Green electricity supply ship and power receiving ship dynamic coordination and safety prevention and control method, device, computer equipment and storage medium

Abstract

The embodiment of the application relates to the field of ship electrical engineering, and provides a green electricity supply ship and power receiving ship dynamic cooperation and safety prevention and control method and device, computer equipment and storage medium, the method comprises the following steps: first, developing a power supply and power receiving cooperation control protocol for ship-shore low-delay communication to deploy a communication architecture, then constructing a multi-dimensional perception hazard identification model to output multi-dimensional perception hazard warning information, and adopting a two-level cooperative safety prevention and control architecture to construct a power supply and power receiving field large model to output two-layer hazard warning information, then modeling a Markov decision model and solving it to output sailing posture control information to control the execution of the charging docking operation, and finally, under the condition that the warning levels represented in the various hazard warning information meet the target level, the execution of the charging docking operation is stopped. The implementation of the method improves the cooperation efficiency and safety of the charging docking operation process.

Description

Technical Field

[0001] This application relates to the field of marine electrical engineering technology, and in particular to a method, device, computer equipment, and storage medium for dynamic coordination and safety control between a green electricity supply vessel and a power receiving vessel. Background Technology

[0002] In the field of marine electrical engineering technology, the shipping industry, as a significant source of global carbon emissions, is inevitably undergoing a green transformation. Green power supply vessels, with their advantages of flexible deployment, comprehensive coverage, and plug-and-play capability, are becoming a core supplementary equipment for zero-emission power supply systems in ports. However, in some busy ports, which receive a large number of cruise ships, the peak charging demand is high during peak tourist berthing periods, and the short berthing window for cruise ships places stringent requirements on charging efficiency and connection time.

[0003] Existing technologies for dynamic collaboration between mobile green energy supply vessels and receiving vessels, especially in typical scenarios such as peak load at cruise homeports, low load power replenishment at anchorages, extreme conditions during typhoons, and collaborative operations during busy port periods, still suffer from problems such as insufficient real-time performance and compatibility of ship-shore communication protocols, insufficient fusion capabilities of multi-source heterogeneous sensor data, insufficient precision in dynamic collaborative control between green energy supply vessels and receiving vessels, and weak safety and control systems. As a result, the collaborative efficiency and safety of the supply vessel and receiving vessel in the charging docking operation process are relatively low. Summary of the Invention

[0004] This application provides a method, apparatus, computer equipment, and storage medium for dynamic coordination and safety control between a green energy supply vessel and a power receiving vessel. More specifically, this application provides a method, apparatus, computer equipment, computer storage medium, and computer program product for dynamic coordination and safety control between a green energy supply vessel and a power receiving vessel, thereby improving the coordination efficiency and safety of the charging docking operation process.

[0005] In a first aspect, embodiments of this application provide a method for dynamic coordination and safety control between a green electricity supply vessel and a power receiving vessel, including:

[0006] Based on the protocol adaptive adaptation model, a power supply and receiving coordination control protocol for ship-shore low-latency communication is developed for the deployment of the communication architecture.

[0007] Based on the deployed communication architecture, multimodal sensing data is acquired during the charging docking operation between the supply vessel and the power receiving vessel. A multidimensional perception hazard identification model is constructed based on the multimodal sensing data to output multidimensional perception hazard early warning information associated with the charging docking operation.

[0008] Based on the multimodal sensing data and domain knowledge graph, a two-level collaborative security control architecture is adopted to construct a large model of the power supply and receiving domain, so as to output the global cognitive reasoning layer hidden danger warning information and the real-time control layer hidden danger warning information associated with the charging docking operation;

[0009] A Markov decision model is obtained by modeling the charging docking operation. The Markov decision model is solved by a dynamic cooperative matching algorithm based on reinforcement learning to output navigation attitude control information. The navigation attitude control information is used to control the execution of the charging docking operation.

[0010] If the warning level represented by the multi-dimensional perception hazard warning information, the global cognitive reasoning layer hazard warning information, and the real-time control layer hazard warning information meets the target level, the charging docking operation shall be stopped.

[0011] Optionally, in some embodiments of this application, the step of developing a power supply and receiving coordination control protocol for ship-shore low-latency communication based on a protocol adaptive adaptation model for deploying the communication architecture includes:

[0012] A protocol adaptive adaptation model is constructed based on protocol adaptation rules;

[0013] Based on the sum of the end-to-end total delay of the instruction and the delay of each process, a communication delay optimization model is constructed.

[0014] Based on the delay summation relationship represented in the delay optimization model as characterized in the protocol adaptive adaptation model, the power supply and receiving coordination control protocol for ship-shore low-latency communication is determined.

[0015] Optionally, in some embodiments of this application, the step of acquiring multimodal sensing data during the charging docking operation between the supply vessel and the power receiving vessel, constructing a multidimensional perception hazard identification model based on the multimodal sensing data, and outputting multidimensional perception hazard early warning information associated with the charging docking operation includes:

[0016] The multimodal sensing data collected during the charging docking operation is obtained. Based on the multimodal sensing data and the green electricity supply and receiving full life cycle operating condition data, a multimodal training dataset is constructed. A data cleaning model, a feature alignment model, and a small sample enhancement model are constructed to optimize the multimodal training dataset to obtain an optimized dataset.

[0017] A multimodal joint embedding model is constructed, and data transformation is performed on the optimized dataset to obtain a feature vector of a unified dimension. Based on the feature vector of the unified dimension, an improved multidimensional perception hazard identification model with an attention mechanism is constructed.

