Intelligent Ship Management System Based on Automatic Operation

Abstract

The present invention discloses an intelligent ship management system based on automatic operation, specifically related to the field of ship management, including a variable operation data acquisition module, a dynamic working condition perception module, an elastic rule update module, a decision-making reasoning module, a multi-objective dynamic optimization module, a decision-making monitoring module, and a digital twin verification module; the intelligent ship management system based on automatic operation significantly improves the adaptability of the system to the variable operation environment by updating the elastic rule knowledge base; the decision-making weight is dynamically adjusted according to the working conditions through the decision-making reasoning module, avoiding the decision-making rigidity of the preset rules in the case of multi-objective conflicts; through the multi-objective dynamic optimization module to synchronously optimize multiple objectives, avoiding the defect of global sub-optimality caused by single-objective optimization in the traditional method, forming a dynamic optimization system covering the whole process of ship operation, and improving the overall operation efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of ship management, and more specifically, to an intelligent ship management system based on automatic operation. Background Art

[0002] With the acceleration of the intelligent process in the shipping industry, efficient and accurate ship management is particularly important; the traditional ship management architecture mainly consists of various sensors and monitoring devices on the ship side and the management center server on the shore base. By periodically collecting the working efficiency and operation data of the ship and transmitting them to the shore base management center server via satellite links, the normal operation of multiple ships is ensured.

[0003] To break through the bottleneck of the traditional architecture, the existing technology has developed towards automation and intelligence. By using a lightweight LSTM neural network to analyze core parameters such as carbon emissions and fuel consumption, and making preliminary decisions according to preset rules, the frequency of data transmission to the shore base management center is reduced, and the communication pressure is lowered.

[0004] However, in actual use, there are still some drawbacks. For example, the automated operation management decision is mainly based on preset rules, lacking sufficient flexibility. When the system faces complex and numerous transportation tasks, it is difficult to quickly and accurately generate the optimal decision, resulting in a decline in the efficiency of the ship to execute tasks, thereby affecting the overall operation efficiency. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides an intelligent ship management system based on automatic operation, through the following solutions to solve the problems raised in the above background art:

[0006] The intelligent ship management system based on automatic operation includes:

[0007] A variable operation data acquisition module: used to integrate the variable operation data of the target port, the variable operation data including AIS data, task information and energy consumption monitoring data, and output the first ship condition after preprocessing;

[0008] A dynamic condition perception module: based on the first ship condition, using a recurrent neural network to identify the key features of the ship conditions in the target port, and obtaining the first condition feature including the ship condition and the operating environment;

[0009] An elastic rule update module: used to perform online transfer learning based on the first condition feature and update the elastic rule knowledge base, the elastic rule knowledge base storing operation management rules including a multi-layer structure;

[0010] Decision Inference Module: Used to construct a decision inference model, perform multi-objective game operations on the elastic rule knowledge base through the decision inference model, and generate the first operation management decision;

[0011] Multi-objective Dynamic Optimization Module: Based on the first operation management decision, synchronously process multiple sub-goals through a multi-objective optimization algorithm to generate the second operation management decision;

[0012] Decision Monitoring Module: Used to convert the second operation management decision into a multi-ship operation management report corresponding to the target port and send it to the shore-based management center, and monitor the execution process of the multi-ship operation management report;

[0013] Digital Twin Verification Module: Used to construct a virtual test field through digital twin technology, and trigger the update of the elastic rule knowledge base based on the execution process of the multi-ship operation management report.

[0014] Preferably, for the dynamic working condition perception module, the recurrent neural network is a hybrid network structure of LSTM and GRU, specifically including:

[0015] The input dimension is aligned with the first ship working condition dimension;

[0016] The LSTM branch embeds a spatio-temporal attention mechanism, and at time step the attention weight The calculation formula is specifically expressed as:

[0017] ;

[0018] Among them, represents the attention weight matrix, represents the hidden state weight matrix, represents the hidden state at time step , represents the input weight matrix, represents the first ship working condition at time step , represents the bias term;

[0019] The outputs of the GRU branch and the LSTM branch are weighted and aggregated through dynamic attention weights.

[0020] Preferably, for the elastic rule update module, the update trigger conditions of the elastic rule knowledge base include the switching signal of the ship condition and the change of the operation environment of the ship from the dynamic working condition perception module, the decision deviation detected by the digital twin verification module exceeding the preset threshold, and the manual rule revision instruction input by the user through the user information terminal.

