Layered anti-interference control method for autonomous towing ship path tracking

Abstract

The invention discloses a layered anti-interference control method for autonomous towing ship path tracking. A hierarchical anti-interference control system is constructed, an outer layer plans a reference trajectory by using a reinforcement learning and multi-agent collaborative optimization algorithm, a middle layer estimates interference by using an observer combining an attention mechanism long-short term memory network and adaptive Kalman filtering, and an inner layer performs control by fusing model prediction, adaptive sliding mode and adaptive fuzzy control. The method solves the problems that the ship is influenced by external interference and self dynamic characteristic change, and the path tracking precision is low, can effectively resist interference, adapts to ship dynamic change, remarkably improves the path tracking precision and reliability, and guarantees safe and efficient navigation of the ship in a complex marine environment.

Description

Technical Field

[0001] The present invention relates to the technical field of ship control, and particularly to a hierarchical disturbance rejection control method for path tracking of autonomous tugboats. Background Art

[0002] In the field of ship transportation, the application of autonomous tugboats is of great significance for improving transportation efficiency and reducing labor costs. However, the path tracking control of current autonomous tugboats faces severe challenges. On the one hand, the marine environment in which ships sail is extremely complex. The sea waves, sea winds, and water currents not only have continuously changing magnitudes and directions of disturbing forces, but also have highly nonlinear and uncertain disturbance characteristics. Traditional control methods are based on simple linear models and are difficult to accurately estimate and effectively compensate for these complex disturbances, resulting in a significant deviation between the actual sailing path of the ship and the preset path, and it is difficult to ensure the accuracy and safety of the towing task. On the other hand, the ship itself has complex dynamic characteristics, such as inertial changes caused by uneven mass distribution of the ship, nonlinear responses of the propulsion system and the rudder, and changes in hydrodynamic characteristics under different loading conditions of the ship, making it difficult to establish an accurate ship control model and further increasing the difficulty of path tracking control. Existing control methods mostly adopt conventional feedback control strategies and lack an effective response mechanism to disturbance factors and dynamic changes in the characteristics of the ship itself, and cannot meet the strict requirements of high-precision path tracking, greatly limiting the wide application of autonomous tugboats in complex marine environments. Summary of the Invention

[0003] The purpose of the present invention is to provide a hierarchical disturbance rejection control method for path tracking of autonomous tugboats. By constructing a new hierarchical disturbance rejection control method, the problems brought about by external disturbances and changes in the dynamic characteristics of the ship itself are overcome, high-precision and high-stability path tracking of the ship in a complex marine environment is achieved, the operation efficiency and safety of autonomous tugboats are significantly improved, and the technical gap in dealing with complex working conditions in this field is filled.

[0004] The present invention is achieved through the following technical solutions:

[0005] A hierarchical disturbance rejection control method for path tracking of autonomous tugboats, by constructing a hierarchical disturbance rejection control system, including:

[0006] An outer layer trajectory planning layer, using a dynamic path planning algorithm based on reinforcement learning and multi-agent collaborative optimization, integrating the starting position, target position, real-time sailing environment information of the ship and the performance constraints in different sailing stages of the ship, and introducing a multi-agent information interaction and cooperation mechanism;

[0007] The middle-layer interference estimation layer uses an adaptive extended state observer improved by combining a long short-term memory network and a quantum particle swarm optimization algorithm to perform multi-dimensional measurements on the six-degree-of-freedom motion parameters of the ship, the sensor measurement noise, and the changes in environmental parameters;

[0008] The inner-layer control execution layer operates based on a fusion strategy of model predictive control (MPC), adaptive sliding mode control (ASMC), and fuzzy neural network control (FNC), according to the reference trajectory and the interference estimation result.

[0009] Further, as an improvement to the technical solution of the present invention, the navigation environment information includes wave height, wave period, water flow velocity and direction, sea wind intensity and direction.

[0010] Further, as an improvement to the technical solution of the present invention, the external interference includes the high-order components of the wave interference force and the sea wind interference torque that changes with time, and the self-uncertainty factors include the model parameter perturbations of the ship under different working conditions.

[0011] Further, as an improvement to the technical solution of the present invention, the fusion strategy operates based on the real-time motion state of the ship, the interference estimation result, and the real-time loading state of the ship and the changes in sea conditions.