[0018] Optionally, in some embodiments of this application, the step of constructing a large model of the power supply and receiving domain based on the multimodal sensing data and domain knowledge graph using a two-level collaborative security control architecture, and outputting global cognitive reasoning layer hidden danger warning information and real-time control layer hidden danger warning information associated with the charging docking operation, includes:

[0019] Based on multimodal fusion features, prediction duration and world model, a global cognitive reasoning layer sub-model is constructed to output the hidden danger warning information and global security constraints of the global cognitive reasoning layer;

[0020] A lightweight model after distillation is used to construct a real-time control layer sub-model based on real-time collected high-frequency features and the global security constraints, so as to output the real-time control layer hidden danger early warning information.

[0021] Optionally, in some embodiments of this application, the method further includes:

[0022] Based on the hidden danger warning information of the global cognitive reasoning layer and the hidden danger warning information of the real-time control layer, cognitive reasoning ability is performed based on the large model to generate reasoning information of the root cause of the fault.

[0023] Based on the reasoning information of the root cause of the fault, domain knowledge, and historical handling cases, operation and maintenance handling suggestions are generated based on the operation and maintenance decision model.

[0024] Optionally, in some embodiments of this application, the step of modeling the charging docking operation to obtain a Markov decision model, and solving the Markov decision model based on a reinforcement learning-based dynamic cooperative matching algorithm to output navigation attitude control information, includes:

[0025] The charging docking operation is model defined to obtain the state space, action space, and multi-objective reward function;

[0026] The Markov decision model is determined based on the state space, the action space, and the multi-objective reward function.

[0027] A dynamic collaborative matching algorithm for reinforcement learning is constructed using a multi-agent deep reinforcement learning algorithm to solve the Markov decision model and output navigation attitude control information.

[0028] Optionally, in some embodiments of this application, the method further includes:

[0029] An adaptive power regulation control algorithm, a power dynamic allocation strategy for multiple ships supplying power simultaneously, and a grid-connected synchronization control strategy were constructed.

[0030] The execution of the charging docking operation is controlled by the adaptive power adjustment control algorithm, the power dynamic allocation strategy for simultaneous power supply from multiple ships, the grid-connected synchronization control strategy, and the navigation attitude control information.

[0031] Secondly, embodiments of this application provide a dynamic coordination and safety control device for green electricity supply vessels and power receiving vessels, which has the function of realizing the dynamic coordination and safety control method for green electricity supply vessels and power receiving vessels provided in the first aspect above. The function can be implemented by hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above function, and the modules can be software and / or hardware.

[0032] In one possible design, the device includes:

[0033] The communication architecture deployment module is used to develop a power supply and receiving coordination control protocol for ship-shore low-latency communication based on a protocol adaptive adaptation model, so as to deploy the communication architecture.

[0034] The multi-dimensional perception and early warning module is used to acquire multi-modal sensing data during the charging docking operation between the supply vessel and the power receiving vessel according to the deployed communication architecture, and to construct a multi-dimensional perception hazard identification model based on the multi-modal sensing data, so as to output multi-dimensional perception hazard early warning information associated with the charging docking operation.

[0035] The two-level collaborative early warning module is used to construct a large model of the power supply and receiving field based on the multimodal sensing data and the domain knowledge graph, using a two-level collaborative security control architecture, so as to output the global cognitive reasoning layer hidden danger early warning information and the real-time control layer hidden danger early warning information associated with the charging docking operation.

[0036] The charging docking control module is used to model the charging docking operation to obtain a Markov decision model, and solve the Markov decision model based on a dynamic cooperative matching algorithm of reinforcement learning to output navigation attitude control information; the navigation attitude control information is used to control the execution of the charging docking operation.

[0037] The alarm stop module is used to stop the execution of the charging docking operation when the warning level represented by the multi-dimensional perception hazard warning information, the global cognitive reasoning layer hazard warning information, and the real-time control layer hazard warning information meets the target level.

[0038] In another aspect, this application provides a computer device including at least one connected processor and a memory, wherein the memory is used to store program code, and the processor is used to call the program code in the memory to execute the methods described in the above aspects.

[0039] In another aspect, embodiments of this application provide a computer storage medium including instructions that, when executed on a computer, cause the computer to perform the methods described in the above aspects.

[0040] In another aspect, this application provides a computer program product containing instructions that, when run on a computer, cause the computer to perform the methods described in the above aspects.

[0041] Compared with traditional technologies, the technical solution of this application improves the effect of identifying hidden dangers under complex working conditions, enhances the interpretability of fault reasoning and operation and maintenance decisions, reduces the safety risks of power supply and receiving operations, and realizes safe and reliable collaborative operation of green power supply vessels and power receiving vessels throughout the entire process, thereby improving the collaborative efficiency and safety of the charging docking operation process. Attached Figure Description

[0042] Figure 1 This is a flowchart of one embodiment.

[0043] Figure 2 This is an overall flowchart of one embodiment.

[0044] Figure 3 This is a structural block diagram of the device in one embodiment.

[0045] Figure 4 This is an internal structural diagram of a computer device in one embodiment.

[0046] Figure 5 This is a diagram of the internal structure of a computer device in another embodiment. Detailed Implementation

[0047] The terms "first," "second," etc., used in the embodiments of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or modules is not necessarily limited to those steps or modules explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products, or devices. The division of modules appearing in the embodiments of this application is only a logical division. In actual applications, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not performed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be through some interface, and the indirect coupling or communication connection between modules may be electrical or other similar forms. None of these are limited in the embodiments of this application. Furthermore, the modules or sub-modules described as separate components may or may not be physically separated, may or may not be physical modules, or may be distributed among multiple circuit modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the embodiments of this application.

