Research On Active Disturbance Rejection Intelligent Control Of Ship Course Based On Parameter Learning

In order to overcome the negative factors such as the nonlinearity and uncertainty of the ship during navigation,and improve the control performance of autopilot,research on control design for the ship course based on linear active disturbance rejection control(ADRC)is carried out.Moreover,the intelligent ADRC ship course control strategy with adaptive parameter is proposed by introducing some other intelligent methods.The research work and innovativeness are as follows:1.Design of ship course linear ADRC controller: The nonlinear Nomoto model is used to design the autopilot.Consider the characteristic of the rudder,and take the rudder and the ship model as a whole.The influence caused by the rudder is regarded as the disturbance of the ship,which is observed and compensated by the extended state observer(ESO).Thus,a linear ADRC ship course controller is given.2.Design of adaptive ADRC controller based on adaptive network-based fuzzy inference system: In order to improve the control performance of the controller,the advantage of adaptive network-based fuzzy inference system is applied to the online adaptive strategy design of ADRC parameters.On this basis,the ADRC controllers with adaptive ESO parameters and adaptive PD parameters are designed,respectively.The performance of two controllers is verified by simulation under different sea conditions.3.Design of adaptive ADRC controller based on deep belief network: As a neural network with feature extraction ability,deep belief network can be utilized to design the parameter adaptive strategy.Based on this idea,the simultaneous online adjustment of the observer bandwidth and the linear feedback gain of the ADRC controller is realized.Besides,the numerical simulation results are compared with the linear ADRC controller under the same sea conditions to verify the superiority of the adaptive ADRC controller.4.Design of adaptive ADRC controller based on Q-learning: Reinforcement learning has attracted much attention in recent years due to its high quality strategy learning effect.In this paper,the problem of ADRC parameters online adjustment is equivalent to the optimization problem of parameters value selection strategy.For a class of ADRC problems,an adaptive algorithm of ADRC parameters based on Q-Learning is proposed and then applied to the adaptive ADRC design for ship course.Besides,considering the practical application of the ship course controller,a method of non-equal probability random initialization is given in the design to select the initial data during offline training.Simulation results show that the controller can effectively suppress external interference with high tracking accuracy and fast tracking speed.