[0021] Preferably, for the decision-making and reasoning module, the construction of the decision-making and reasoning model specifically includes:

[0022] Using a fuzzy Petri net to handle rule conflicts from the elastic rule knowledge base and constructing a dynamic priority determination tree;

[0023] Adopting a deep Q-network to optimize the decision-making and reasoning model through long-term iteration.

[0024] Preferably, for the decision-making and reasoning module, performing a multi-objective game operation on the elastic rule knowledge base through the decision-making and reasoning model specifically includes:

[0025] Defining a task rule container that stores various task type constraints and rules, and the task rule container corresponds to the task type and the corresponding preset operation objectives;

[0026] Receiving the activation degree and mutual relationship of each rule in the elastic rule knowledge base on different ships, where the activation degree is the applicability of the rule in ship task execution and operation objectives, and the mutual relationship is the logical association between different rules;

[0027] Calculating the activation degree of each task rule container through fuzzy reasoning , representing the priority of each task rule container in the multi-ship operation environment at the target port, specifically expressed as:

[0028] Among them, Represents the activation degree of the task rule container , Represents the number of rules received by the task rule container , Represents the rule index received by the task rule container , Represents the influence of the th rule on the activation degree of the task rule container , Represents the degree to which the th rule is triggered;

[0029] Constructing a dynamic priority determination tree and selecting the optimal decision path according to the dynamic weight of the preset operation objective;

[0030] Balancing multiple preset operation objectives by maximizing the cumulative reward.

[0031] Preferably, for the multi-objective dynamic optimization module, the multi-objective optimization algorithms include the NSGA-Ⅲ algorithm, the ant colony algorithm, and the mixed integer method, and the multi-objectives include economic optimization, emergency optimization, and ship scheduling optimization.

[0032] Preferably, the multi-objective dynamic optimization module uses an improved NSGA-Ⅲ algorithm to process multiple conflicting objectives, where the conflicting objectives are multiple mutually conflicting preset operation objectives, specifically including:

[0033] Performing clustering analysis based on the historical operation management decisions corresponding to the conflicting objectives, and generating multiple feature vectors through the results of the clustering analysis, where the feature vectors are represented as the operation management decisions in the historical operation management decisions;

[0034] Adopting simulated binary crossover operation and polynomial mutation, and introducing a safety constraint violation penalty term in the selection pressure mechanism;

[0035] Constructing a dynamic multi-objective optimization weight matrix to adjust the priorities of the conflicting objectives.

[0036] Preferably, the multi-objective dynamic optimization module uses an improved ant colony algorithm to process multiple conflicting objectives, specifically including:

[0037] The pheromone update rule in the improved ant colony algorithm is used to balance the selection probabilities of multiple conflicting objectives, specifically expressed as:

[0038] ;

[0039] Wherein, represents the pheromone concentration from the node corresponding to the conflicting objective to the node to at time step represents the pheromone evaporation coefficient, represents the pheromone concentration from the node corresponding to the conflicting objective to the node to at time step represents the pheromone intensity constant, represents the total probability of executing the conflicting objective for the th time, represents the risk coefficient corresponding to the conflicting objective;

[0040] The heuristic factor in the improved ant colony algorithm is specifically expressed as:

[0041] ;

[0042] Wherein, represents the attraction of the node corresponding to the conflicting objective to , represents the distance from the node corresponding to the conflicting objective to to represents the node corresponding to the conflicting objective to steering angle, indicating the execution direction of conflicting goals.

[0043] Technical effects and advantages of the present invention:

[0044] 1. The present invention updates the elastic rule knowledge base through the elastic rule update module, solves the problem that preset rules are difficult to adapt to sudden tasks, avoids the lag of manual intervention, and significantly improves the adaptability of the system to the changing operation environment;

[0045] 2. The present invention realizes intelligent decision-making under multi-objective game through the decision-making reasoning module, dynamically adjusts the decision-making weight according to the working conditions, avoids the decision-making rigidity of preset rules in the case of multi-objective conflicts, and effectively solves the problem of sub-optimal decision-making in the case of multi-objective conflicts;

[0046] 3. The present invention synchronously optimizes multiple goals through the target dynamic optimization module, avoids the defect of global sub-optimality caused by single-goal optimization in traditional methods, forms a dynamic optimization system covering the whole process of ship operation, and improves the overall operation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 is a flowchart of an intelligent ship management system based on automatic operation according to an embodiment of the present application.

