Aiming at the problem that the manned submersible vehicle is affected by ocean currents and over-actuated characteristics, a double-layer control structure is proposed for the manned submersible vehicle. First of all, a six-degree-of-freedoms dynamic model of the manned submersible vehicle system under the influence of ocean currents is established. Secondly, the disturbances of the submersible vehicle are divided into two parts: the disturbance caused by ocean currents and other disturbances. The ocean current observer and nonlinear disturbance observer are designed to estimate the ocean current disturbance and other disturbances, respectively. Based on the estimated disturbance information, a motion controller is designed using the back-stepping algorithm for the manned submersible vehicle. Then, a thrust allocation method based on neural network quadratic programming is proposed to address the over-actuated problem, as well as to meet the thrust constrains of the manned submersible vehicle. Finally, some numerical simulation results are given to demonstrate the effectiveness and superiority of the proposed control scheme. The superiorities of the proposed control scheme based on refined disturbance estimation and neural network thrust allocation are threefold: its capabilities to obtain more accurate disturbance estimation result, to achieve the optimal energy consumption, and to prevent the thrust exceeding the limit.