An intelligent ship control system

Abstract

The application discloses an intelligent ship control system, which comprises a ship interactive data acquisition module, a ship state refining module, a navigation experience screening module, a multi-ship cooperative control intelligent agent training module, a ship parameter self-adaptive adjustment module and a ship control module. The application belongs to the field of intelligent control and specifically relates to an intelligent ship control system. According to the scheme, core features are extracted through double-branch refining, a dynamic transfer probability similarity coefficient is introduced, and similarity judgment is adapted to interactive intensity. For a positive sample pair, the low-dimensional features of ships with similar behaviors are close, and for a negative sample pair, the feature distance is enlarged, so that the collision avoidance strategy can be transferred across ship types. Safe sampling guided by a reference path is used to screen navigation experience, and dangerous data interference is avoided. An extreme risk penalty is added to strengthen the fitting of close-range collision samples. A dynamic cycle of environment perception is introduced, a post-adjustment safety reward is set, environment changes are dynamically adapted, and the ship control effect is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent control, specifically to an intelligent ship control system. Background Technology

[0002] Generally, ship control systems collect information about the ship's own state and the marine environment through sensors, process it using control algorithms, and then output commands to actuators such as the steering gear and throttle to achieve precise control of the ship's motion. However, typical ship control systems suffer from several drawbacks: ignoring the dynamic nature of environmental disturbances, using static measurements of the similarity of behavior among multiple ships, weak migration capabilities, and insufficient anti-interference capabilities. These issues exacerbate energy consumption and course instability, leading to low ship control reliability. Furthermore, typical ship control systems suffer from blindly relying on experience-based selection, low utilization of high-value data, singular goals in multi-ship collaboration, imbalance between individual and group needs, rigid parameter adjustments, and difficulty adapting to dynamic environments, ultimately resulting in poor ship control performance. Summary of the Invention

[0003] To address the aforementioned issues and overcome the shortcomings of existing technologies, this invention provides an intelligent ship control system. Addressing the problems of general ship control systems, such as ignoring the dynamic nature of environmental interference, static multi-ship behavior similarity measurement, weak transferability, and insufficient anti-interference capabilities, which exacerbate energy consumption and course instability, leading to low ship control reliability, this solution extracts core features through a dual-branch extraction process. It introduces a dynamic transition probability similarity coefficient into the dynamic behavior similarity measurement, enabling similarity judgments to adapt to interaction intensity and scenario type. By dividing ships into positive and negative sample pairs and setting loss functions for each, for positive sample pairs, the low-dimensional features of ships with similar behaviors are forced to be close; for negative sample pairs, the feature distance is increased, thereby enabling collision avoidance strategies to be transferred across ship types. Through a three-layer design of physical modeling, dynamic measurement, and full data, the ship control strategy is adaptable to complex sea conditions and multi-ship interaction scenarios. This approach possesses physical feasibility, scenario adaptability, and cross-ship type mobility, thereby improving ship control reliability. Addressing the issues of blind experience-based selection, low utilization of high-value data, singular objectives in multi-ship collaboration, imbalance between individual and group needs, rigid parameter adjustments, and difficulty adapting to dynamic environments—leading to poor ship control performance—this solution employs reference path-guided safety sampling to screen navigation experience and avoid interference from hazardous data. Simultaneously, it constrains path tracking accuracy, energy consumption, and collision avoidance safety, guiding multi-objective optimization of the main network. It avoids value overlap by learning the Q-value separation and risk penalties of sub-networks; it enhances the fitting of close-range collision samples by incorporating extreme risk penalties; and it dynamically adapts to environmental changes through hierarchical parameter adjustments, introducing a dynamic cycle of environmental perception and setting safety rewards after adjustments, thus resolving strategy rigidity and safety risk issues, ultimately improving ship control performance.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides an intelligent ship control system, including a ship interactive data acquisition module, a ship status extraction module, a navigation experience screening module, a multi-ship collaborative control intelligent agent training module, a ship parameter adaptive adjustment module, and a ship control module;

[0005] The ship interactive data acquisition module acquires the ship's own status data and surrounding ship observation data to build a ship navigation experience repository.

[0006] The ship status extraction module optimizes the features of samples from the ship navigation experience repository through dual-branch training.

[0007] The navigation experience screening module defines a value extraction vector for ship route segments and filters samples from the ship navigation experience repository.