[0048] Figure 2 is a block diagram of the architecture of the elastic rule knowledge base in the intelligent ship management system based on automatic operation according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0049] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0050] The terms used in the following embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. As used in the specification of the present application, the singular forms "a", "an", "the", "above", "said", "this" are also intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in the present application refers to and includes any or all possible combinations of one or more of the listed items.

[0051] Hereinafter, the terms "first" and "second" are for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the embodiments of the present application, unless otherwise specified, the meaning of "a plurality" is two or more.

[0052] As shown in the appended Figure 1 intelligent ship management system based on automatic operation, including but not limited to a system central processing module, a communication bus, a historical operation database, a user information terminal, etc., further including a variable operation data acquisition module, a dynamic condition perception module, an elastic rule update module, a decision-making reasoning module, a multi-objective dynamic optimization module, a decision-making monitoring module, and a digital twin verification module.

[0053] Among them, the system central processing module is the core operation and control unit of the entire intelligent ship operation management system, including one or more heterogeneous processing cores, connecting each functional module in the system through a high-speed serial bus and a PCIe interface; by running the multi-modal decision algorithm code set stored in the historical operation database, calling the constraint condition matrix stored in the elastic rule knowledge base and the hydrodynamic model in the digital twin parameter library, performing ship dynamic path planning, multi-objective collaborative optimization, and safety risk prediction operations to ensure that multi-objective decision-making generation in a variable operation environment is completed within 200 ms, the monitoring system load rate does not exceed the threshold of 70%, and parallel processing of at least 32 channels of sensor data is supported.

[0054] Among them, the communication bus is used to implement deterministic communication between components using a time-sensitive network protocol, including but not limited to a CAN bus connecting ship equipment IoT nodes, an Ethernet fiber optic ring network connecting each intelligent module, and a Profibus DP bus controlling actuators.

[0055] Among them, the historical operation database is used to store a large amount of data related to intelligent ship management based on automatic operation, including data related to mathematical models established based on physical principles such as ship kinematics and hydrodynamics principles, storing a large amount of historical operation data, including a ten-year AIS trajectory database, a main engine energy consumption characteristic curve library, an emergency handling case atlas, etc.; when the system central processing module executes dynamic rule updates, it calls the historical ocean current pattern data and sensor data in the database for comparison and analysis, so as to realize closed-loop optimization control of the ship speed.

[0056] Among them, the user information terminal provides an interface for users to interact with the system by connecting external devices such as a display screen and a camera through a standard wired interface or a wireless interface.

[0057] Among them, the variable operation data acquisition module is used to integrate the variable operation data of the target port. The variable operation data includes AIS data, task information, and energy consumption monitoring data, and outputs the first ship condition after preprocessing.

[0058] It should be noted that the variable operation data acquisition module depends on the operation information of the target port. Among them, the AIS data receives the position, speed, course, etc. of multiple ships through the AIS of the target port; task information: the task information is obtained by the user entering information, including but not limited to the plan for loading and unloading goods, docking time, route arrangement, etc.; the energy consumption monitoring data includes but not limited to the energy consumption of the dock equipment corresponding to the target port, the energy supply situation, etc.

[0059] Furthermore, the dynamic data preprocessing operation on the variable operation data includes: using the variational Bayesian filtering technology to perform noise reduction processing on the variable operation data, calculating the posterior distribution through variational inference iteration, and dynamically adjusting the filtering bandwidth; realizing through the NTP-PTP hybrid clock synchronization protocol, and aligning the timestamp of the data to the microsecond-level accuracy.

[0060] Among them, the dynamic condition perception module, based on the first ship condition, uses a recurrent neural network to identify the key features of the conditions of each ship in the target port, and obtains the first condition feature including the ship condition and the operating environment.

[0061] It should be noted that the dynamic condition perception module uses an improved recurrent neural network to identify the conditions of each ship in the target port and its operating environment based on the first ship condition transmitted by the variable operation data acquisition module. Among them, the ship conditions in the first condition feature include but are not limited to the standby state of waiting for task allocation in a specified area, the pre-execution state of having received a task instruction, the task execution state of being in the process of executing the assigned task, the emergency state of encountering unexpected situations, etc., and the operating environment includes but is not limited to the task state, the traffic condition of the port, the berth occupancy situation, the operating state of the key equipment of the ship, the communication quality between the ship and the port and the shore base, etc.