[0012] Further, as an improvement to the technical solution of the present invention, when the ship is sailing, the ship's position, speed, acceleration, heading angle, roll angle, and pitch angle information are collected in real time as the input of the middle-layer interference estimation layer.

[0013] Further, as an improvement to the technical solution of the present invention, when the ship deviates significantly from the reference trajectory, the outer-layer trajectory planning layer uses a dynamic path planning algorithm that combines reinforcement learning and multi-agent collaborative optimization to re-plan the reference trajectory.

[0014] Further, as an improvement to the technical solution of the present invention, when the interference situation changes greatly, the inner-layer control execution layer adjusts the interference compensation strategy and control parameters.

[0015] Further, as an improvement to the technical solution of the present invention, the ship's mass, moment of inertia, and detailed self-parameters of the hydrodynamic coefficients under different loading states are input during the initialization stage.

[0016] Further, as an improvement to the technical solution of the present invention, the control input calculated by the inner-layer control execution layer includes the dynamic adjustment of the rudder angle and the optimal distribution of the thruster thrust.

[0017] Further, as an improvement to the technical solution of the present invention, in the ship formation towing scenario, a multi-agent collaborative optimization algorithm is used to achieve collaborative path planning and interference coordination control among ships.

[0018] It should be noted that:

[0019] Hierarchical Structure Design: Pioneering a multi-scale hierarchical anti-interference control architecture, which is collaboratively composed of an outer layer trajectory planning layer, a middle layer interference estimation layer, and an inner layer control execution layer. The outer layer trajectory planning layer applies a dynamic path planning strategy that combines deep reinforcement learning and genetic algorithms. This strategy is based on the ship's starting position, target position, real-time updated big data of the marine environment, and the performance constraint conditions of the ship at different navigation stages. It uses the global search ability of genetic algorithms to optimize the reinforcement learning model and real-time plan the optimal reference trajectory that takes into account navigation safety, efficiency, and energy consumption. At the same time, considering the collision avoidance requirements of the ship with surrounding obstacles and other ships in complex environments, it ensures the feasibility and reliability of the trajectory.

[0020] The middle layer interference estimation layer adopts an improved adaptive extended state observer (AESO-AL-AKF) that combines an attention mechanism long short-term memory network (Attention-LSTM) and an adaptive Kalman filter. Through multi-source sensor fusion technology, it comprehensively perceives and analyzes the dynamic changes of the ship's six-degree-of-freedom motion parameters, sensor measurement noise, and marine environment parameters (such as waves, sea winds, currents, etc.). The Attention-LSTM network can automatically focus on the key features in the ship's motion state information, and combine the adaptive Kalman filter to correct and estimate the measurement data in real time, so as to accurately capture the complex time-varying characteristics of the external interference suffered by the ship, as well as the uncertain parameter changes of the ship itself due to factors such as loading state and equipment wear, such as the high-order nonlinear interference force of waves, the sea wind interference torque that changes with time and space, and the dynamic drift of the ship's hydrodynamic coefficients.

[0021] The inner layer control execution layer is based on a deep fusion strategy of model predictive control (MPC), adaptive sliding mode control (ASMC), and adaptive fuzzy control (AFC). According to the reference trajectory and the accurate interference estimation results output by the middle layer interference estimation layer, MPC performs multi-step prediction and optimization of the ship's motion state within a finite future time. ASMC uses its fast response and strong robustness to adjust the control input in real time to resist interference. AFC dynamically adjusts the control rules and parameters according to the ship's real-time motion state, interference characteristics, and the real-time loading state and sea condition changes of the ship, realizing refined and intelligent control of the ship's propulsion system and steering gear. For example, according to the interference intensity and direction in different sea conditions, it intelligently adjusts the dynamic response range of the rudder angle and the distribution ratio of the thruster thrust to ensure that the ship accurately tracks the reference trajectory.

[0022] Interference Estimation and Compensation: In the middle-level interference estimation stage, AESO-AL-AKF uses multi-source data fusion and deep neural network feature extraction capabilities to accurately estimate the magnitude, direction, frequency characteristics, and dynamic evolution trend of the interference. In the inner control execution layer, the estimated interference is used as a feedforward compensation term and combined with the control signal generated based on the MPC-ASMC-AFC fusion strategy to achieve real-time dynamic compensation for the interference. At the same time, the boundary layer design of adaptive sliding mode control effectively reduces control vibration, improves the ship's navigation stability and control accuracy in interference environments, and ensures the ship's safe navigation and path tracking accuracy even in extreme sea conditions (such as hurricanes and huge waves).