[0008] The multi-ship cooperative control agent training module is based on a filtered and optimized ship navigation experience repository to construct a ship guidance main network and a control learning sub-network.

[0009] The ship parameter adaptive adjustment module controls the network parameters into a core security layer and a non-core optimization layer, and calculates the dynamic adjustment cycle according to the intensity of environmental interference.

[0010] The ship control module acquires real-time ship operation data, inputs it into the ship status refinement module for optimization, and then inputs it into the multi-ship collaborative control intelligent agent to achieve real-time ship control.

[0011] Furthermore, the ship interaction data acquisition module acquires the interaction data of the ship in the digital twin environment. Each ship collects its own status and the observation status of surrounding ships in real time through sensors; after executing control actions, it records the next status and reward; and stores the joint data of multiple ships in the ship navigation experience repository.

[0012] Furthermore, the ship condition refining module specifically includes:

[0013] Branch 1, State Extraction Training Based on Ship Dynamics: Based on the ship's MMG model, the initial ship state is obtained by predicting the next state and environmental disturbances from the current state and actions; loss function... for: ;in, This is the predicted next state of the ship. The superscript te indicates the actual value, the superscript pd indicates the predicted value, t indicates the time, and i indicates the ship. These are predicted water flow forces and wind forces, where 'c' represents the water flow force. Indicates wind force; It is the weight of the equilibrium state prediction and the disturbance force prediction;

[0014] Branch 2, State Extraction Training Based on Ship Mutual Simulation: This measures the behavioral similarity between two ships and introduces a dynamic transition probability similarity coefficient; expressed as: ; ;in, is a measure of the behavioral similarity between ship i and ship j; the smaller the value, the more similar the behaviors. 'a' is the control action, and 'A' is the set of control actions. and It is the reward obtained by ships i and j for performing action a; In action a, water flow Wind speed Under these conditions, ship i transitions to state i. The probability of; The ship j is transferred to state The probability of; It is the transition probability similarity coefficient; It is the benchmark coefficient; This is a reference distance; It is the distance between the centers of mass of the two ships; It is the JS divergence; It is the scene correction coefficient; learn a behavior feature encoder to map the initial ship extraction to low-dimensional behavior features, so that ships with similar behaviors have similar low-dimensional behavior features, and divide ship data into positive and negative sample pairs by behavior threshold; define the core loss for positive sample pairs; define the contrast loss for negative sample pairs; sample an equal number of positive and negative sample pairs in each iteration, update the behavior feature encoder parameters with Adam optimizer, and iterate until the loss converges.

[0015] Furthermore, the navigation experience screening module specifically includes:

[0016] The definition of ship path segment extraction is that extraction is the difference between the initial state extraction and the final state extraction;

[0017] Similarity sampling guided by reference path segments is used to randomly select one path segment from the experience storage pool; the similarity between all candidate path segments and the selected path segment is calculated, and path segments with similarity ∈ [0.3, 0.7] are filtered; if the number of filtered path segments is greater than the batch size, the top K path segments are retained in descending order of reward value, and each path segment is split into single-step training samples.

[0018] Furthermore, the multi-ship cooperative control agent training module specifically includes:

[0019] The ship guidance master network is constructed with path tracking under multi-ship collision avoidance constraints as the objective, and an optimization objective function is constructed.

[0020] The ship control learning subnetwork is constructed, and its Q-value function is divided into state value and action advantage. The state value corresponds to the overall value of the marine environment, and the action advantage corresponds to the local value of the control action. Both the state value and the action advantage are obtained by fitting through a neural network.

[0021] The loss of the ship control learning subnetwork is subject to an extreme risk penalty.

[0022] Furthermore, the ship parameter adaptive adjustment module specifically includes:

[0023] Some parameters were adjusted, dividing the ship control network parameters into a core safety layer and a non-core optimization layer, and adjustments were only made to the non-core layer.

[0024] A dynamic adjustment cycle mechanism based on environmental perception is introduced; a safety reward is obtained after the adjustment.

[0025] Furthermore, the ship control module acquires real-time ship operation data, inputs it into the ship status refinement module for optimization, and inputs it into the multi-ship collaborative control intelligent agent module to realize real-time ship control based on the module output.

