Research On Ship Trajectory Control Based On Neural Network Predictive Control

As an important marine basic industry,the shipbuilding industry plays an important strategic role in national economic development and defense construction.Trajectory tracking is a prerequisite for ships to accomplish transportation,monitoring,military activities,and other tasks.Superior control algorithms can greatly improve the efficiency and accuracy of ship trajectory tracking.Traditional mechanism-based ship motion control requires the establishment of mathematical motion models based on hydrodynamic parameters.The accuracy of the model depends on the identification effect of hydrodynamic parameters.But,existing hydrodynamic parameters identification methods cannot obtain complete ship information dynamics,and unmodeled dynamics will affect the control effect.So,this paper proposes a data driven ship track control method,which forms a ship track control strategy based on neural network predictive control by constructing a neural network predictive model of the ship.The main research content and results of this paper are as follows:(1)Research on real-time weight matrices optimization method of nonlinear model predictive control based on the improved genetic algorithm(GA).At each moment,according to the real-time ship trajectory tracking status,GA is used to online optimize each weight coefficient of the NMPC objective function in real time,solving the problem of determining constant weight values using experimental or empirical methods,which improves the efficiency of parameter setting and control effectiveness.By improving the crossover operator,mutation operator,crossover rate,and mutation rate in genetic algorithms,the optimization ability of the algorithm is improved.(2)Research on ship motion model building method based on the recurrent neural network(RNN).Considering that velocities in the ship body frame is difficult to measure,a data driven neural network model of ship motion is established to estimate the future motion state of the ship,using only the current control input and the historical position and heading data.Simulations are conducted for different combinations of activation function and optimization methods,and it is determined that RNN has the best prediction performance when using the activation function of Leakyrelu and the optimization method of Adam.(3)Research on model predictive control method based on neural network prediction model.A neural network prediction controller is constructed by using neural network as the prediction model of NMPC for ship trajectory tracking control.Neural network predictive control(NNPC)is designed based on the principle of data driven,and it establishes a ship motion model by searching for the mapping relationship between input and output data.The motion model predicts the position and heading information at multiple future moments to establish an online optimization objective function,and uses genetic algorithms to optimize the nonlinear objective function.Therefore,the optimal control action for ship trajectory tracking control is obtained.Simulation results verify the feasibility and effectiveness of the proposed NNPC strategy and the online weight matrices optimization approach.