This paper presents an iterative approximate dynamic programming (ADP) algorithm to obtain the optimal control for microgrid, where a novel discrete adaptive law is used to learn the unknown weights in the leaning scheme. The objective is to minimize the cost of the microgrid and prolong battery lifespan. First, the microgrid system model and objective function are established. Then, aiming at the optimal control of microgrid, an iterative ADP with the novel adaptive law is constructed. To facilitate the implementation of the pro-posed method, neural networks are used to approximate value function and optimal control law. Numerical experiments illustrate the performance of the proposed method.