[0026] The beneficial effects achieved by the present invention using the above solution are as follows:

[0027] (1) In view of the problems that general ship control systems ignore the dynamics of environmental interference, the static measurement of the similarity of multi-ship behavior, the weak transferability and the insufficient anti-interference capability, which in turn aggravate energy consumption and course instability, resulting in low ship control reliability, this solution extracts the core features through dual-branch extraction and introduces the dynamic transfer probability similarity coefficient in the dynamic behavior similarity measurement, so that the similarity judgment is adapted to the interaction intensity and scenario type; by dividing the ship into positive and negative sample pairs and setting loss functions respectively, for positive sample pairs, the low-dimensional features of ships with similar behaviors are forced to be close, and for negative sample pairs, the feature distance is widened, so as to realize the collision avoidance strategy can be transferred across ship types; through the three-layer design of physical modeling-dynamic measurement-full data, the ship control strategy has physical feasibility, scenario adaptability and cross-ship type transferability in complex sea conditions and multi-ship interaction scenarios, thereby improving the reliability of ship control.

[0028] (2) To address the problems of blind experience screening, low utilization of high-value data, singular objectives of multi-ship collaboration, imbalance between individual and group needs, rigid parameter adjustment, and difficulty in adapting to dynamic environments in general ship control systems, which lead to poor ship control performance, this scheme uses safety sampling guided by reference paths to screen navigation experience and avoid interference from dangerous data; at the same time, it constrains path tracking accuracy, energy consumption and collision avoidance safety to guide multi-objective optimization of the main network; it avoids value overlap by learning the Q-value separation and risk penalty of the sub-network; it strengthens the fitting of close-range collision samples by adding extreme risk penalties; it introduces the dynamic cycle of environmental perception through hierarchical parameter adjustment, sets safety rewards after adjustment, dynamically adapts to environmental changes, and solves the problems of rigid strategies and safety risks; thereby improving the ship control performance. Attached Figure Description

[0029] Figure 1 This is a flowchart illustrating an intelligent ship control system provided by the present invention.

[0030] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0031] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0032] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0033] Example 1, see Figure 1 The present invention provides an intelligent ship control system, including a ship interactive data acquisition module, a ship status extraction module, a navigation experience screening module, a multi-ship collaborative control agent training module, a ship parameter adaptive adjustment module, and a ship control module.

[0034] The ship interactive data acquisition module acquires the ship's own status data and surrounding ship observation data to build a ship navigation experience repository; and sends the data to the ship status extraction module.

[0035] The ship status extraction module optimizes the features of samples in the ship navigation experience repository through dual-branch training and sends the data to the navigation experience screening module.

[0036] The navigation experience screening module defines a value extraction vector for ship path segments, filters samples from the ship navigation experience repository, and sends the data to the multi-ship collaborative control agent training module.

[0037] The multi-ship collaborative control agent training module constructs a ship guidance main network and a control learning sub-network based on a filtered and optimized ship navigation experience repository; and sends the data to the ship parameter adaptive adjustment module.

[0038] The ship parameter adaptive adjustment module divides the control network parameters into a core security layer and a non-core optimization layer, calculates the dynamic adjustment cycle according to the intensity of environmental interference, and sends the data to the ship control module.

[0039] The ship control module acquires real-time ship operation data, inputs it into the ship status refinement module for optimization, and then inputs it into the multi-ship collaborative control intelligent agent to achieve real-time ship control.

[0040] Example 2, see Figure 1 This embodiment is based on the above embodiment. The ship interaction data acquisition module acquires the ship's interaction data in the digital twin environment, covering different sea conditions and scenarios.

[0041] The specific operation is as follows: Each ship collects its own status in real time through sensors. The surrounding ship observation vector is ; These are the ship's position coordinates; It refers to the ship's speed; It is the ship's heading angle; It is the draft; It is the heading of ship j relative to ship i. It is relative speed; It is the distance between the centers of gravity of the two ships; executing control actions. Then, record the next state. With rewards Combined data from multiple ships Stored in the ship navigation experience repository; , and These are the changes in rudder angle, throttle opening, and propeller speed, respectively. , , These are the ship's joint state, joint actions, and joint rewards at time t; This represents the combined ship status at time t+1; the data needs to be normalized.

