LSTM-Based Prediction Control of Free-Surface Tank

Abstract

The free-surface tank is a typical ship roll reduction device. It can also play a role in ballast in some large ro-ro ships, and has the advantage of a wide range of periodic adjustments. The performance of the free-surface tank is mainly affected by the characteristics of the fluid in the tank and the control strategy. In order to obtain the ideal anti-roll effect, computational fluid dynamics (CFD) simulations are used to obtain the fluid characteristics within the constraints of the hull space and layout, taking into account the actual shape and internal baffles. On this basis, in order to overcome the shortcomings of the traditional control method with poor real-time performance, a period prediction method based on long short-term memory (LSTM) neural networks is developed to adapt to the period of the free tank on the surface and the rolling motion of the ship, and the proportional-integral-differential (PID) controller is used to realize the control of the liquid level in the tank. The simulation results show that the proposed predictive control strategy based on LSTM can improve the control performance by reducing the phase lag, which can not only eliminate the rolling phenomenon in the low frequency area of the anti-rolling tank, but also better reduce the ship rolling, and the anti-rolling effect can reach 30%.