[0048] Figure 1 This is a flowchart illustrating one embodiment, such as... Figure 1 As shown in the embodiments of this application, the method for dynamic coordination and safety control between green electricity supply vessels and power receiving vessels includes:

[0049] S1, based on the protocol adaptive adaptation model, develops a power supply and receiving coordination control protocol for ship-shore low-latency communication for the deployment of the communication architecture.

[0050] Among them, the protocol adaptive adaptation model refers to the computational model that automatically matches and converts communication protocols for different ship types.

[0051] Among them, ship-to-shore low-latency communication refers to the communication method that enables millisecond-level command transmission between ship and shore; the power supply and receiving coordination control protocol, also known as the unified power supply and receiving coordination control protocol, refers to the standardized communication rules for the interaction of power supply and receiving commands between ship and shore.

[0052] Among them, the communication architecture refers to the overall deployment structure of ship-shore multi-link redundant communication.

[0053] S2, based on the deployed communication architecture, acquires multimodal sensing data during the charging docking operation between the supply vessel and the power receiving vessel, constructs a multidimensional perception hazard identification model based on the multimodal sensing data, and outputs multidimensional perception hazard early warning information related to the charging docking operation.

[0054] Multimodal sensing data refers to various types of operational monitoring data, including electrical, interface, environmental, and ship status data. More specifically, the monitoring dimensions of multimodal sensing data include: electrical parameter monitoring (voltage, current, power, frequency, phase difference), interface status monitoring (contact pressure, interface temperature, insertion / removal stroke, wear), environmental status monitoring (ambient humidity, salt spray concentration, wind and wave level, visibility), and ship status monitoring (ship attitude, relative position, draft, electrical load, channel position).

[0055] Among them, the multi-dimensional perception hazard identification model, also known as the multi-modal joint embedding model and the AI ​​multi-dimensional perception model, refers to an AI model that integrates multi-modal data to identify power supply hazards.

[0056] Among them, the multi-dimensional perception hazard early warning information refers to the hazard level prompt information output by the model in the charging docking operation.

[0057] S3, based on multimodal sensor data and domain knowledge graph, adopts a two-level collaborative security control architecture to construct a large model for the power supply and receiving domain, so as to output global cognitive reasoning layer hidden danger warning information and real-time control layer hidden danger warning information related to charging docking operations.

[0058] Among them, the domain knowledge graph refers to a structured knowledge base composed of rules and relationships in the power supply and receiving domains.

[0059] Among them, the two-level collaborative security control architecture refers to a security control framework that combines a global cognitive reasoning layer and a real-time management and control layer.

[0060] Among them, the large-scale power supply and receiving model refers to a dedicated cognitive reasoning model for ship power supply scenarios.

[0061] Among them, the global cognitive reasoning layer hazard warning information refers to long-term time-series global risk prediction and alert information. The real-time control layer hazard warning information refers to on-site high-frequency risk monitoring and alert information.

[0062] S4. A Markov decision model is obtained by modeling the charging docking operation. The Markov decision model is solved by a dynamic cooperative matching algorithm based on reinforcement learning to output navigation attitude control information. The navigation attitude control information is used to control the execution of the charging docking operation.

[0063] Among them, modeling the charging docking operation refers to the operation of abstracting the ship docking power supply process into a mathematical model; Markov decision model refers to the time series model describing the ship's dynamic docking decision process.

[0064] Among them, the reinforcement learning-based dynamic collaborative matching algorithm, also known as the reinforcement learning-based AI dynamic collaborative matching algorithm, refers to the reinforcement learning-based ship attitude precision control algorithm.

[0065] Among these, navigation attitude control information refers to control commands such as the ship's position, heading, and inclination, used to enable the green electricity supply vessel to autonomously plan its navigation path, achieve precise six-degree-of-freedom attitude control, and dynamically match its relative position from anchorage to the cruise home port. Specifically, navigation attitude control information includes control commands for the propellers, side thrusters, and rudder.

[0066] S5, if the warning level represented by the multi-dimensional perception hazard warning information, the global cognitive reasoning layer hazard warning information, and the real-time control layer hazard warning information meets the target level, the charging docking operation shall be stopped.

[0067] The warning level refers to the risk level classified according to the degree of danger of the hidden danger. For example, the warning level includes level one, level two, and level three.

[0068] The target level refers to the threshold for determining potential risks that trigger emergency protection, while stopping charging docking operations refers to the safety protection action of cutting off power supply and terminating docking. For example, if the target level is level three, an emergency disconnection will be executed when the warning level reaches level three, stopping the charging docking operation.

[0069] Compared to traditional technologies, this application's embodiments first develop a low-latency ship-to-shore power supply and receiving collaborative control protocol for deploying the communication architecture. Then, a multi-dimensional perception hazard identification model is constructed to output multi-dimensional perception hazard early warning information. A two-level collaborative safety control architecture is adopted to construct a large-scale model for the power supply and receiving domain, outputting two layers of hazard early warning information. A Markov decision model is then modeled and solved to output navigation attitude control information to control the execution of the charging docking operation. Finally, if the warning level represented by each hazard early warning information meets the target level, the charging docking operation is stopped. The technical solution of this application's embodiments improves the effectiveness of hazard identification under complex working conditions, enhances the interpretability of fault reasoning and operation and maintenance decisions, reduces the safety risks of power supply and receiving operations, and achieves safe and reliable collaborative operation of the green electricity supply vessel and the receiving vessel throughout the entire process, thereby improving the collaborative efficiency and safety of the charging docking operation process.

