Abstract: For a multi-agent system with unknown disturbances, an extended state observer is designed to estimate the state of each agent and the unknown disturbances, and the estimations are sent to the neighbor agent through the wireless network. A Stackelberg game is utilized to describe the bandwidth allocation problem of the multi-agent system. A predictive cloud control scheme that can compensate for network delay is designed. The Nash equilibrium solution of Stackelberg game and the consistency and stability conditions of the multi-agent system are given. The validity is verified by simulation examples.