[0042] Example 3, see Figure 1 This embodiment is based on the above embodiment. The ship status extraction module has strong coupling with the ship status, including the impact of speed changes on course stability, and environmental interference, including water flow and wind and waves. It extracts core features through dual-branch extraction and realizes feature recombination and physical alignment of multi-ship joint data.

[0043] The specific steps are as follows:

[0044] Branch 1, State Extraction Training Based on Ship Dynamics: Based on the ship's MMG model, predict the next state and environmental disturbances using the current state and actions. The states include position, heading, and speed; environmental disturbances include current forces and wind forces, resulting in the initial ship state extraction; loss function... for: ;in, This is the predicted next state of the ship. The superscript te indicates the actual value, the superscript pd indicates the predicted value, t indicates the time, and i indicates the ship. These are predicted water flow forces and wind forces, where 'c' represents the water flow force. Indicates wind force; It is the weight of the equilibrium state prediction and the disturbance force prediction;

[0045] Training data: Small batches of ship data are sampled from the experience storage pool. A ship spatiotemporal feature extraction network is used to train and refine the model. CNN layers extract spatial features such as relative position, and LSTM layers capture temporal dynamics such as speed changes. Model structure: The CNN layer consists of two convolutional layers and a pooling layer, outputting spatial feature vectors; the LSTM layer consists of one bidirectional LSTM layer, outputting temporal feature vectors. Prediction head includes a state prediction head: a fully connected layer outputting the next state prediction. Interference force prediction head: Fully connected layer output environmental interference force prediction Forward propagation: Calculation of output from ship spatiotemporal feature extraction Backpropagation is performed using the Adam optimizer until... convergence;

[0046] Branch 2, State Extraction Training Based on Ship Mutual Simulation: Measuring the behavioral similarity between two ships i and j involves measuring whether their responses to the same control actions are consistent under similar sea conditions, and introducing a dynamic transition probability similarity coefficient, expressed as: ; ;in, is a measure of the behavioral similarity between ship i and ship j; the smaller the value, the more similar the behaviors. 'a' is the control action, and 'A' is the set of control actions. and It is the reward obtained by ships i and j for performing action a; In action a, water flow Wind speed Under these conditions, ship i transitions to state i. The probability of; The ship j is transferred to state The probability of; It is the transition probability similarity coefficient; It is the benchmark coefficient; This is a reference distance, used to define the threshold between interaction and no interaction. Beyond this distance, the two ships are considered to have no mutual influence. It is the JS divergence; It is a scenario correction factor. During free navigation, the ship's behavior is independent. When avoiding a collision Emergency collision avoidance In close-range, high-interaction scenarios, the focus is on the JS divergence of state transition probabilities; in emergency collision avoidance scenarios, the weight of physical response is strengthened to ensure that collision avoidance actions are compliant.

[0047] Learn a behavior feature encoder to map the initial ship data into low-dimensional behavior features, so that ships with similar behaviors have similar low-dimensional behavior features. Divide ship data into positive and negative sample pairs by behavior threshold; construct a behavior feature mapping network, including a 3-layer fully connected network, to output low-dimensional behavior features; predict the state transition probability distribution based on the low-dimensional behavior features, model it with a Gaussian distribution, and output the mean and variance.

[0048] For positive sample pairs, the core loss is defined as follows: ; and It is the extraction of the ship's low-dimensional state; and The ship received the award; and It is a conditional state transition probability distribution based on low-dimensional features, actions, and water flow, for all possible states;

[0049] For negative sample pairs, define the contrastive loss. , represented as: m is the marginal parameter;

[0050] In each iteration, an equal number of positive and negative sample pairs are sampled, and the parameters of the behavior feature encoder are updated using the Adam optimizer. The iteration continues until the loss converges.

[0051] The loss function based on the MMG model avoids the disconnect between black-box extraction and the actual motion of the ship, ensuring that the subsequent control strategy conforms to the physical laws of ship handling; the mutual simulation extraction supports transfer, captures the common behaviors of multi-ship cooperation, and can transfer the collision avoidance strategy of cargo ships to passenger ships, reducing redundant training.

