system.