[0070] In a specific embodiment, step S1 can also be described as: constructing a power supply and receiving collaborative control protocol and communication architecture based on ship-shore low-latency communication.

[0071] Step S1 addresses the core issues of insufficient real-time performance and compatibility in ship-shore communication by constructing a three-pronged ship-shore communication system: multi-link redundant communication, standardized protocol adaptation, and low-latency optimization. This system achieves full communication coverage across the entire voyage from port anchorages to the cruise home port, adapts to the high-concurrency communication demands during peak port hours, and provides a stable, low-latency, and highly reliable communication foundation for power supply and receiving coordinated control. This addresses the key technical challenges of ship-shore collaborative communication. Step S1 specifically includes S11, S12, and S13.

[0072] S11. Selection and Redundancy Deployment of Ship-Shore Collaborative Communication System.

[0073] The selection and deployment of a multi-link redundant communication system for green electricity supply vessels, receiving vessels, and shore-based control centers were completed. The core configuration includes: the main link adopts 5G+TSN (Time-Sensitive Networking) fiber optic communication, supplemented by shore-based 5G base station solutions in various ports (such as Xiamen Port); the backup link adopts a maritime satellite communication enhancement solution to achieve full-area communication coverage without dead zones in port anchorages, waterways, and cruise homeport operation areas; the IEEE 1588 PTP high-precision time synchronization protocol is deployed to achieve a ship-shore data sampling time deviation of ≤1ms, ensuring the time synchronization of power supply and receiving coordinated control commands; and electromagnetic shielding, moisture-proof and anti-interference hardware design is provided to adapt to the complex operating environment of ports with high humidity, strong electromagnetic interference, and high salt spray.

[0074] Establish unified standards for multi-disciplinary data in green electricity supply and reception: time-series electrical and sensor data adopt TSDB format, unstructured operation and maintenance logs and safety specification text data adopt JSON format, ship power system topology and physical model data adopt STEP format, and port hydrological, meteorological and berthing condition data adopt NetCDF format, so as to achieve standardized definition and unified interaction of multi-source heterogeneous data.

[0075] S12, Development of a unified power supply and receiving coordination control protocol.

[0076] Based on the ISO 18131 international standard for ship-to-shore power communication, a unified power supply and receiving coordination control protocol with a publish-subscribe architecture was developed to achieve seamless compatibility between the green power supply vessel and the IPMS system of the receiving vessel. The core of the protocol defines four major interaction logics: ① Ship parameter matching interaction during the pre-docking phase (rated voltage, frequency, rated power, interface specifications, charging time requirements, etc. of the receiving vessel); ② Ship-to-shore scheduling and coordinated collision avoidance command interaction during the navigation phase; ③ Attitude synchronization and motion coordination command interaction during the docking phase; ④ Real-time synchronization of power regulation, status monitoring, and emergency protection commands during the power supply phase.

[0077] A protocol adaptive adaptation model is constructed to achieve plug-and-play compatibility between communication interfaces and protocols of different ship types. The model is as follows:

[0078] ;

[0079] In the formula, For the native communication protocol and data format of the powered ship, As per the ISO 18131 standard protocol specification, This is a protocol feature matching function used to complete the feature mapping and compatibility verification between the powered ship protocol and the standard protocol. This forms the basic framework of the collaborative control protocol of the present invention. This is a protocol conversion function that enables lossless conversion between native protocols of different ship types and the unified collaborative control protocol; This enables protocol adaptation and splicing operations, achieving dynamic adaptation and seamless integration of protocol interfaces.

[0080] S13, Low latency and high reliability optimization for ship-to-shore communication.

[0081] To address the millisecond-level response requirements of power supply and receiving coordinated control, a communication latency optimization model is constructed to control the end-to-end transmission latency of coordinated control commands. The model is as follows:

[0082] ;

[0083] In the formula, The total end-to-end delay of the instruction. The time consumed by instruction encoding For link transmission time, The time consumed by instruction decoding The time consumed for time synchronization calibration; the model optimization objective is to minimize the total end-to-end delay of the cooperative control commands. Emergency protection command transmission delay The packet loss rate is ≤0.001%, meeting the real-time requirements of power supply and receiving coordinated control and emergency protection.

[0084] Simultaneously employing edge computing node local data processing, lightweight protocol frame compression, and redundant link seamless switching technologies, a dynamic scheduling mechanism for communication resources is designed to address communication congestion scenarios during peak port hours. This ensures priority transmission of emergency protection commands and collaborative control commands, avoids command delays under high concurrency, and further enhances the stability and anti-interference capabilities of the communication link.

[0085] Optionally, in some embodiments of this application, a power supply and receiving coordination control protocol for ship-shore low-latency communication is developed based on a protocol adaptive adaptation model for deployment of the communication architecture. This includes: constructing a protocol adaptive adaptation model based on protocol adaptation rules; constructing a communication latency optimization model based on the latency summation relationship between the total end-to-end latency of the command and the latency of each process; and determining the power supply and receiving coordination control protocol for ship-shore low-latency communication according to the latency summation relationship represented in the latency optimization model as represented in the protocol adaptive adaptation model.