[0023] Adaptive Control Strategy: To address the complex and volatile dynamic characteristics of a vessel, the MPC-ASMC-AFC fusion strategy at the inner control execution layer monitors the vessel's motion, interference estimation, loading status, and sea conditions in real time. It leverages the self-adjusting capabilities of adaptive fuzzy control rules to automatically optimize control parameters and rules online. The MPC rolling optimization mechanism dynamically plans and adjusts the vessel's future motion. Combined with the rapid response of ASMC, this ensures timely and effective control, enabling precise and efficient control of the vessel under varying operating conditions. This adapts to the vessel's navigation needs under various complex conditions without requiring human intervention.

[0024] In summary, the present invention has the following beneficial effects:

[0025] Super strong anti-interference ability: With its unique layered structure, advanced multi-source data fusion interference estimation technology and efficient interference compensation mechanism, the present invention can effectively resist complex and changeable external interference such as waves, sea breezes, water currents, etc. In harsh sea conditions, the path tracking error is reduced by more than 80% compared with traditional methods, significantly improving the path tracking accuracy and navigation safety of ships in complex marine environments.

[0026] Highly adaptive to ship dynamic characteristics: The adaptive control strategy can automatically adapt to the complex changes in the ship's own dynamic characteristics in real time, including changes in ship performance under different loading states, different navigation phases, and equipment aging and failure conditions. It greatly improves the reliability and stability of control and ensures that the ship can sail safely, stably and efficiently under various operating conditions.

[0027] Significantly improve control accuracy and reliability: The present invention achieves an order-of-magnitude improvement in the path tracking accuracy of autonomous towing vessels, greatly enhancing the accuracy and safety of towing tasks and reducing the risk of accidents such as ship collisions and groundings. At the same time, this method has high generality and scalability, can be widely applied to vessels of different types and tonnages, has extremely high engineering application value, leads the innovative development of autonomous towing vessel technology, and provides core technical support for the intelligent and safe development of the global marine transportation industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Other features, objects, and advantages of the present invention will become more apparent by reading the detailed description of non-limiting embodiments with reference to the following drawings:

[0029] Figure 1 It is a flow framework diagram of a hierarchical disturbance rejection control method for path tracking of an autonomous towing vessel according to an embodiment of the present invention;

[0030] Figure 2 It is a structural framework diagram of a hierarchical disturbance rejection control system according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0031] To make the objectives, features, and advantages of the present application more obvious and understandable, the technical solutions in the embodiments of the present application are described clearly and completely. Obviously, the embodiments described below are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0032] Referring to Figure 1 and Figure 2 , a hierarchical disturbance rejection control method for path tracking of an autonomous towing vessel, by constructing a hierarchical disturbance rejection control system, includes:

[0033] The outer layer trajectory planning layer uses a dynamic path planning algorithm based on reinforcement learning and multi-agent collaborative optimization to integrate the starting position, target position, real-time navigation environment information of the vessel, and performance constraints in different navigation stages of the vessel, and introduces a multi-agent information interaction and cooperation mechanism;

[0034] The middle layer disturbance estimation layer uses an improved adaptive extended state observer combined with a long short-term memory network and a quantum particle swarm optimization algorithm to perform multi-dimensional measurements on the six-degree-of-freedom motion parameters of the vessel, sensor measurement noise, and environmental parameter changes;

[0035] The inner layer control execution layer is based on a fusion strategy of model predictive control MPC, adaptive sliding mode control ASMC, and fuzzy neural network control FNC, and operates according to the reference trajectory and disturbance estimation results.

[0036] Specifically, in the solution of this embodiment, the navigation environment information includes wave height, wave period, water flow velocity and direction, sea wind intensity and direction.

[0037] Specifically, in the solution of this embodiment, the external disturbances include high-order components of wave interference forces and sea wind interference moments that change with time, and the self-uncertainty factors include model parameter perturbations of the ship under different working conditions.

[0038] Specifically, in the solution of this embodiment, the fusion strategy operates based on the real-time motion state of the ship, the interference estimation results, as well as the real-time loading state of the ship and the changes in sea conditions.