[0062] In a possible implementation manner, the recurrent neural network adopts a hybrid network structure of LSTM and GRU. Among them, the input dimension in the input layer of the hybrid network structure is aligned with the dimension of the first ship condition; a spatio-temporal attention mechanism is embedded in the LSTM branch. Among them, at the time step the attention weight The calculation formula is specifically expressed as:

[0063] Among them, is expressed as the attention weight matrix, is expressed as the hidden state weight matrix, is expressed as at the time step The hidden state, is represented as the input weight matrix, is represented at time step of the first ship condition, is represented as the bias term; it should be noted that the attention weight matrix , the hidden state weight matrix and the input weight matrix are all trainable parameter matrices. Among them, the attention weight matrix is used to map the hidden state and the first ship condition to the attention weights, capturing the correlation between the hidden state and the first ship condition; the hidden state weight matrix is used to represent the contribution of the hidden state to the calculation of the attention weights at the current time step, and the input weight matrix is used to represent the contribution of the current input of the first ship condition to the calculation of the attention weights at the current time step; the outputs of the GRU branch and the LSTM branch are weighted and aggregated through the dynamic attention weights.

[0064] In this embodiment, the LSTM branch consists of a 3-layer stacked structure, with each layer containing 64 hidden units, and the forget gate bias is initialized to 1; the GRU branch adopts a 2-layer structure, with each layer containing 48 units, and CudnnGRU is used to achieve GPU acceleration.

[0065] Among them, the elastic rule update module is used to perform online transfer learning based on the first condition feature and update the elastic rule knowledge base, and the elastic rule knowledge base stores the operation management rules including a multi-layer structure.

[0066] It should be noted that the elastic rule update module receives the switching signal of the ship condition and the change of the operating environment where the ship is located from the dynamic condition perception module, as well as the verification feedback information from the digital twin verification module, and adjusts and updates the elastic rule knowledge base.

[0067] Furthermore, the elastic rule knowledge base stores operation management rules including a multi-layer structure. In this embodiment, a three-layer dynamic evolution architecture is adopted, including a static safety rule layer, a dynamic optimization rule layer, and a self-learning emergency strategy layer. Among them, the static safety rule layer stores basic and mandatory safety rule sets, which are derived from the mandatory regulations of the International Maritime Organization, the structural and equipment specifications of classification societies, and the safety operation thresholds provided by ship equipment manufacturers. The dynamic optimization rule layer contains optimization parameters and rules for improving ship operation efficiency. In this embodiment, it includes an energy efficiency-speed curve generated based on data for guiding the selection of economic speed, a route economic evaluation function dynamically adjusted according to factors such as fuel price and time cost, and a device scheduling priority table generated based on the prediction of the remaining service life of the device and the current load condition. The self-learning emergency strategy layer is used to store and generate emergency disposal strategies for coping with sudden, abnormal, or extreme working conditions. The emergency disposal strategies include historical accident disposal records, virtual simulation plans generated in a virtual environment through the digital twin module, and rules encoded and extracted from expert experience reports through natural language processing technology.

[0068] It should be noted that when conflicts occur between different rule layers, a conflict resolution mechanism is adopted. The conflict resolution mechanism includes establishing a rule priority determination tree to clarify the priorities among multiple rules. In this embodiment, the static safety rules > dynamic optimization rules > economic rules > self-learning emergency strategies are adopted. If there are conflicts among rules with the same priority, a data-driven dynamic rule weight adjustment mechanism is adopted, with an adjustment frequency of 1 Hz. By dynamically allocating weights according to the current working conditions and the historical performance of the rules, a fuzzy reasoning method is adopted in this embodiment to define the membership function of rule applicability and solve the fuzzy output through the centroid method to obtain the final rule.

[0069] Furthermore, the update mechanism of the elastic rule knowledge base is mainly realized through online transfer learning, and incremental updates are performed every 5 minutes. The conditions for incremental updates include but are not limited to the switching signal of the ship condition in the dynamic working condition perception module and the change of the operating environment where the ship is located, the decision deviation detected by the digital twin verification module exceeding the preset threshold, and the manual rule revision instruction input by the user through the user information terminal, etc.