[0086] Among them, protocol adaptation rules refer to the constraint criteria for matching and converting different ship communication protocols. For example, the mathematical expression of protocol adaptation rules is:

[0087] ; For the native communication protocol and data format of the powered ship, As a standard protocol specification, For protocol feature matching function; This forms the basic framework for collaborative control protocols. This is a protocol conversion function; For protocol adaptation splicing operations.

[0088] Among them, the total end-to-end delay of the instruction This refers to the total time elapsed from sending to receiving an instruction. Process latency refers to the time spent on instruction encoding / decoding, transmission, synchronization, and other related processes. Process latency includes... Instruction encoding time Link transmission time Instruction decoding time Time synchronization calibration time. The communication latency optimization model refers to a mathematical model that calculates and reduces the total time spent on command transmission. Specifically, the mathematical expression of the model can be:

[0089] ;

[0090] In this embodiment, by constructing a protocol adaptive adaptation model and a communication latency optimization model, communication transmission efficiency is improved, command latency is reduced, ship-shore interaction reliability is enhanced, and low-latency stable communication is achieved.

[0091] In a specific embodiment, step S2 can also be described as: constructing a multimodal fusion AI multidimensional perception and real-time monitoring system for power supply and receiving hazards.

[0092] Step S2 addresses the core issues of poor adaptability and insufficient hazard identification capabilities of AI monitoring models under complex operating conditions. It constructs an intelligent sensing system encompassing "multimodal data fusion, unified feature embedding, AI hazard identification, and tiered early warning triggering." This system enhances feature extraction and hazard identification capabilities under extreme environments, particularly during complex conditions such as typhoons, high tides, and peak hours at various ports. It adapts to the full-condition monitoring needs of low-load power replenishment at anchorages and high-power fast charging at cruise ship homeports, enabling real-time monitoring, early identification, and precise location of hazards throughout the power supply and reception process. This addresses key technical challenges related to multi-source heterogeneous data fusion and intelligent sensing. Step S2 specifically includes S21, S22, and S23.

[0093] S21. Deployment and dataset construction of a multimodal sensing system for the entire power supply and receiving process.

[0094] The selection, calibration, and deployment of a multimodal sensing system covering the entire green electricity supply and reception process have been completed. The core monitoring dimensions include: electrical parameter monitoring (voltage, current, power, frequency, phase difference), interface status monitoring (contact pressure, interface temperature, insertion and removal stroke, wear), environmental status monitoring (ambient humidity, salt spray concentration, wind and wave level, visibility), and ship status monitoring (ship attitude, relative position, draft, electrical load, and channel position), achieving full coverage of data collection for all operating conditions and dimensions during navigation, berthing, docking, and power supply.

[0095] By combining green electricity supply and reception data throughout their entire lifecycle, a multimodal training dataset is constructed. This dataset includes: multi-source sensor time-series data, ship power system operation data, hydrological, meteorological, and waterway geographic data for various ports, port traffic flow and berthing operation data, textual data of ship power supply and reception safety specifications and design standards, historical power supply and reception operation data, electrical fault and hazard handling cases, and expert operation and maintenance experience data. The dataset covers all scenarios of operation for various ports, including typhoon days, rainstorm days, high tide levels, and busy waterway periods, as well as typical power supply and reception conditions such as low-load power replenishment at anchorages and simultaneous power supply to multiple ships at cruise homeports.

[0096] A data cleaning, feature alignment, and few-shot augmentation model is constructed to optimize the dataset. The model is as follows:

[0097] ;

[0098] In the formula, For multimodal sensor operating data, For ship electrical system operation data, For port environmental and hydro-meteorological data, For data on berthing and power load of powered vessels, To provide power receiving safety specifications and design standards text data, This includes data on historical fault and potential hazard handling cases; It is a spatiotemporal alignment function for multi-source data, which realizes format unification and feature association of heterogeneous data through spatiotemporal benchmark calibration and semantic matching; This data represents low-quality / small sample extreme operating conditions (such as short-circuit faults, hidden poor contacts, and extreme wind and wave conditions during typhoons). For data augmentation functions, methods such as time-series feature interpolation, fault scenario synthesis, and working condition disturbance simulation are used to expand the dataset and solve the problem of small sample training. This is a feature splicing operation that enables deep fusion of multimodal features.

[0099] S22. Construction of a multimodal feature fusion and AI-based hazard identification model.

[0100] By integrating the implicit features of multi-source sensor data with the explicit rules of power supply and receiving domain knowledge, a multimodal joint embedding model and an AI multidimensional perception model are constructed to achieve real-time identification and accurate location of potential power supply and receiving hazards.

[0101] First, a multimodal joint embedding model is constructed to transform monitoring data of different dimensions and modalities into a unified-dimensional feature vector. The model is as follows:

[0102] ;

[0103] In the formula, To embed the temporal features of multimodal sensor data, the TCN temporal convolutional network is used to extract anomalous change features of time-series data such as voltage, current, temperature, and pressure. Electrical features are embedded into ship power system data, and electrical anomalies such as power fluctuations, phase deviations, and harmonic distortions are extracted using a CNN convolutional neural network. To embed the temporal features of port environmental data, environmental features and fluctuation patterns such as humidity, salt spray, wind and waves are extracted through an LSTM network. To embed the spatiotemporal features of the powered vessel status data, a graph neural network (GNN) is used to extract the dynamic changes in the vessel's attitude, relative position, electrical load, and channel position. To embed semantic features of power supply and receiving safety specification text data, a BERT pre-trained model is used to generate the semantic logic of safety rules and hazard judgment standards. To achieve structured embedding of knowledge graphs in the power supply and receiving domains, the TransE algorithm is used to transform entities and relationships into vector representations, incorporating domain knowledge association information. The modality weight matrix is ​​adaptively adjusted through training to optimize feature extraction weights for extreme scenarios such as typhoon days and high humidity, highlighting the contribution of each modality to the hazard identification task. The knowledge embedding fusion coefficient is set to a value of 0.4 to 0.8, with 0.6 being the preferred value, to balance the fusion ratio of multimodal data features and structured knowledge.

