As one of the most promising new energy sources today, ocean current energy has become an important part of energy strategies. There are short-term disorderly fluctuations in the flow rate of ocean current energy. The uncontrolled input of kinetic energy from the ocean current can lead to poor quality of power generated by current energy generators. The existing technology of current energy generation uses mechanical rigid transmission, which is prone to fatigue damage and low reliability under variable load fluctuations. In this paper, a joint simulation platform based on AMESim and Simulink is constructed based on 50kW hydraulic transmission and control power generation equipment. This paper establishes a mathematical model of the hydraulic transmission control system and proposes a constant frequency control algorithm based on a deep learning prediction model to improve the steady-state accuracy of the hydraulic motor speed. This paper proposes a deep learning prediction model based on EWT (Empirical Wavelet Transform)-LSTM (Long Short-Term Memory)-CNN(Convolutional Neural Network), which improves the prediction accuracy by 12.26% compared to short-term memory neural network. The model improves the motor speed dynamic accuracy by 90%, the standard deviation index by 78.11%, and the maximum deviation by 86.10% compared to the feed-forward Proportional Integral Derivative (PID)control algorithm. Therefore, the model can effectively improve the quality of the system’s power generation. At the same time, the time cost of the model for a single prediction is less than the sampling time of other control algorithms. In this paper, the simulation results are verified by a 50kW hydraulic transmission control experimental bench. The constant frequency control algorithm based on the deep learning prediction model can effectively improve the constant frequency dynamic accuracy of the motor of the hydraulic transmission control power generation equipment, which in turn improves the system power generation quality.