To solve the problem that electric ships with batteries as a single power supply can't adapt to complex operating conditions, a composite power supply system structure and an energy management strategy with lithium batteries as the main power supply and ultracapacitors as the auxiliary power supply are proposed. A composite power electric ship model is established, and an energy management strategy based on model predictive control is formulated. A radial basis function neural network is used to predict future power demands； aiming to minimize the energy loss of the composite power supply system, a dynamic programming algorithm is used to optimize the output power of ultracapacitors in the prediction interval. The energy management strategy is simulated on MATLAB/Simulink platform. The results show that the energy management strategy based on model predictive control has good real-time performance, and its energy loss is 14.57% lower than that of the rule-based energy management strategy.