In this paper, a novel optimal control method by broad learning system (BLS)-based adaptive dynamic programming is proposed. This method is used to course-keeping control of ship under the conditions of unknown system dynamics, energy saving and reduced equipment waste. First, it is necessary to identify the unknown nonlinear dynamics in the ship's course keeping system, so a model network is established by BLS. Then, a BLS based optimal control scheme is proposed, the data used for the BLSs training is composed of current data and recorded data. The connection weights of the approximator is obtained by real system without need of iteration. Therefore, compared with the traditional adaptive dynamic programming (ADP) algorithm that requires multiple iterations, BLS-based ADP prove the effectiveness and high performance of the proposed optimal control law for course-keeping in ship autonomous driving.