[0070] Among them, the decision reasoning module is used to construct a decision reasoning model, perform multi-objective game operations on the elastic rule knowledge base through the decision reasoning model, and generate a first operation management decision.

[0071] It should be noted that the decision reasoning model is a pre-constructed learning model. Perform multi-objective game operations on the elastic rule knowledge base through the decision reasoning model to obtain an optimized decision-making plan that meets multi-objective constraints, that is, the first operation management decision. The first operation management decision includes but is not limited to the operation adjustment instructions of the ship, the task scheduling plan, the resource allocation plan, etc.

[0072] Furthermore, the decision-making and reasoning model adopts a hybrid architecture of fuzzy Petri net and deep Q-network for multi-objective game; uses the fuzzy Petri net to handle rule conflicts from the flexible rule knowledge base, constructs a dynamic priority determination tree, and generates a preliminary decision-making tendency; adopts the deep Q-network to optimize the decision-making and reasoning model in the long term to adapt to unknown working conditions.

[0073] In a possible implementation, the multi-objective game operation includes: defining a task rule container that stores various task type constraints and rules, where the task rule container corresponds to the task type and the corresponding preset operation objectives, and the content of the task rule container includes but is not limited to COLREG rules, equipment safety thresholds, fuel consumption-speed curves, route economy parameters, case handling sequences, expert experience rules, etc.; the task rule container establishes a corresponding relationship with the task type and operation objectives through a classification and coding mapping mechanism, and each task rule container corresponds to a set of parameters involved in the content of a type of the task rule container and the task type corresponding to the preset operation objective; receiving the activation degree and mutual relationship of each rule in the flexible rule knowledge base on different ships, where the activation degree is the applicability of the rule in ship task execution and operation objectives, and the mutual relationship is the logical association between different rules; calculating the activation degree of each task rule container through fuzzy reasoning , specifically expressed as:

[0074] ;

[0075] Among them, represents the activation degree of the task rule container , with a value range of [0, 1], represents the number of rules received by the task rule container , represents the rule index received by the task rule container , represents the influence of the th rule on the activation degree of the task rule container , with a value range of [0, 1], represents the degree of activation of the th rule, with a value range of [0, 1], represents the "or" operation in fuzzy logic, which is used to calculate the maximum value in the contributions of multiple rules to the activation degree of the task rule container ; constructing a dynamic priority determination tree, and according to the dynamic weights of the preset operation objectives, the preset operation objectives in this embodiment include safety, economy, and emergency, and select the optimal decision-making path , specifically expressed as:

[0076] ;

[0077] Among them, represents all possible paths in the constructed dynamic priority decision tree, represents the dynamic weight of the preset operation target, represents the path in the target score; balance multiple preset operation targets by maximizing the cumulative reward;

[0078] Among them, the multi-objective dynamic optimization module, based on the first operation management decision, synchronously processes multiple sub-goals through a multi-objective optimization algorithm to generate a second operation management decision.

[0079] It should be noted that the multi-objective dynamic optimization module uses a multi-objective optimization algorithm to coordinate multiple conflicting preset operation targets in the first operation management decision to generate a final optimized decision-making scheme, that is, the second operation management decision.

[0080] In a possible implementation manner, the multi-objective optimization algorithm includes the NSGA-Ⅲ algorithm, the ant colony algorithm, and the mixed integer method. The multi-objectives include, but are not limited to, economic optimization, energy efficiency optimization, task scheduling optimization, path planning optimization, emergency optimization, collision risk minimization, equipment scheduling optimization, resource allocation optimization, and maintenance plan optimization, etc.

[0081] Furthermore, an improved NSGA-Ⅲ algorithm is used to process multiple conflicting objectives. It should be noted that the conflicting objectives are multiple conflicting preset operation targets, and a series of non-dominated solution sets that achieve a balance among the objectives are obtained, including: performing cluster analysis based on the historical operation management decisions corresponding to the conflicting objectives, generating multiple feature vectors through the results of the cluster analysis, and the feature vectors are represented as the operation management decisions in the historical operation management decisions; using simulated binary crossover operations that can generate new candidate solutions and polynomial mutations that increase the diversity of the population, and introducing a safety constraint violation penalty term in the selection pressure mechanism to punish the solutions that violate the safety constraints and reduce their probability of being selected. Among them, the simulated binary crossover operation combines the characteristics of two parent individuals by simulating the crossover phenomenon in biological evolution to generate diverse offspring individuals; constructing a dynamic multi-objective optimization weight matrix to adjust the priorities of each conflicting objective. In this embodiment, the dynamic adjustment factor of the ship safety weight is positively correlated with the wind and wave level; the dynamic adjustment factor of the fuel economy weight is negatively correlated with the oil price volatility; the equipment life weight is adjusted based on the prediction of the remaining service life of the equipment.

