CN-121976577-A - Bow blowing positioning control method for trailing suction hopper dredger

CN121976577ACN 121976577 ACN121976577 ACN 121976577ACN-121976577-A

Abstract

The invention provides a bow blowing positioning control method of a trailing suction hopper dredger. The method builds a mechanism environment during bow spraying operation of the trailing suction hopper dredger based on a simplified ship three-degree-of-freedom dynamics and kinematic model, adopts a maximum entropy reinforcement learning algorithm to build a reinforcement learning network, greatly improves the environment global exploration capacity of the model, has higher robustness, and is beneficial to solving the problem of inaccurate positioning of the bow spraying operation of the traditional trailing suction hopper dredger by introducing a priority experience playback mechanism, a mixed sampling factor and strategy delay updating into the model, and relieving the problem of sample diversity reduction caused by excessive bias by utilizing the mixed sampling factor on the basis of maintaining priority experience playback of high-efficiency focusing key samples, and simultaneously reducing interference of value estimation errors to strategy updating by a DPU (differential pulse width unit).

Inventors

LUAN KAI
YU MENGHONG
DONG YANLI
CHENG SHUANG
YU KAI

Assignees

江苏科技大学

Dates

Publication Date: 20260505
Application Date: 20260121

Claims (10)

1. The bow blowing positioning control method of the trailing suction hopper dredger is characterized by comprising the following steps of: s1, establishing a three-degree-of-freedom dynamics model and a bow spraying reaction force model of the trailing suction hopper dredger; S2, constructing a bow blowing positioning simulation environment based on a simulation platform according to a three-degree-of-freedom dynamics model and a bow blowing reaction force model, and defining a state space, an action space and a reward function of the bow blowing positioning simulation environment; S3, establishing a MaxEnt RL control model by adopting a priority experience playback mechanism and a strategy delay update mechanism; And S4, deploying a MaxEnt RL control model in the constructed bow blowing positioning simulation environment, training, and performing bow blowing positioning control by using the trained MaxEnt RL control model.
2. The method for controlling bow blowing positioning of the trailing suction hopper dredger according to claim 1, wherein the three-degree-of-freedom dynamics model formula is as follows: wherein: For the position pose in the NED coordinate system, Is in the north-oriented position and is in the normal position, In order to be in the eastern direction, Is the heading; For the transformation matrix of the coordinate system, For the speed of the ship, Is a matrix of inertia which is a matrix of inertia, In order to provide a matrix of centripetal forces, In order to provide a damping matrix, In order to control the force of the force, Is an environmental disturbance force.
3. The method for controlling bow blowing positioning of the trailing suction hopper dredger according to claim 2, wherein the inertia matrix is: wherein: For the mass of the ship, the weight of the ship, For the mass of the ship, the weight of the ship, The longitudinal offset of the ship centroid relative to the origin of the reference coordinate system; In order to be a mass moment of inertia, , , , Is the hydrodynamic coefficient related to inertia; The damping matrix is as follows: wherein: , , , , is a damping-dependent hydrodynamic coefficient.
4. The method for positioning and controlling bow blowing of the trailing suction hopper dredger according to claim 2, wherein the environmental disturbance force comprises wind disturbance force and wave disturbance force, and the wind disturbance force formula is as follows: wherein: For the longitudinal wind-disturbance force, In order to be a transverse wind-disturbing force, Is a bow moment corresponding to wind disturbance force, Is the orthogonal area above the waterline, Is the projected area above the waterline, In order to achieve an air density of the air, For the relative wind speed, 、 And As a wind coefficient, the wind power generation system, Is the relative wind direction angle; The wave disturbance force formula is: wherein: For the longitudinal wave-disturbing force, For the transverse wave-disturbing force, Is the bow moment corresponding to the wave disturbance force, For the length of the vessel, Is the density of the seawater, and the seawater is the density of the seawater, The acceleration of the gravity is that, For the average amplitude of the waves, 、 And Is the wave coefficient of the wave, and the wave coefficient of the wave, Is the wave meeting angle.
5. The bow blowing positioning control method of the trailing suction hopper dredger according to claim 1, wherein the bow blowing reaction force model formula is as follows: wherein: In order to provide a bow-jet reaction force, For the total reaction force generated by the injection momentum, For the sloshing force caused by the bow spray, For the sloshing force caused by the bow spray, Is the moment of the bow swing caused by the bow spray, Is the counter-force coefficient in the heave direction, Is the moment coefficient in the bow-sway direction, Is the longitudinal offset of the center of the spout relative to the center of mass of the hull.
6. The method for controlling bow blowing positioning of the trailing suction hopper dredger according to claim 1, wherein the formula of the state space is: wherein: In order to be a state space, Is in the north-oriented position and is in the normal position, In order to be in the eastern direction, In order to be the angle of the bow direction, In order to achieve a heave velocity, In order to achieve the speed of the swaying, For the angular velocity of the bow swing, For the north-oriented target position, In order to orient the subject to the east, For the target bow angle, Is the rotation speed of the mud pump.
7. The method for controlling bow blowing positioning of the trailing suction hopper dredger according to claim 1, wherein the formula of the action space is: wherein: In order to be a space for the motion, A control thrust applied longitudinally to the hull; a control thrust applied transversely to the hull; a control moment applied to the hull about a vertical axis; Is a rotation speed control instruction of the mud pump.
8. The method for controlling bow blowing positioning of a trailing suction hopper dredger according to claim 1, wherein the reward function formula is: wherein: as a function of the reward, In order to reward items for the heading, For the location-rewarding item, Rewarding items for spraying mud; The heading reward term formula is: the position reward term formula is: the formula of the mud spraying reward term is as follows: wherein: 、、 And As the weight coefficient of the light-emitting diode, In order to be the angle of the bow direction, For a yaw bonus item, For the yaw bonus term, For a normalized distance error in the eastern direction between the current hull and the target position, For a normalized distance error in the western direction between the current hull and the target position, As the current rotation speed of the mud pump, And (3) with The minimum and maximum allowable working speeds of the mud pump are respectively.
9. The method for positioning and controlling bow blowing of a suction hopper dredger according to claim 1, wherein the MaxEnt RL control model comprises 5 neural networks, namely 1 Actor network for action selection, 2 Critic networks for action value estimation and 2 target networks corresponding to the Critic networks.
10. The method for controlling bow blowing positioning of the suction dredger according to claim 9, wherein the step S4 comprises initializing a hyper-parameter value of a MaxEnt RL control model, interacting with a bow blowing positioning simulation environment by a random strategy or Gaussian noise strategy, collecting experience quaternion into a priority experience playback pool, selecting a set batch of experience quaternion, inputting a state of the experience quaternion into an Actor network to obtain action and strategy entropy, taking the state and the action of the experience quaternion as input parameters, respectively inputting the states and the actions into two Critic networks to obtain value, calculating a loss value of the Actor network, carrying out parameter updating on the Actor network by adopting a gradient strategy, calculating strategy entropy loss by adopting strategy entropy and combining a target strategy value, copying Critic network parameters and the Actor network parameters, updating the hyper-parameter of the MaxEnt RL control model by utilizing a genetic algorithm, selecting an optimal hyper-parameter group after setting training rounds, and carrying out bow blowing positioning control by utilizing the corresponding MaxEnt RL control model.