[0104] Based on unified embedded features (a model with unified feature vectors), an improved VisionTransformer (VIT) AI multidimensional perception hazard identification model with an attention mechanism is constructed to enhance the ability to extract weak features of hidden faults. This model enables real-time identification of hazards such as short circuits, leakage, poor contact, interface overheating, and abnormal voltage. The hazard identification accuracy is ≥99.5%, and the early detection lead time for poor contact is ≥30 seconds, allowing sufficient time for emergency response.

[0105] S23. Design of a hazard classification, early warning, and emergency triggering mechanism.

[0106] Based on the hazard identification results, a three-level early warning and triggering mechanism is constructed: Level 1 early warning targets latent hazards such as early signs of poor contact and slight temperature rise, triggering audible and visual warnings and maintenance prompts; Level 2 degrading targets moderate hazards such as excessive current fluctuations and abnormal contact pressure, triggering degraded protection such as power derating and dynamic load limiting; Level 3 emergency disconnection targets severe sudden faults such as short circuits, leakage, and arcing, triggering a millisecond-level emergency disconnection mechanism to cut off the power supply circuit and ensure the safety of equipment and personnel.

[0107] Optionally, in some embodiments of this application, multimodal sensing data is acquired during the charging docking operation between the supply vessel and the power receiving vessel, and a multidimensional perception hazard identification model is constructed based on the multimodal sensing data to output multidimensional perception hazard early warning information associated with the charging docking operation. This includes: collecting multimodal sensing data during the charging docking operation; constructing a multimodal training dataset based on the multimodal sensing data and green electricity supply and receiving full life cycle operating condition data; optimizing the multimodal training dataset by constructing a data cleaning model, a feature alignment model, and a few-shot augmentation model to obtain an optimized dataset; constructing a multimodal joint embedding model; performing data transformation based on the optimized dataset to obtain a feature vector of a unified dimension; and constructing an improved multidimensional perception hazard identification model that incorporates an attention mechanism based on the feature vector of the unified dimension.

[0108] Among them, the green electricity supply and reception full life cycle operating condition data refers to the data related to the operating status of the entire power supply process; the multimodal training dataset refers to the collection of multiple types of monitoring data used for model training.

[0109] Among them, the data cleaning model refers to the optimization model that removes abnormal noise data; the feature alignment model refers to the model that achieves unified spatiotemporal correlation of multi-source data; the small sample augmentation model refers to the optimization model that expands the amount of scarce working condition data; and the optimized dataset refers to the standardized training data after cleaning, alignment and augmentation.

[0110] Among them, the multimodal joint embedding model refers to a model that transforms multiple types of data into a unified feature vector; the unified dimensional feature vector refers to the standardized feature representation after the transformation of different modal data; and the attention mechanism refers to the algorithm mechanism that highlights key features to improve recognition accuracy.

[0111] Among them, the improved multidimensional perception hazard identification model that introduces an attention mechanism refers to the improved Vision Transformer (VIT) AI multidimensional perception hazard identification model that introduces an attention mechanism, which is used to output multidimensional perception hazard early warning information.

[0112] In this embodiment, data optimization and multimodal modeling improved the accuracy of hazard identification, enhanced the ability to perceive complex working conditions, and improved the accuracy of early warning.

[0113] In a specific embodiment, step S3 can also be described as the construction of a full-process security control system of "AI real-time monitoring + large model inference".

[0114] Step S3 addresses the core issues of insufficient real-time performance, interpretability, and fault tolerance in safety control. It constructs a two-tiered collaborative safety control architecture: a large-scale model for global cognitive reasoning and a small-scale model for real-time control. This involves building full-scenario world models for each port to enable operational process simulation and risk prediction, achieving in-depth reasoning of fault root causes, generating interpretable maintenance decisions, and implementing closed-loop safety control throughout the entire process. This addresses key technical and theoretical issues related to power supply and reception safety control. Step S3 specifically includes S31, S32, and S33.

[0115] S31. Large-scale model construction and RAG retrieval enhancement mechanism in the power supply and receiving field.

[0116] Based on the constructed multimodal dataset of green electricity supply and reception and the domain knowledge graph, a large-scale model of the shipping vertical domain is used for pre-training and fine-tuning to build a large-scale model dedicated to power supply and reception, realizing three core capabilities: deep reasoning of fault root causes, optimization of operation and maintenance decisions, and global risk assessment. Based on the geographical, hydrological and meteorological, port operation rules, and historical ship operation data of each port, a full-scenario world model of green electricity supply and reception for each port is constructed to realize the pre-simulation of operation processes and risk prediction under different operation scenarios and extreme conditions.

[0117] By introducing the RAG retrieval enhancement mechanism and constructing a dedicated knowledge base for the power supply and receiving domains, the system can retrieve and match safety regulations, fault cases, operation and maintenance experience, handling procedures, and scenario-based operation plans for various ports in real time during the large model inference process. This provides accurate domain knowledge support for global inference, suppresses the illusion of large models, and ensures that the inference results comply with ship electrical safety regulations and port operation requirements.

