Abstract: In order to analyze the stability of hybrid semi-tensor product model of the Internet-based teleoperation system, it's first to convert the hybrid semi-tensor product model into an impulse switched system, and then discusses stability. Next a nonlinear model predictive controller is designed based on the hybrid semi-tensor product reference model. In the time delay estimation and compensation for the model predictive controller, the Internet time delay is divided into several parts and then puts forward a Gaussian mixture first, the entropy maximization following unsupervised learning time delay estimation method, by which the time delay is compensated and the remote robot is controlled. At last, an experiment is carried out, and the experimental results show that the method advanced in this paper can meet the requirements of the Internet-based teleoperation system.