Description

Bow blowing positioning control method for trailing suction hopper dredger Technical Field The invention relates to the technical field of intelligent ship control, in particular to a bow blowing positioning control method of a trailing suction hopper dredger. Background As common large-scale dredging equipment, the drag suction dredger has multiple functions of dredging, loading, transporting, hydraulic filling and the like, wherein bow blowing is one of the common dredging modes of the drag suction dredger, and is suitable for the engineering of offshore hydraulic filling land making, channel widening, management and the like. The bow blowing operation sprays silt slurry to a target area at high speed through a bow nozzle so as to realize land making or maintenance of a channel, however, in the process, the ship body is in shallow water on the shore and in a complex hydrodynamic environment, is subjected to multi-source disturbance such as wind, wave, flow, jet reaction force and the like, ship drift and bow direction deviation are extremely easy to occur, position errors of a hydraulic reclamation area are caused, and filling precision and operation efficiency are further reduced. The existing bow blowing positioning control method mainly depends on traditional methods such as proportional-integral-derivative (PID) control, self-adaptive control or Model Predictive Control (MPC) and the like. The method can realize basic positioning control under the condition of ideal environment or known dynamic parameters, but has the defects that the method is highly dependent on a ship dynamic model and parameter identification, the ship mass dynamically changes along with the mud discharging process in actual construction, the randomness of the environment interference is high, the modeling precision is difficult to ensure, when the method is subjected to strong disturbance such as stormy waves and currents, the controller is easy to saturate or lose effectiveness, the ship attitude is difficult to maintain, and the method generally lacks self-correction capability on long-term accumulated deviation based on short-term prediction or instantaneous error feedback, and is difficult to maintain the positioning precision in the long-term operation process. With the development of artificial intelligence, reinforcement learning (Reinforcement Learning, RL) is increasingly finding an advantage in complex dynamic system control as a data-driven intelligent decision method. Reinforcement learning is able to gain experience through continuous interaction with the environment and continuously optimize decision strategies based on rewarding mechanisms, thereby achieving effective control of nonlinear, uncertainty systems without requiring an accurate mathematical model. Reinforcement learning has proven to achieve superior performance in complex environments over traditional controllers in the areas of autopilot, unmanned aerial vehicle formation, robotic handling, etc. However, compared with unmanned vehicles and unmanned aerial vehicles which have largely applied reinforcement learning, the working environment of the trailing suction hopper dredger is more complex, and various uncertain factors such as large inertia of the ship body, significant change of dynamic parameters along with the working process, superimposed interference of wind and wave currents, jet reaction force and the like exist. These characteristics place higher demands on the robustness, adaptation and long-term stability of the control algorithm. Disclosure of Invention The invention aims to provide a bow blowing positioning control method of a drag suction dredger, which adopts reinforcement learning to improve precision. The technical scheme is that the bow blowing positioning control method of the trailing suction hopper dredger comprises the following steps: s1, establishing a three-degree-of-freedom dynamics model and a bow spraying reaction force model of the trailing suction hopper dredger; S2, constructing a bow blowing positioning simulation environment based on a simulation platform according to a three-degree-of-freedom dynamics model and a bow blowing reaction force model, and defining a state space, an action space and a reward function of the bow blowing positioning simulation environment; S3, establishing a MaxEnt RL control model by adopting a priority experience playback mechanism and a strategy delay update mechanism; And S4, deploying a MaxEnt RL control model in the constructed bow blowing positioning simulation environment, training, and performing bow blowing positioning control by using the trained MaxEnt RL control model. Specifically, the three degree of freedom dynamics model formula is as follows: wherein: For the position pose in the NED coordinate system, Is in the north-oriented position and is in the normal position,In order to be in the eastern direction,Is the heading; For the transformation matrix of the coordinate syste