[0039] Specifically, in the solution of this embodiment, when the ship is sailing, the ship position, speed, acceleration, heading angle, roll angle, and pitch angle information are collected in real time as the input of the middle-layer interference estimation layer.

[0040] Specifically, in the solution of this embodiment, when the ship deviates significantly from the reference trajectory, the outer-layer trajectory planning layer uses a dynamic path planning algorithm that combines reinforcement learning and multi-agent collaborative optimization to re-plan the reference trajectory.

[0041] Specifically, in the solution of this embodiment, when the interference situation changes greatly, the inner-layer control execution layer adjusts the interference compensation strategy and control parameters.

[0042] Specifically, in the solution of this embodiment, in the initialization stage, the ship mass, moment of inertia, and detailed self-parameters of hydrodynamic coefficients under different loading states are input.

[0043] Specifically, in the solution of this embodiment, the control inputs calculated by the inner-layer control execution layer include dynamic adjustment of the rudder angle and optimal distribution of the thruster thrust.

[0044] Specifically, in the solution of this embodiment, in the ship formation towing scenario, multi-agent collaborative optimization algorithm is used to achieve collaborative path planning and interference coordination control among ships.

[0045] It should be noted that: before the ship sails, a comprehensive initialization of the entire hierarchical disturbance rejection control system is carried out. First, detailed ship information is input to the outer-layer trajectory planning layer, including the precise geometric dimensions of the ship, power system parameters (such as engine power, thruster efficiency curve), the center of gravity position and moment of inertia data under different loading states. At the same time, high-precision electronic nautical chart data of the navigation area are entered, covering information such as water depth, reef distribution, restricted navigation areas, etc., as well as real-time updated marine environment data, such as wave height, wavelength, period of waves, wind speed, wind direction of sea wind, flow velocity, flow direction of water flow, etc.

[0046] Using these data, through the fusion algorithm of deep reinforcement learning and genetic algorithm, the initial path is planned. In the reinforcement learning model, appropriate state space s, action space a, and reward function R are defined. The state space s = [x, y, θ, v x , v y , ω, cnv], where x and y are the current position coordinates of the ship, θ is the heading angle, v x 、v y are the velocity components of the ship in the x and y directions, ω is the angular velocity, and env is the surrounding environment information; the action space a = [Δθ, ΔT], where Δθ is the steering angle adjustment amount of the ship, and ΔT is the thruster thrust adjustment amount. The reward function R comprehensively considers factors such as navigation safety, efficiency, and energy consumption. For example, it can be expressed as:

[0047]

[0048] where d is the distance between the current position of the ship and the target position, fuel is the fuel consumption per unit time, risk is the risk assessment value of the ship approaching dangerous areas (such as reefs, restricted navigation areas), and α, β, γ are weight coefficients, which are adjusted according to actual needs. The genetic algorithm is used to optimize the parameters of the reinforcement learning model. Through operations such as population initialization, selection, crossover, and mutation, it continuously iterates to find the optimal path planning strategy.

[0049] Estimation of interference during navigation

[0050] During the ship's navigation, the middle-layer interference estimation layer starts to work. The multi-source sensors continuously collect the six-degree-of-freedom motion parameters of the ship, including the translational displacements and rotational angles along the x, y, and z axes, as well as the sensor measurement noise data. At the same time, the real-time changes of the marine environment parameters are obtained through meteorological sensors, hydrological sensors, etc.

[0051] These data are input into the improved adaptive extended state observer (AESO-AL-AKF) that combines the attention mechanism long short-term memory network (Attention-LSTM) and adaptive Kalman filter. Assuming the motion state equation of the ship is x k = Fx k-1 + Gu k-1 + w k-1 +, and the observation equation is z k = Hx k + v k , where x k is the state vector at time k, F is the state transition matrix, G is the control input matrix, u k-1 is the control input at time k - 1, w k-1 is the process noise, z k is the observation vector at time k, H is the observation matrix, and vk is the observation noise. Adaptive Kalman filtering adjusts the filtering parameters in real time according to the statistical characteristics of the measurement data. Its core steps include prediction and update:

[0052] Prediction:

[0053]

[0054] P k|k-1 = FP k-1|k-1 F T + Q k-1

[0055] Update:

[0056] K k = P k|k-1 H T (HP k|k-1 H T + R k ) -1

[0057]

[0058] P k|k = (I - K k H)P k|k-1

[0059] Among them, is the predicted state at time k based on the state at time k - 1, is the updated state at time k, P k|k-1 is the predicted covariance, P k|k is the updated covariance, K k is the Kalman gain, Q k-1 is the process noise covariance, R k is the observation noise covariance, and I is the identity matrix. The Attention-LSTM network extracts features from the input data and automatically focuses on key information, such as the abnormal motion characteristics of the ship when encountering strong sea waves. When the movement of the loaded cargo on the ship causes a change in the center of gravity, AESO-AL-AKF can quickly estimate the resulting changes in the ship's dynamic parameters.

[0060] Control Execution and Adjustment

[0061] The inner control execution layer adopts a deep fusion strategy of model predictive control (MPC), adaptive sliding mode control (ASMC), and adaptive fuzzy control (AFC) according to the reference trajectory and the interference estimation result output by the middle-layer interference estimation layer for control.

[0062] MPC predicts the ship's motion state in multiple future time steps based on the current state of the ship and the interference estimation. Assume the prediction time domain is Np , the control time domain is N c , construct an optimization problem including the ship dynamics model, control input constraints and state constraints, and the objective function can be expressed as:

[0063]

[0064] Among them, x k|k is the predicted state at time k, is the reference state at time k, Q is the state weight matrix, and R is the control input weight matrix. Solving this optimization problem yields the optimal control input sequence—the rudder angle and thrust adjustment values for multiple future time steps. However, given that actual control can only execute the control inputs at the current moment, a rolling optimization strategy is employed, re-predicting and optimizing after each control cycle.

[0065] ASMC adjusts the control input in real time to resist the disturbance based on the disturbance estimation results. The sliding surface s = Cx is designed, where C is the sliding surface coefficient matrix and x is the state vector. The reaching law adopts the exponential reaching law s = -ε sgn (s)-ηs, where ε and η are positive numbers and sgn(s) is the sign function. When encountering strong sea wind disturbance, ASMC can quickly adjust the rudder angle to keep the ship on the intended course.

[0066] AFC dynamically adjusts control rules and parameters according to the ship's real-time motion state, interference characteristics, ship's real-time loading state and sea conditions. A fuzzy rule base is established to convert the ship's motion deviation e, deviation change rate Using input variables such as the disturbance intensity d as input, the system undergoes fuzzification, fuzzy inference, and defuzzification to generate adaptively adjusted control parameters, enabling refined control of the ship's propulsion system and steering gear. For example, under varying sea conditions, the system intelligently adjusts the rudder angle response gain and propeller thrust distribution ratio based on wave height and period.

[0067] Throughout the voyage, the vessel's position, speed, heading, and other status information are monitored in real time and compared with a reference trajectory. If the vessel deviates from the reference trajectory by more than a set threshold, the outer trajectory planning layer immediately replans the path using the latest environmental information and vessel status, ensuring the vessel remains safely and efficiently on target.

[0068] Example:

[0069] Implementation Case: A container ship with a deadweight of 5,000 tons plans to sail from Shanghai Port to Singapore Port. In the initialization stage, the operator enters the detailed parameters of the ship, such as a length of 180 meters, a beam of 25 meters, a main engine power of 8,000 kilowatts, and the data of the center of gravity position under different loading conditions into the system. At the same time, the system obtains the electronic nautical chart of the navigation area, showing a reef area nearby, and real-time marine environment data, such as encountering waves 3 - 4 meters high, a wind speed of 15 knots, and a flow direction of northeast. Through the fusion of deep reinforcement learning and genetic algorithm, the system plans an initial path that avoids the reef area and can reasonably adjust the course and speed according to the wave and current conditions, enabling the ship to save fuel and time as much as possible while ensuring safety.

[0070] Before the ship sails, a comprehensive initialization of the entire hierarchical disturbance rejection control system is carried out. First, detailed ship information is input into the outer trajectory planning layer, including the precise geometric dimensions of the ship, the parameters of the power system (such as engine power, propeller efficiency curve), the center of gravity position and moment of inertia data under different loading conditions. At the same time, high-precision electronic nautical chart data of the navigation area is entered, covering information such as water depth, reef distribution, restricted navigation areas, etc., as well as real-time updated marine environment data, such as wave height, wavelength, period of waves, wind speed, wind direction of sea breeze, flow velocity, flow direction of water flow, etc.