[0052] By performing the above operations, this solution addresses the problems of general ship control systems, such as ignoring the dynamic nature of environmental interference, static multi-ship behavior similarity measurement, weak transferability, and insufficient anti-interference capability, which exacerbate energy consumption and course instability, leading to low ship control reliability. This solution extracts core features through a dual-branch extraction process and introduces a dynamic transition probability similarity coefficient into the dynamic behavior similarity measurement, making the similarity judgment adaptable to the interaction intensity and scenario type. By dividing ships into positive and negative sample pairs and setting loss functions for each, for positive sample pairs, the low-dimensional features of ships with similar behaviors are forced to be close, while for negative sample pairs, the feature distance is increased, thereby enabling collision avoidance strategies to be transferred across ship types. Through a three-layer design of physical modeling, dynamic measurement, and full data, the ship control strategy possesses physical feasibility, scenario adaptability, and cross-ship type transferability in complex sea conditions and multi-ship interaction scenarios, thereby improving ship control reliability.

[0053] Example 4, see Figure 1 This embodiment is based on the above embodiment. The navigation experience screening module has a large amount of redundant data in the ship experience storage pool. It prioritizes sampling high-value path segments, that is, path segments that include collision avoidance decisions, extreme sea state responses, and path corrections. It adopts a sampling method of ship path segment-level value extraction + reference path segment guidance. It realizes the optimization of multi-ship joint dataset and selectively screens the path segment data that is most valuable to the control strategy.

[0054] The specific steps are: defining and refining ship path segments; ship path segments... This is a continuous navigation segment, extracted as the difference between the initial state and the final state. This difference directly reflects the control significance of the path segment; the larger the difference, the more critical the decisions involved in the path segment. It is expressed as: ; It is a ship extraction at the end of a path segment, including the end point location, course, and cumulative rewards; It is the ship refining at the initial moment of the path segment; It is a single continuous navigation path segment of vessel i; It is a value extraction vector for path segments;

[0055] Similarity sampling guided by reference path segments: Randomly select one safe and high-value path segment from the experience storage pool. To ensure the safety of the sampling direction, the following conditions must be met: collision avoidance success rate = 100%, path deviation less than 50m, and energy consumption less than 10% of the average; all candidate path segments must be calculated. and The similarity is used to filter path segments with similarity ∈ [0.3, 0.7], avoiding completely redundant straight-line navigation path segments and dangerous path segments that differ too much from safe path segments, as shown below: If the number of path segments to be selected is greater than the batch size, the top K path segments are retained in descending order of reward value, and each path segment is split into single-step training samples. It is the path segment similarity.

[0056] Example 5, see Figure 1 This embodiment is based on the above embodiment. The multi-ship cooperative control agent training module is a system that needs to balance the individual accuracy of single-ship course tracking and the cooperative safety of multi-ship collision avoidance in ship control. For the optimized ship navigation experience repository of the ship state extraction module and navigation experience screening module, a ship guidance main network-ship control learning sub-network is established.

[0057] The specific steps are as follows:

[0058] The ship guidance master network is constructed with the goal of path tracking under multi-ship collision avoidance constraints to achieve group collaborative safety. An optimization objective function is constructed, expressed as: ; It is the control action sequence for the future H steps; It is the squared path deviation of ship i at step t+k; It is the square of the throttle opening; H is the total number of steps, and k is the step index; , and It is the target weight coefficient; It is the safe distance for collision avoidance; It is the distance between the centers of mass of ship i and ship j at step t+k.

[0059] The ship control learning subnetwork is constructed, and its Q-value function is divided into state value and action advantage. State value corresponds to the overall value of the marine environment, while action advantage corresponds to the local value of control actions, enhancing individual control effectiveness; expressed as: ;in, It is a state value; select the safety level for the current sea state. The advantage lies in the maneuverability, specifically the effect of adjusting the rudder angle on course correction. It is the Q-value function of ship i, and the output of the ship control learning sub-network; both state value and action advantage are obtained through neural network fitting, trained independently, and the network structure is a fully connected network + batch normalization layer. and These are network parameters; It is the action space Any action within;

[0060] Ship control learning subnetwork loss Add extreme risk penalty items To enhance the fitting of near-collision samples, it is expressed as: ; ;in, It is the guidance weight of the main network for ship collaborative control guidance; It is the Q-value evaluation of the learning subnetwork's actions by the ship's collaborative control guidance master network, ensuring that the learning subnetwork's strategy does not deviate from the safe direction; M is the number of samples; j is the sample index; It is the target Q value of the j-th sample.

