To remedy the defects of the poor power grid connection and its poor stability at ports, we adopt a control strategy based on radial basis function (RBF) neural network adaptive sliding mode control. In addition, the sliding mode control is optimized by using a proportional–integral (PI) sliding surface and following a fractional sliding mode law. The neural network gives a general approximation: the parameter error is approximated by the neural network to compensate errors. Owing to the good anti-interference and robustness of sliding mode control, the stability of the shore-to-ship power grid connection is improved. The sliding mode law is proved to be able to ensure the stability of the system when an RBF neural network is adopted to approximate errors. In the environment of a MATLAB simulation, a simulation model of a shore-to-ship power grid connection is built. A simulation experiment is performed under a low voltage of 440 V, and the simulation results at different frequencies are compared with the sliding mode control and proportional–integral–derivative (PID) results without an RBF neural network. As revealed by the results, the control strategy is effective and feasible.