CN-121978958-A - Model-free intelligent self-adaptive vector control method for chemical carrier propulsion system

CN121978958ACN 121978958 ACN121978958 ACN 121978958ACN-121978958-A

Abstract

The invention discloses a model-free intelligent self-adaptive vector control method of a chemical product ship propulsion system, and relates to the technical field of ship control. The method comprises the steps of establishing a simulation environment of a chemical ship propulsion system, calculating load torque of a propulsion motor in the propulsion system according to a ship-machine-paddle interaction model, establishing a compact dynamic linearization data model, describing a current and rotating speed dynamic relation by using a pseudo jacobian matrix, establishing a TD3 reinforcement learning intelligent body, outputting four key parameters of a model-free self-adaptive controller (MFAC) in real time according to a system state, substituting the parameters into the controller through action mapping, updating an estimated value on line, calculating a quadrature current instruction, and driving the chemical ship propulsion system by combining vector control and SVPWM technology. The invention solves the problem of control instability of the chemical ship caused by severe perturbation of load under complex working conditions such as high-risk liquid cargo carrying and over-connection in sailing, and realizes the running stability and safety of the chemical ship propulsion system.

Inventors

YAO WENLONG
FENG JIANLIANG
LU JINYU
YANG LEI
LIU YUCHUAN
LIU SHILONG

Assignees

青岛科技大学

Dates

Publication Date: 20260505
Application Date: 20260210

Claims (4)

1. A model-free intelligent adaptive vector control method for a chemical tanker propulsion system, the method comprising the steps of: S1, establishing a simulation environment of a chemical ship propulsion system, and calculating the load torque of a propulsion motor in the propulsion system according to the characteristics of a propeller and a ship-machine-propeller interaction model ; S2, defining a nonlinear discrete time dynamic expression of the propulsion system, constructing a Compact Form Dynamic Linearization (CFDL) data model and utilizing a pseudo-Jacobian matrix Description of the quadrature current With motor speed A dynamic change relationship between the two; S3, constructing a reinforcement learning agent based on a dual-delay depth deterministic strategy gradient (TD 3) algorithm, defining a state space, an action space and a reward function containing parameter smooth constraint, wherein the agent outputs four key parameters required by a model-free adaptive controller (MFAC) in real time according to the current system state, namely a control law step factor Control law weight factor Estimation of law step factor Estimation law weight factor ; S4, converting the optimized parameters output by the reinforcement learning agent into physical parameters through an action mapping mechanism, substituting the physical parameters into the model-free self-adaptive controller in real time, and updating the estimated value of the pseudo-Jacobian matrix on line And calculating the quadrature current command of the propulsion motor ; S5, commanding the quadrature axis current And combining a vector control strategy, generating a control signal through a Space Vector Pulse Width Modulation (SVPWM) technology, driving a chemical ship propulsion system to operate, and realizing accurate tracking of the expected rotating speed.
2. The model-free intelligent self-adaptive vector control method of a chemical tanker propulsion system according to claim 1, wherein in step S2, the construction process of the nonlinear discrete time dynamic expression, the compact format dynamic linearization data model and the control algorithm specifically comprises: (1) Defining a nonlinear discrete time dynamic expression of the propulsion system considering the complex working condition of the chemical carrier: Wherein, the Is that The output rotating speed of the propulsion system at moment; Representing the system past A historical output rotating speed sequence at each moment; Indicating that the system is at the current time In the past Inputting a quadrature current sequence at each moment; Representing an unknown nonlinear time-varying function and representing comprehensive dynamics characteristics including nonlinear friction, propeller hydrodynamic interference, liquid cargo shaking impact and parameter time-varying drift in a chemical ship propulsion system; (2) Based on the expression, a tightly-formatted dynamic linearization data model suitable for the severe-load perturbation working condition is constructed: Wherein, the And Respectively is Time of day and time of day The output rotating speed of the propulsion system at moment; Is that Control increment of the time cross-axis current; the dynamic sensitivity of the chemical ship propulsion system to the change of the input of the quadrature current is represented by a pseudo jacobian matrix at the current working point, and the parameter can implicitly capture the unmodeled dynamic and external environmental disturbance in the ship-machine-paddle system; (3) Based on the extremum principle, constructing a pseudo jacobian matrix estimation law and a model-free self-adaptive control law: the constructed pseudo jacobian matrix estimation law formula is as follows: Wherein, the In order to estimate the rhythmic step size factor, Is an estimation law weight factor; the built model-free self-adaptive control law formula is as follows: Wherein, the In order to control the rate step size factor, Is a control law weight factor.
3. The model-free intelligent self-adaptive vector control method of a chemical tanker propulsion system according to claim 1, wherein in step S3, the reinforcement learning agent is constructed by using a TD3 algorithm, and the specific design and core update mechanism comprises: (1) State space Selecting a rotational speed tracking error Rate of error change Motion vector outputted by agent at last moment As a state variable; (2) Action space Four parameters of the model-free adaptive controller are selected as motion vectors, namely ; (3) Reward function Aiming at the characteristic that the fluctuation of the rotating speed of the propulsion system is easy to be caused by the shaking coupling effect of the liquid cargo when the chemical ship transports high-risk liquid cargo or oversea refutes, the comprehensive rewarding function comprising multiple constraints is constructed for restraining disturbance and simultaneously considering the running smoothness of the propulsion system and the evolution convergence of the parameters of the controller: Wherein, the , , Respectively the weight coefficients; punishment of rotational speed tracking errors; Punishment controls the sharp fluctuations of energy; For parameter smoothing penalty term, for suppressing controller parameters , , , Abrupt change at adjacent moment, prevent the system from oscillating; (4) Establishing two Critic networks with the same structure but different parameter initial values And The smaller of the two network output values is taken when the target value is calculated: the mechanism is used for solving the problem of overestimated deviation of the cost function in the traditional DDPG algorithm and improving the accuracy of parameter learning; (5) Policy delay updating mechanism, wherein Critic network parameters are updated frequently, and the updating frequency of an Actor network is lower than that of Critic network, and only when a value network is updated The next time is ) And then, the strategy network parameters and the target network parameters are updated once so as to ensure that the strategy update is established on the basis of convergence of the value estimation, and thus the more stable and smooth MFAC control parameters are output.
4. The method for model-free intelligent adaptive vector control of a chemical tanker propulsion system according to claim 1, wherein in step S4, the real-time calculation process of the motion mapping mechanism and the model-free adaptive controller comprises: (1) The action mapping mechanism converts the normalized parameters output by the reinforcement learning agent into four physical control parameters of the model-free self-adaptive controller, and the calculation formula is as follows: Wherein, the Output for agent The normalized action value is within the range of [ -1, 1]; Is the mapped first A MFAC physical parameter; [ the physical effective interval preset for the parameter ensures that the mapped parameter is constantly greater than zero and is in a system stable domain; (2) Real-time computation of pseudo-jacobian matrix estimation using mapped parameters of agent output 、 Updating the estimated value according to the following formula : (3) Real-time calculation of control laws by mapped parameters of agent output 、 Combining updated Calculating the quadrature current command at the current moment : Wherein, the Is the desired rotational speed at the next moment.

