In order to deal with nonlinear, time-varying, and multivariable constrained characteristics in closed-loop industrial processes, a multivariable constrained adaptive predictive control (CAPC) method based on closed-loop subspace identification is proposed. The state-space model is obtained through the closed-loop subspace identification algorithm, which is regarded as the system model. The algorithm is implemented online to update the R matrix with a receding window. By comparing the prediction errors before and after updating, it considers whether or not to update the system model. The model is then used to design the model predictive controller, which involves the solution of a quadratic program solving multivariable constraints. This paper presents a comparison between the performance of the proposed control method when applied to a 2-CSTR system, and that of an open-loop subspace CAPC method. The superiority of the proposed method is illustrated by the simulation results.