[0118] S32, Two-level collaborative security control architecture design.

[0119] To address the security management needs at different levels of green electricity supply and reception, a two-tier collaborative security control architecture consisting of a "global cognitive reasoning layer and a real-time control layer" is constructed. This architecture achieves the rational allocation of computing resources and the precise matching of security management. The model is as follows:

[0120] ;

[0121] In the formula, For the global cognitive reasoning layer, a dual-driven architecture of "large-scale power supply and receiving model + full-scene world model of port" is adopted, based on multi-modal fusion features. Predicted duration Based on the world model, it achieves three core functions: First, long-term time-series prediction of risks throughout the power supply and reception process, including risk evolution caused by power load fluctuations, interface condition degradation, and environmental changes; second, operation process simulation, fault simulation, and risk pre-verification based on the world model, simulating the system response and risk evolution of different operation strategies and fault handling strategies under different scenarios such as typhoon days, high tides, and busy periods, predicting potential hazards, and optimizing operation plans and safety control strategies in advance; third, global safety control strategy optimization, outputting global safety constraints (including safe voltage / frequency range, interface temperature threshold, safe contact pressure range, and emergency response boundary conditions); among them, the world model of the entire scenario of green power supply and reception in ports integrates the geography, hydrology and meteorology, port traffic flow, ship power systems, and mechanical docking mechanisms of various ports, and simulates the dynamic impact of changes in electrical parameters, mechanical disturbances, and environmental interference on the power supply and reception process in real time.

[0122] For the real-time control layer, a lightweight AI model after distillation is used, based on high-frequency features acquired in real time. (Sampling frequency 50Hz~200Hz) and global security constraints It achieves millisecond-level real-time hazard monitoring, graded protection triggering, and emergency disconnection control, with an emergency protection response time of ≤50ms; To achieve the fusion of results and realize the hierarchical correlation and coordination between global safety optimization goals and real-time control actions, the world model pre-simulation results provide pre-verification data support for the global cognitive reasoning layer, and complete the risk simulation and optimization of the operation plan before the actual operation, thereby reducing the risk of the actual operation.

[0123] Simultaneously, a layered triggering and updating mechanism is designed. The global cognitive reasoning layer updates once every 10 seconds and once every 30 seconds, triggering emergency updates only when there are sudden changes in operating conditions, escalation of hidden dangers, or sudden changes in the environment. The control frequency of the real-time control layer is 50Hz-200Hz, realizing continuous closed-loop control and balancing the depth of global reasoning with the response speed of real-time control.

[0124] S33, Fault Root Cause Reasoning and Interpretable Operation and Maintenance Decision Generation.

[0125] Based on the cognitive reasoning capabilities of the large model and combined with the results of hazard identification, it can accurately reason about the root causes of faults, and locate core fault sources such as poor contact, aging lines, environmental humidity, mechanical wear, and parameter mismatch. It introduces interpretable AI (XAI) technology, and uses SHAP value and LIME method to generate an interpretable report with fault reasoning logic, risk basis, and handling priority, which solves the problem of insufficient interpretability of black box models and meets the needs of ship safety supervision and operation and maintenance decision-making.

[0126] An AI-powered operation and maintenance decision-making model is built. Based on the root cause reasoning results of faults, combined with domain knowledge and historical handling cases, it pushes accurate operation and maintenance handling suggestions, including cleaning interface dust, replacing aging lines, optimizing sealing protection, calibrating docking mechanisms, and adjusting power supply parameters. At the same time, it generates operation and maintenance operation manuals and safety precautions, realizing a closed loop of the entire process of "hazard identification - root cause reasoning - handling suggestions - operation and maintenance implementation".

[0127] Optionally, in some embodiments of this application, a two-level collaborative security control architecture is adopted based on multimodal sensing data and domain knowledge graphs to construct a large model of the power supply and receiving domain, so as to output global cognitive reasoning layer hidden danger warning information and real-time control layer hidden danger warning information related to charging docking operations. This includes: constructing a global cognitive reasoning layer sub-model based on multimodal fusion features, prediction duration and world model, so as to output global cognitive reasoning layer hidden danger warning information and global security constraints; and using a distilled lightweight model, constructing a real-time control layer sub-model based on real-time collected high-frequency features and global security constraints, so as to output real-time control layer hidden danger warning information.

[0128] Among them, multimodal fusion features This refers to the comprehensive feature information obtained after fusing multiple types of monitoring data. Multimodal fusion features can be obtained from multimodal sensor data through feature extraction. Prediction duration. This refers to the length of time the model takes to anticipate risks. A world model is a digital model that simulates the port operating environment and ship operations.

[0129] Among them, the global cognitive reasoning sub-model refers to the upper-level model that realizes global risk inference; the real-time control sub-model refers to the lower-level model that realizes rapid on-site response. Global safety constraints refer to the overall safety boundary conditions that power supply operations must follow.

[0130] The distilled lightweight model refers to a compressed and simplified real-time running small model. High-frequency features refer to real-time operating status features acquired at high frequencies, which can be obtained through feature extraction based on high-frequency acquired multimodal sensor data.

[0131] In this embodiment, the collaboration of the two-layer sub-model improves the accuracy of risk prediction, increases the speed of emergency response, enhances the real-time performance of the system, and achieves safety management that takes into account both the overall situation and the actual situation.

