In this work, we propose a revised Chaboche model to describe the static recovery behavior of 316H at 550oC. From experiments, 316H stainless steel exhibits two characteristics: the increased stress relaxation behavior with loading cycles under the fatigue-creep condition and the steady state of relaxed stress after long strain holding time. Accordingly, a new constitutive model in the framework of Chaboche model has been proposed by improving both kinematic hardening rule and isotropic hardening rule with static recovery items. The parameters related to the static recovery items could not be identified directly. To overcome this problem, we use an inverse approach that involves the following two steps: (i) establishing a Gaussian Process surrogate model by using the data generated from finite element method, and (ii) obtaining the value of parameters by using Markov Chain Monto Carle sampling under the Bayesian framework. As a result, the stress relaxation behaviors under different loading conditions can be well predicted by the proposed model.