[0061] Example 6, see Figure 1 Based on the above embodiments, the ship parameter adaptive adjustment module addresses the problem that the control strategy is prone to becoming rigid due to dynamic changes in the environment during long-term ship navigation. It designs a three-layer mechanism of partial parameter adjustment + periodic timing + safety reward. After adjustment, small actions are guided by rewards to prevent loss of control.

[0062] The specific operation is as follows: Adjustments are made to some parameters, dividing the ship control network parameters into a core safety layer (determining navigation safety and not adjustable) and a non-core optimization layer (affecting efficiency / energy consumption and adjustable). Adjustments are only made to the non-core layer to ensure that core functions such as collision avoidance and course stability are not interrupted. This is represented as follows: ; These are the control network parameters of ship i at time t; These are the initial parameters for the non-core layer; These are the control network parameters for ship i at time t+1, and are the adjusted control network parameters; the non-core layer includes all weight coefficients; the core layer includes network parameters and collision avoidance safety distance.

[0063] A dynamic adjustment cycle mechanism based on environmental awareness is introduced to prevent lag in response to sudden environmental changes and over-adjustment during periods of environmental stability. This is expressed as: ; ; This is the moment that triggers parameter adjustments; It is the basic adjustment period; n is the adjustment period ordinal number; It is the interference coefficient; It is the intensity of environmental disturbance; , and These are the water flow velocity, wind speed, and significant wave height at time t, respectively, which need to be normalized.

[0064] After adjustments, a safety bonus is awarded. Following parameter adjustments, the ship's control strategy may experience short-term fluctuations. Additional bonuses are needed to guide the ship to make minor adjustments and maintain safety, avoiding dangerous encounters. This is represented as follows: ;in, This is the total reward for ship i after parameter adjustments; It is a basic security reward; It is an indicator function; It is the threshold for rudder angle change; It is the change in rudder angle;

[0065] The mission termination condition for a collaborative mission is that all three conditions must be met simultaneously. If any one condition is not met, the mission will not terminate and must return to the multi-ship collaborative interaction data collection stage to collect new experience and retrain the multi-ship collaborative control model. The conditions include: the berthing position accuracy of all ships meets the standard, the berthing position deviation of a single ship is lower than the berthing threshold, the total mission duration does not exceed the limit, and there is no risk of collision between the ships.

[0066] By performing the above operations, this solution addresses the problems of blind experience screening, low utilization of high-value data, singular objectives in multi-ship collaboration, imbalance between individual and group needs, rigid parameter adjustments, and difficulty in adapting to dynamic environments, which ultimately lead to poor ship control performance in general ship control systems. This solution uses reference path-guided safety sampling to screen navigation experience and avoid interference from hazardous data. Simultaneously, it constrains path tracking accuracy, energy consumption, and collision avoidance safety, guiding multi-objective optimization of the main network. It avoids value overlap by learning the Q-value separation and risk penalties of sub-networks. It strengthens the fitting of close-range collision samples by adding extreme risk penalties. Through hierarchical parameter adjustment, it introduces a dynamic cycle of environmental perception, sets safety rewards after adjustments, and dynamically adapts to environmental changes, solving the problems of strategy rigidity and safety risks. Ultimately, this improves ship control performance.

[0067] Example 7, see Figure 1 This embodiment is based on the above embodiment. The ship control module acquires real-time ship operation data, inputs it into the ship status refinement module for optimization, and inputs it into the multi-ship collaborative control intelligent agent module. Real-time ship control is realized based on the module output. The intelligent agent combines the optimized ship status with the multi-ship collision avoidance rules to generate control strategies for rudder angle, throttle, and propeller speed. The strategy commands are then sent to the hardware of the steering gear and actuators. At the same time, the execution effect is fed back in real time for dynamic fine-tuning. Finally, the ship achieves real-time precise control, ensuring stable single-ship maneuvering and safe multi-ship collaborative operation.