[0071] Using these data, an initial path is planned through the fusion algorithm of deep reinforcement learning and genetic algorithm. In the reinforcement learning model, a suitable state space s, action space a, and reward function R are defined. The state space s = [x, y, θ, v x , v y , ω, cnv], where x and y are the current position coordinates of the ship, θ is the course angle, v x , v y are the velocity components of the ship in the x and y directions, ω is the angular velocity, and cnv is the surrounding environment information; the action space a = [Δθ, ΔT], Δθ is the adjustment amount of the ship's turning angle, and ΔT is the adjustment amount of the propeller thrust. The reward function R comprehensively considers factors such as navigation safety, efficiency, and energy consumption. For example, it can be expressed as:

[0072]

[0073] where d is the distance between the current position of the ship and the target position, fucl is the fuel consumption per unit time, risk is the risk assessment value of the ship approaching dangerous areas (such as reefs, restricted navigation areas), and α, β, γ are weight coefficients, which are adjusted according to actual needs. The genetic algorithm is used to optimize the parameters of the reinforcement learning model. Through operations such as population initialization, selection, crossover, and mutation, it continuously iterates to find the optimal path planning strategy.

[0074] Disturbance Estimation during Navigation

[0075] Implementation case: When the above container ship encounters huge waves of 5 meters high and strong winds. Multi-source sensors continuously collect the motion data of the ship, such as the displacement change of the ship in the x direction, the speed fluctuation in the y direction, the changes in the roll angle and pitch angle, etc., as well as the sensor measurement noise. AESO-AL-AKF adjusts the parameters in real time through adaptive Kalman filtering according to these data, and accurately estimates the disturbing forces and disturbing torques generated by the sea waves and sea winds on the ship. The Attention-LSTM network focuses on the abnormal roll and pitch motion characteristics of the ship under the impact of huge waves, and quickly identifies the complex interference patterns suffered by the ship. At the same time, due to the slight displacement of some goods on the ship during navigation, resulting in a change in the center of gravity, AESO-AL-AKF timely estimates the changes in the ship's dynamic parameters, providing an accurate basis for subsequent control adjustment.

[0076] During the ship's navigation, the intermediate interference estimation layer starts to work. Multi-source sensors continuously collect the six-degree-of-freedom motion parameters of the ship, including the translational displacements and rotational angles along the x, y, and z axes, as well as the sensor measurement noise data. At the same time, the real-time changes in ocean environmental parameters are obtained through meteorological sensors, hydrological sensors, etc.

[0077] These data are input into an improved adaptive extended state observer (AESO-AL-AKF) that combines an attention mechanism long short-term memory network (Attention-LSTM) with adaptive Kalman filtering. Assume that the motion state equation of the ship is x k = Fx k-1 + Gu k-1 + w k-1 +, and the observation equation is z k = Hx k + v k , where x k is the state vector at time k, F is the state transition matrix, G is the control input matrix, u k-1 is the control input at time k-1, w k-1 is the process noise, z k is the observation vector at time k, H is the observation matrix, and v k is the observation noise. Adaptive Kalman filtering adjusts the filtering parameters in real time according to the statistical characteristics of the measurement data, and its core steps include prediction and update:

[0078] Prediction:

[0079]

[0080] P k|k-1 = FP k-1|k-1 F T + Q k-1

[0081] Update:

[0082] K k = P k|k-1 H T (HP k|k-1 H T + R k ) -1

[0083]

[0084] P k|k = (I - K k H)P k|k-1

[0085] Wherein, is the predicted state at time k based on the state at time k - 1, is the updated state at time k, P k|k-1 is the predicted covariance, P k|k is the updated covariance, K k is the Kalman gain, Q k-1 is the process noise covariance, R k is the observation noise covariance, and I is the identity matrix. The Attention-LSTM network extracts features from the input data and automatically focuses on key information, such as the abnormal motion characteristics of the ship when encountering strong sea waves. When the cargo on the ship moves and causes a change in the center of gravity, AESO-AL-AKF can quickly estimate the resulting changes in the ship's dynamic parameters.