[0082] In this embodiment, for sub-goals that require weighing multiple cost and efficiency factors, such as economic optimization, energy efficiency optimization, and task scheduling optimization, the NSGA-Ⅲ algorithm is mainly adopted. Among them, for economic optimization, by considering multiple cost dimensions such as fuel cost, equipment wear cost, and labor cost simultaneously, an operation strategy set that minimizes the total operation cost is solved. For energy efficiency optimization, by optimizing parameters such as the speed, route, and equipment configuration of the ship, the energy utilization efficiency is maximized, and fuel consumption and carbon emissions are reduced. For task scheduling optimization, by processing the complex task sequences of the ship in the target port or target area, considering the dependencies, time window constraints, and resource availability between tasks, the task execution order is reasonably arranged, thereby minimizing the total task completion time and improving the overall operation efficiency.

[0083] Furthermore, an improved ant colony algorithm is used to handle multiple conflicting goals. In the improved ant colony algorithm, the pheromone update rule is used to balance the selection probabilities of multiple conflicting goals, which is specifically expressed as:

[0084] Among them, represents the pheromone concentration of the node corresponding to the conflicting goal at time step to , represents the pheromone evaporation coefficient, represents the pheromone concentration of the node corresponding to the conflicting goal at time step to , represents the pheromone intensity constant, represents the total probability of the th execution of the conflicting goal, represents the risk coefficient corresponding to the conflicting goal; in this embodiment, the calculation method of the path risk coefficient is specifically expressed as:

[0085] ;

[0086] Among them, represents the ocean current interference degree expressed by the ratio of the current flow rate to the maximum anti-current capacity of the ship, represents the visibility attenuation coefficient calculated based on the fusion of camera and radar data, and respectively represent the weights of the ocean current interference degree and the visibility attenuation coefficient;

[0087] The heuristic factor in the improved ant colony algorithm is specifically expressed as:

[0088] ;

[0089] Among them, The node corresponding to the conflicting target The attraction to The node corresponding to the conflicting target The distance to The node corresponding to the conflicting target The steering angle to It represents the execution direction of the conflicting target; it should be noted that when the pre-execution direction is consistent with the execution direction of the conflicting target, that is Approaching 0, the heuristic value increases, and the pre-execution direction is preferentially selected; when the pre-execution direction deviates from the execution direction of the conflicting target, that is Approaching 180°, the heuristic value decays, and the execution direction of the conflicting target is avoided.

[0090] In this embodiment, for sub-goals such as path planning optimization, urgency optimization, and collision risk minimization, an improved ant colony algorithm is mainly used; by simulating the behavior of ants communicating and selecting paths through pheromones during foraging, in terms of path planning optimization, according to real-time meteorological, ocean current, traffic density and other information, the pheromone concentration and heuristic factor are dynamically adjusted to plan the shortest and safest navigation path for the ship, effectively reducing the navigation time and fuel consumption; in terms of urgency optimization, when receiving an emergency task or unexpected situation, quickly respond to environmental changes, by adjusting pheromones and heuristic factors, preferentially explore and select paths and actions that can quickly respond to emergencies to ensure the priority handling of emergency tasks; in terms of collision risk minimization, integrate real-time traffic information from sensors such as AIS and radar, set the pheromone concentration in high-density traffic areas to a lower value, or introduce a risk penalty term in the heuristic factor to guide the algorithm to avoid high-risk areas, thereby reducing the risk of ship collisions.

[0091] Furthermore, a mixed integer method is constructed. For sub-goals such as equipment scheduling optimization and resource allocation optimization, the optimization problem containing continuous variables and discrete variables is accurately modeled, and complex linear constraints are processed.