[0068] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0069] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. An intelligent ship control system, characterized in that: The system includes a ship interactive data acquisition module, a ship status extraction module, a navigation experience screening module, a multi-ship collaborative control agent training module, a ship parameter adaptive adjustment module, and a ship control module. The ship interactive data acquisition module acquires the ship's own status data and surrounding ship observation data to build a ship navigation experience repository. The ship status extraction module optimizes the features of samples from the ship navigation experience repository through dual-branch training. The navigation experience screening module defines a value extraction vector for ship route segments and filters samples from the ship navigation experience repository. The multi-ship cooperative control agent training module is based on a filtered and optimized ship navigation experience repository to construct a ship guidance main network and a control learning sub-network. The ship parameter adaptive adjustment module controls the network parameters into a core security layer and a non-core optimization layer, and calculates the dynamic adjustment cycle according to the intensity of environmental interference. The ship control module acquires real-time ship operation data, inputs it into the ship status refinement module for optimization, and then inputs it into the multi-ship collaborative control intelligent agent to achieve real-time ship control. The ship condition refining module specifically includes: Branch 1, State Extraction Training Based on Ship Dynamics: Based on the ship's MMG model, the initial ship state is obtained by predicting the next state and environmental disturbances from the current state and actions; loss function... for: ;in, This is the predicted next state of the ship. The superscript te indicates the actual value, the superscript pd indicates the predicted value, t indicates the time, and i indicates the ship. These are predicted water flow forces and wind forces, where 'c' represents the water flow force. Indicates wind force; It is the weight of the equilibrium state prediction and the disturbance force prediction; Branch 2, State Extraction Training Based on Ship Mutual Simulation: This measures the behavioral similarity between two ships and introduces a dynamic transition probability similarity coefficient; expressed as: ; ;in, is a measure of the behavioral similarity between ship i and ship j; the smaller the value, the more similar the behaviors. 'a' is the control action, and 'A' is the set of control actions. and It is the reward obtained by ships i and j for performing action a; In action a, water flow Wind speed Under these conditions, ship i transitions to state i. The probability of; The ship j is transferred to state The probability of; It is the transition probability similarity coefficient; It is the benchmark coefficient; This is a reference distance; It is the distance between the centers of mass of the two ships; It is the JS divergence; It is the scene correction coefficient; learn a behavior feature encoder to map the initial ship extraction to low-dimensional behavior features, so that ships with similar behaviors have similar low-dimensional behavior features, and divide ship data into positive and negative sample pairs by behavior threshold; define the core loss for positive sample pairs; define the contrast loss for negative sample pairs; sample an equal number of positive and negative sample pairs in each iteration, update the behavior feature encoder parameters with Adam optimizer, and iterate until the loss converges.

2. The intelligent ship control system according to claim 1, characterized in that: The navigation experience screening module specifically includes: The definition of ship path segment extraction is that extraction is the difference between the initial state extraction and the final state extraction; Similarity sampling guided by reference path segments is used to randomly select one path segment from the experience storage pool; the similarity between all candidate path segments and the selected path segment is calculated, and path segments with similarity ∈ [0.3, 0.7] are filtered; if the number of filtered path segments is greater than the batch size, the top K path segments are retained in descending order of reward value, and each path segment is split into single-step training samples.

3. The intelligent ship control system according to claim 2, characterized in that: The multi-ship cooperative control agent training module specifically includes: The ship guidance master network is constructed with path tracking under multi-ship collision avoidance constraints as the objective, and an optimization objective function is constructed. The ship control learning subnetwork is constructed, and its Q-value function is divided into state value and action advantage. The state value corresponds to the overall value of the marine environment, and the action advantage corresponds to the local value of the control action. Both the state value and the action advantage are obtained by fitting through a neural network. The loss of the ship control learning subnetwork is added to the extreme risk penalty item.

4. The intelligent ship control system according to claim 3, characterized in that: The ship parameter adaptive adjustment module specifically includes: Some parameters were adjusted, dividing the ship control network parameters into a core safety layer and a non-core optimization layer, and adjustments were only made to the non-core layer. A dynamic adjustment cycle mechanism based on environmental perception is introduced; a safety reward is obtained after the adjustment.

5. The intelligent ship control system according to claim 4, characterized in that: The ship interaction data acquisition module acquires the interaction data of ships in the digital twin environment. Each ship collects its own status and the observation status of surrounding ships in real time through sensors; after executing control actions, it records the next status and reward; and stores the joint data of multiple ships in the ship navigation experience repository.

6. The intelligent ship control system according to claim 5, characterized in that: The ship control module acquires real-time ship operation data, inputs it into the ship status refinement module for optimization, and inputs it into the multi-ship collaborative control intelligent agent module. Real-time ship control is achieved based on the module output.

Images