[0086] Control Execution and Adjustment

[0087] Implementation Case: In severe sea conditions, MPC predicts the ship's motion state for the next 10 time steps (N p = 10) according to the interference result estimated by AESO-AL-AKF, and the control time domain is set to 5 time steps (N c(= 5). By solving the optimization problem, the adjustment values of the rudder angle and thruster thrust for the next 5 time steps are calculated to resist the interference of sea waves and sea winds and maintain the stable navigation of the ship. ASMC quickly adjusts the rudder angle according to the interference estimation result. When the ship's heading deviates due to strong sea wind interference, the ship's state is quickly adjusted back to the sliding mode surface through the exponential reaching law to maintain the predetermined heading. AFC dynamically adjusts the control rules according to the motion deviation, deviation change rate and interference intensity of the ship in huge waves. For example, when the wave height reaches 5 meters and the period is short, AFC automatically increases the response gain of the rudder angle, enabling the ship to adjust its heading more sensitively, while reasonably distributing the thruster thrust to ensure the ship sails steadily in the wind and waves. During the voyage, if the ship deviates from the reference trajectory by more than the set threshold of 50 meters due to sudden strong currents, the outer trajectory planning layer immediately uses the latest sea current information and ship state to re-plan the path and guide the ship back to a safe and efficient navigation route. Finally, the container ship successfully arrives at the Port of Singapore, verifying the effectiveness and reliability of this control method.

[0088] The inner control execution layer adopts a deep fusion strategy of model predictive control (MPC), adaptive sliding mode control (ASMC) and adaptive fuzzy control (AFC) according to the reference trajectory and the interference estimation result output by the middle layer interference estimation layer for control.

[0089] MPC predicts the ship's motion state for multiple future time steps based on the ship's current state and interference estimation. Assume the prediction horizon is N p , and the control horizon is N c , and an optimization problem including the ship's dynamics model, control input constraints and state constraints is constructed. The objective function can be expressed as:

[0090]

[0091] where, x k|k is the predicted state at time step k, is the reference state at time step k, Q is the state weight matrix, and R is the control input weight matrix. By solving this optimization problem, the optimal control input sequence, that is, the adjustment values of the rudder angle and thruster thrust for multiple future time steps, is obtained. However, considering that only the control input at the current moment can be executed in actual control, a rolling optimization strategy is adopted, and prediction and optimization are re-performed every control cycle.

[0092] ASMC adjusts the control input in real time according to the interference estimation result to resist interference. The sliding mode surface is designed as s = Cx, where C is the sliding mode surface coefficient matrix and x is the state vector. The reaching law adopts the exponential reaching law s = -ε sgn(s)-ηs, where ε and η are positive numbers and sgn(s) is the sign function. When encountering strong sea wind disturbance, ASMC can quickly adjust the rudder angle to keep the ship on the intended course.

[0093] AFC dynamically adjusts control rules and parameters according to the ship's real-time motion state, interference characteristics, ship's real-time loading state and sea conditions. A fuzzy rule base is established to convert the ship's motion deviation e, deviation change rate Using input variables such as the disturbance intensity d as input, the system undergoes fuzzification, fuzzy inference, and defuzzification to generate adaptively adjusted control parameters, enabling refined control of the ship's propulsion system and steering gear. For example, under varying sea conditions, the system intelligently adjusts the rudder angle response gain and propeller thrust distribution ratio based on wave height and period.

[0094] Throughout the voyage, the vessel's position, speed, heading, and other status information are monitored in real time and compared with a reference trajectory. If the vessel deviates from the reference trajectory by more than a set threshold, the outer trajectory planning layer immediately replans the path using the latest environmental information and vessel status, ensuring the vessel remains safely and efficiently on target.

[0095] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0096] Super strong anti-interference ability: With its unique layered structure, advanced multi-source data fusion interference estimation technology and efficient interference compensation mechanism, the present invention can effectively resist complex and changeable external interference such as waves, sea breezes, water currents, etc. In harsh sea conditions, the path tracking error is reduced by more than 80% compared with traditional methods, significantly improving the path tracking accuracy and navigation safety of ships in complex marine environments.