[0092] ​​​In this embodiment, in terms of equipment scheduling optimization, by accurately characterizing the operating characteristics, start-stop costs, maintenance requirements of each equipment in the ship, as well as the power balance constraints among them, the start-stop time and sequence of each equipment are arranged to achieve load balancing of the power system, improve equipment utilization rate, and reduce equipment wear and energy consumption; in terms of resource allocation optimization, the allocation problems of resources such as manpower and materials in ship operation are handled, considering the availability of resources, task requirements, and cost constraints, to improve resource utilization efficiency and reduce operation costs; in terms of maintenance plan optimization, combined with the operating status of equipment, historical failure data, and predicted remaining service life, an optimal equipment maintenance plan is formulated to balance maintenance costs and equipment failure risks and extend the service life of equipment.

[0093] Among them, the decision monitoring module is used to convert the second operation management decision into a multi-ship operation management report corresponding to the target port and send it to the shore-based management center, and monitor the execution process of the multi-ship operation management report.

[0094] It should be noted that the decision monitoring module converts the second operation management decision generated by the multi-objective dynamic optimization module into a multi-ship operation management report corresponding to the target port; the content of the multi-ship operation management report includes the ship identification involved, specific decision instructions, the execution time and estimated completion time of the instructions, the resource allocation situation involved, task scheduling adjustment, and relevant safety tips and precautions; and it is sent to the shore-based management center corresponding to the target port; at the same time, the execution process of the multi-ship operation management report is monitored in real time and feedback information is collected, and the feedback information is the progress data of the task execution of each ship, including but not limited to task completion progress, task execution efficiency, delay situation, actual start time, actual end time, estimated completion time, real-time fuel consumption rate, equipment startup time, personnel work status, etc.

[0095] Among them, the digital twin verification module is used to construct a virtual test field through digital twin technology and trigger the update of the elastic rule knowledge base based on the execution process of the multi-ship operation management report.

[0096] It should be noted that the virtual test field is based on a digital twin model constructed by a ship hydrodynamic model, an equipment dynamics model, an environmental model, and data obtained from the variable operation data acquisition module and the dynamic working condition perception module; by inputting the feedback information collected from the decision monitoring module into the digital virtual test field to simulate the actual scenario after the decision is executed; decision evaluation is carried out based on the simulation results of the virtual test field and the feedback information, and the decision evaluation includes but not limited to comparing the ship state predicted by the virtual test field simulation with the actually collected feedback information and calculating the decision deviation of the execution; when it is detected through evaluation that the decision deviation exceeds the preset threshold, the digital twin verification module triggers the update of the elastic rule knowledge base.

[0097] Secondly, in the attached drawings of the disclosed embodiments of the present invention, only the structures related to the disclosed embodiments are involved. For other structures, reference may be made to the general design. Without conflict, the same embodiment and different embodiments of the present invention can be combined with each other;

[0098] Finally, the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. An intelligent ship management system based on automatic operation, characterized in that Including: Variable operation data acquisition module: used to integrate the variable operation data of the target port, the variable operation data includes AIS data, task information and energy consumption monitoring data, and outputs the first ship condition after preprocessing; Dynamic condition perception module: Based on the first ship condition, a recurrent neural network is used to identify the key features of the ship conditions in the target port, and the first condition features including the ship condition and the operating environment are obtained; Elastic rule update module: used to perform online transfer learning based on the first condition features and update the elastic rule knowledge base, and the elastic rule knowledge base stores operation management rules including a multi-layer structure; Decision-making inference module: used to construct a decision-making inference model, and perform multi-objective game operations on the elastic rule knowledge base through the decision-making inference model to generate the first operation management decision; Multi-objective dynamic optimization module: Based on the first operation management decision, multiple sub-goals are synchronously processed through a multi-objective optimization algorithm to generate the second operation management decision; Decision monitoring module: used to convert the second operation management decision into a multi-ship operation management report corresponding to the target port and send it to the shore-based management center, and monitor the execution process of the multi-ship operation management report; Digital twin verification module: used to construct a virtual test field through digital twin technology, and trigger the update of the elastic rule knowledge base based on the execution process of the multi-ship operation management report.

