A data-driven iterative Adaptive Dynamic Programming (ADP) algorithm including online sampling and offline learning is proposed. The H∞ control problem of the nonlinear system with input constraints and unknown models is solved only by online data. The model-free Hamiton-Jacobi-Isaacs (HJI) equation is derived by using the methods of Policy Iteration (PI) and Iteration Reinforcement Learning (IRL).Three neural networks are constructed. After the online acquisition of system data is completed, the off-line learning method is used to approximately solve the model-free HJI equation, and then the value function, control strategy and disturbance strategy are obtained. The parameter of the neural network is solved by the least squares method. Simulation results verify the feasibility of the algorithm.