[0097] Highly adaptive to ship dynamic characteristics: The adaptive control strategy can automatically adapt to the complex changes in the ship's own dynamic characteristics in real time, including changes in ship performance under different loading states, different navigation phases, and equipment aging and failure conditions. It greatly improves the reliability and stability of control and ensures that the ship can sail safely, stably and efficiently under various operating conditions.

[0098] Significantly Improved Control Precision and Reliability: This method achieves an order-of-magnitude improvement in the path tracking accuracy of autonomous towing vessels, significantly enhancing the accuracy and safety of towing missions and reducing the risk of accidents such as collisions and groundings. Furthermore, this method is highly versatile and scalable, applicable to a wide range of vessels of different types and tonnages. It possesses significant engineering application value, leading the innovative development of autonomous towing vessel technology and providing core technical support for the intelligent and secure development of the global maritime transport industry.

[0099] The above has introduced in detail the technical solutions provided by the embodiments of the present invention. Specific examples are used herein to elaborate on the principles and implementation manners of the embodiments of the present invention. The description of the above embodiments is only applicable to helping understand the principles of the embodiments of the present invention. At the same time, for those of ordinary skill in the art, according to the embodiments of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation on the present invention.

Claims

1. A hierarchical disturbance rejection control method for path tracking of an autonomous towing ship, characterized in that Construct a hierarchical anti-disturbance control system, including: The outer layer trajectory planning layer uses a dynamic path planning algorithm based on reinforcement learning and multi-agent collaborative optimization to integrate the starting position of the ship, the target position, the real-time navigation environment information, and the performance constraints of different navigation stages of the ship, and introduces a multi-agent information interaction and collaboration mechanism; The middle layer disturbance estimation layer uses an adaptive extended state observer improved by combining a long short-term memory network and a quantum particle swarm optimization algorithm to perform multi-dimensional measurements on the six-degree-of-freedom motion parameters of the ship, the sensor measurement noise, and the changes in environmental parameters; The inner layer control execution layer operates based on a fusion strategy of model predictive control, adaptive sliding mode control, and fuzzy neural network control, and operates according to the reference trajectory and the disturbance estimation result.

2. The hierarchical disturbance rejection control method for autonomous towing ship path tracking according to claim 1, characterized in that, The navigation environment information includes wave height, wave period, water flow velocity and direction, sea breeze intensity and direction.

3. The hierarchical disturbance rejection control method for autonomous towing ship path tracking according to claim 1, characterized in that: The external disturbances include the high-order components of the wave disturbance force and the sea breeze disturbance torque that changes with time, and the self-uncertainty factors include the model parameter perturbation of the ship under different working conditions.

4. A hierarchical disturbance rejection control method for autonomous towing ship path tracking according to claim 1, characterized in that: The fusion strategy operates based on the real-time motion state of the ship, the disturbance estimation result, and the real-time loading state of the ship and the changes in sea conditions.

5. The hierarchical disturbance rejection control method for autonomous towing ship path tracking according to claim 1, characterized in that: When the ship is sailing, the ship position, speed, acceleration, heading angle, roll angle, and pitch angle information are collected in real time as the input of the middle layer disturbance estimation layer.

6. The hierarchical disturbance rejection control method for autonomous towing ship path tracking according to claim 1, characterized in that: When the ship deviates significantly from the reference trajectory, the outer layer trajectory planning layer uses a dynamic path planning algorithm based on reinforcement learning and multi-agent collaborative optimization to re-plan the reference trajectory.

7. A hierarchical disturbance rejection control method for autonomous towing ship path tracking according to claim 1, characterized in that: When the disturbance situation changes greatly, the inner layer control execution layer adjusts the disturbance compensation strategy and control parameters.

8. The hierarchical disturbance rejection control method for autonomous tugboat path tracking according to claim 1, characterized in that: In the initialization stage, the detailed parameters of the ship's own mass, moment of inertia, and hydrodynamic coefficients under different loading states are input.

9. The hierarchical disturbance rejection control method for autonomous towing ship path tracking according to any one of claims 1-8, characterized in that: The control inputs calculated by the inner layer control execution layer include the dynamic adjustment of the rudder angle and the optimal distribution of the thruster thrust.

10. A hierarchical disturbance rejection control method for autonomous towing ship path tracking according to claim 1, characterized in that: In the ship formation towing scenario, the multi-agent collaborative optimization algorithm is used to achieve collaborative path planning and disturbance coordination control among ships.

Images