2. The intelligent ship management system based on automatic operation according to claim 1, characterized in that: For the dynamic condition perception module, the recurrent neural network is a hybrid network structure of LSTM and GRU, specifically including: The input dimension is aligned with the dimension of the first ship condition; The LSTM branch embeds a spatio-temporal attention mechanism, and the attention weights at each time step are calculated according to the following formula, which is specifically expressed as: ； Among them, is represented as the attention weight matrix, is represented as the hidden state weight matrix, is represented as at time step the hidden state, is represented as the input weight matrix, is represented as at time step the first ship condition, is represented as the bias term; The outputs of the GRU branch and the LSTM branch are weighted and aggregated through dynamic attention weights.

3. The intelligent ship management system based on automatic operation according to claim 1, characterized in that: For the elastic rule update module, the update trigger conditions of the elastic rule knowledge base include the switching signal of the ship condition in the dynamic condition perception module and the change of the operating environment where the ship is located, the decision deviation detected by the digital twin verification module exceeding the preset threshold, and the manual rule revision instruction input by the user through the user information terminal.

4. The intelligent ship management system based on automatic operation according to claim 1, characterized in that: For the decision-making inference module, the construction of the decision-making inference model specifically includes: Using a fuzzy Petri net to handle rule conflicts from the elastic rule knowledge base and constructing a dynamic priority determination tree; Adopting a deep Q network to iteratively optimize the decision-making inference model in the long term.

5. The intelligent ship management system based on automatic operation according to claim 1, wherein: For the decision-making inference module, performing multi-objective game operations on the elastic rule knowledge base through the decision-making inference model specifically includes: Defining a task rule container that stores various task type constraints and rules, and the task rule container corresponds to the task type and the corresponding preset operation goals; Receiving the activation degree and mutual relationship of each rule in the elastic rule knowledge base on different ships, the activation degree is the applicability of the rule in ship task execution and operation goals, and the mutual relationship is the logical association between different rules; Calculate the activation degree of each task rule container through fuzzy reasoning , indicating the priority of each task rule container in the multi-ship operation environment at the target port, specifically expressed as: ； Among them, represents the activation degree of the task rule container ; represents the number of rules received by the task rule container ; represents the rule index received by the task rule container ; represents the influence of the th rule on the activation degree of the task rule container ; represents the degree to which the th rule is triggered; Constructing a dynamic priority determination tree, and selecting the optimal decision path according to the dynamic weight of the preset operation goal; Balancing multiple preset operation goals by maximizing the cumulative reward.

6. The intelligent ship management system based on automatic operation according to claim 1, characterized in that: The multi-objective dynamic optimization module, and the multi-objective optimization algorithms include the NSGA-Ⅲ algorithm, the ant colony algorithm, and the mixed integer method. The multi-objectives include economic optimization, urgency optimization, and ship scheduling optimization.

7. The intelligent ship management system based on automatic operation according to claim 6, characterized in that: The multi-objective dynamic optimization module uses the improved NSGA-Ⅲ algorithm to process multiple conflicting objectives. The conflicting objectives are multiple preset operation objectives that conflict with each other, specifically including: Performing clustering analysis based on the historical operation management decisions corresponding to the conflicting objectives, and generating multiple feature vectors through the results of the clustering analysis. The feature vectors are expressed as the operation management decisions in the historical operation management decisions; Adopting simulated binary crossover operation and polynomial mutation, and introducing a safety constraint violation penalty term in the selection pressure mechanism; Constructing a dynamic multi-objective optimization weight matrix to adjust the priorities of each conflicting objective.

8. The intelligent ship management system based on automatic operation according to claim 6, characterized in that: The multi-objective dynamic optimization module uses the improved ant colony algorithm to process multiple conflicting objectives, specifically including: The pheromone update rule in the improved ant colony algorithm is used to balance the selection probabilities of multiple conflicting objectives, specifically expressed as: ； Among them, is expressed as the pheromone concentration of the node corresponding to the conflicting target at the time step to ; The pheromone evaporation coefficient is expressed as ; is expressed as the pheromone concentration of the node corresponding to the conflicting target at the time step to ; The pheromone intensity constant is expressed as ; is expressed as the total probability of executing the conflicting target for the th time, and the risk coefficient corresponding to the conflicting target is expressed as The heuristic factor in the improved ant colony algorithm, specifically expressed as: ； Among them, represents the node corresponding to the conflict target The attraction to represents the node corresponding to the conflict target to the distance to represents the node corresponding to the conflict target to the steering angle to represents the execution direction of the conflict target.

Images