Description

Model-free intelligent self-adaptive vector control method for chemical carrier propulsion system Technical Field The invention belongs to the technical field of ship electric propulsion and intelligent control, and particularly relates to a control method for carrying out online real-time optimization on key parameters of a model-free adaptive controller (MFAC) by using a reinforcement learning algorithm aiming at a chemical ship propulsion system. Background The chemical ship is used as a special liquid cargo transport ship, mainly bears the global logistics transport task of flammable, explosive, toxic or highly corrosive liquid chemicals, and is an indispensable key link in petrochemical industry chains. Under the background of the current global shipping industry for greatly advancing green ships and intelligent shipping, the energy conservation, emission reduction and accurate control performance of a ship propulsion system become core focuses of technical development. A Permanent Magnet Synchronous Motor (PMSM) is generally adopted as main propulsion power for large and medium-sized modern chemical ships, and the remarkable advantages of high power density, high efficiency, high overload capacity, wide speed regulation range and the like are utilized, so that a full-electric propulsion system is constructed by matching with a high-performance vector control technology, and the requirements of extremely high dynamic response and control precision of the power system under the conditions of complex navigation, port entering and exiting operation and special operation of the ships are met. It is worth pointing out that chemical ships face extremely challenging complex conditions when performing high-risk operations such as refuting in offshore navigation. The two vessels need to keep close synchronization under the low-speed sailing state, and at the moment, the ship suction effect generated by ship-ship fluid coupling and the intense sloshing caused by the free liquid level effect of liquid cargo in the cabin are overlapped with each other, so that the propeller load torque presents strong nonlinearity and time-varying characteristics. Conventional vector control strategies based on proportional-integral (PI) regulators generally rely on fixed controller parameters and cannot accommodate real-time drift of vessel draft, center of gravity and moment of inertia due to liquid cargo flow during the overbreak process. The parameter mismatch is very easy to cause the overshoot or oscillation of the rotating speed of the propulsion system when the load is suddenly changed, thereby causing the collision of two ships or the breakage of an oil pipeline, and causing serious chemical leakage or safety accidents. In recent years, model-free adaptive Control (MFAC) has received attention as a data-driven Control method to solve Control problems caused by uncertainty of Model parameters of a propulsion system. The method does not depend on a physical model of a controlled object, only uses input and output data for control, and has strong robustness on unmodeled dynamics. However, the performance of an MFAC controller is highly dependent on its internal control law step size factor, weight factor, and related parameters in the pseudo-jacobian matrix (PJM) estimation law. The existing researches mostly adopt an empirical method to test or introduce Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) for offline optimization, the methods have large calculated amount and cannot realize online real-time adjustment, and the requirements of a chemical ship on the quick dynamic response and absolute stability of a propulsion system under the special working conditions are difficult to meet. The dual-delay depth deterministic strategy gradient (TD 3) algorithm is used as an advanced depth reinforcement learning algorithm, and has the capability of performing powerful perception and decision in a continuous action space. The algorithm effectively solves the problems of overestimation of the cost function and unstable training in the traditional DDPG algorithm by introducing a double Critic network and a strategy delay updating mechanism. Compared with the defect that the traditional offline optimization algorithm cannot adapt to the real-time change of the system, the TD3 algorithm can accurately perform online real-time optimization on four key parameters of the control law and the estimation law in the MFAC controller by utilizing real-time interaction data, and the controller is ensured to always operate under the optimal parameter combination in a complex time-varying environment, so that the method has extremely high self-adaptability. The invention provides a model-free intelligent self-adaptive vector control method of a chemical carrier propulsion system, which considers the special requirements on the stability of the propulsion system when carrying high-risk liquid cargo and performing over-barge operation