[0132] It should be noted that the multi-dimensional perception hazard warning information in step S2 adopts an output logic based on multimodal sensor data, emphasizing rapid and comprehensive output of warnings; while the global cognitive reasoning layer output hazard warning information in S3 includes risks throughout the power supply and receiving process and the hazard level of the natural environment, emphasizing warnings over a large area; the real-time control layer outputs hazard warning information, emphasizing the hazard level of the ship itself, such as a malfunction of a piece of equipment on the ship, emphasizing warnings within a small area. The two levels of steps S2 and S3 output hazard warning information with different judgment methods, different warning ranges, and different output frequencies, improving the richness and comprehensiveness of the warnings.

[0133] Optionally, in some embodiments of this application, the method further includes: performing cognitive reasoning based on a large model to generate reasoning information about the root cause of the fault, based on the global cognitive reasoning layer hidden danger warning information and the real-time control layer hidden danger warning information; and generating operation and maintenance handling suggestions based on the operation and maintenance decision model, based on the reasoning information about the root cause of the fault, domain knowledge and historical handling cases.

[0134] Among them, cognitive reasoning ability refers to the ability of large models to analyze and judge the causes of failures; the reasoning information of the root cause of failures refers to the information on the causes of failures derived from model analysis.

[0135] Among them, historical handling cases refer to experience examples of handling past power supply failures; operation and maintenance decision model refers to intelligent decision-making model that generates failure handling solutions; operation and maintenance handling suggestions refer to specific handling guidance given for failures.

[0136] In this embodiment, the accuracy of fault location is improved, the efficiency of operation and maintenance is enhanced, the rationality of the handling plan is increased, and the cost of fault investigation is reduced by using large model reasoning and decision generation.

[0137] In a specific embodiment, step S4 can also be described as: dynamic collaborative matching and adaptive control between the green electric ship and the powered ship under wind and wave disturbance.

[0138] Step S4 addresses the core issues of insufficient precision in dynamic collaborative control and poor stability in docking and power supply adaptation. It constructs a dynamic collaborative matching algorithm and adaptive control strategy for ships under wind and wave disturbances. This enables green electricity supply vessels to autonomously plan their navigation routes from various port anchorages to the cruise home port, achieve precise berthing and docking control under wind and wave disturbances, and dynamically allocate power for simultaneous power supply from single or multiple ships. It ensures operational efficiency requirements with charging time ≤ 1 hour and adapts to the wind-resistant stability control needs of berthing during typhoons, thus solving key technical and theoretical problems in cross-entity dynamic collaborative control of ships. Step S4 specifically includes S41, S42, and S43.

[0139] S41. Modeling of Markov decision process for ship relative motion and docking process.

[0140] The entire process of green electricity supply vessel's navigation from anchorage to home port, attitude alignment with the receiving vessel, interface docking, and dynamic power supply adaptation is modeled as a Markov decision process (MDP), defining the state space, action space, and multi-objective reward function, providing a foundation for optimizing reinforcement learning collaborative control strategies.

[0141] state space :

[0142] ;

[0143] in The relative lateral and longitudinal positions of the green electricity supply vessel and the power receiving vessel. Relative heading angle The roll angle, The pitch angle is the angle of inclination. For relative lateral and longitudinal velocities, To ensure the contact pressure at the interface, To supply power output voltage and current, To supply the state of charge of the ship's energy storage system for green electricity, The current operating mode is (navigation mode, pre-docking mode, docking mode, power supply mode, emergency mode). Real-time location within the waterway. Provides information on the movement of surrounding vessels.

[0144] Action space : ;

[0145] in For the target rotational speed of the thruster, For the target thrust of the side thruster, The target value for the rudder angle. To align with the organization's target itinerary, To supply the target voltage and current, Power allocation factor for multiple ships The target operation mode.

[0146] The multi-objective reward function design, based on Pareto optimality theory, balances six core objectives: navigation efficiency, docking accuracy, attitude stability, power supply stability, operational efficiency, and operational safety. The model is as follows:

[0147] ;

[0148] In the formula, The reward is based on navigation efficiency, calculated according to channel transit time and path planning deviation; the higher the transit efficiency, the higher the reward. The reward for attitude alignment and docking accuracy is calculated based on the lateral and longitudinal deviations between the actual relative position and the target docking position; the smaller the deviation, the higher the reward. The reward for ship attitude stability is calculated based on the fluctuation range of roll and pitch angles. The smaller the fluctuation, the higher the reward. The weight is adaptively increased under typhoon conditions. The reward for power supply stability is calculated based on the deviation of the output voltage and frequency from the rated values; the smaller the deviation, the higher the reward. The reward is based on the docking time and charging time. The shorter the time, the higher the reward. The weight is automatically increased in the fast charging mode at the cruise home port. To ensure operational safety, a negative penalty is imposed when the vessel's attitude, docking pressure, electrical parameters, or waterway navigation status exceed safety thresholds, thus ensuring that control actions comply with safety constraints. The weighting coefficients are dynamic and adaptively adjusted based on real-time operation modes, port environment, and operational requirements, especially during navigation. Weighting increase, in the docking mode Weighting increase, power supply mode Increased weight, emergency mode With all weights set to maximum, the sum of the weights is 1.

[0149] S42. Construction of AI dynamic collaborative matching algorithm based on reinforcement learning.

[0150] To address the need for autonomous navigation and precise attitude alignment of ships under wind, wave, and tidal disturbances, a multi-agent deep reinforcement learning algorithm (MADDPG) is employed to construct an AI dynamic collaborative matching algorithm. This algorithm enables green electricity supply vessels to autonomously plan their navigation path, achieve precise six-degree-of-freedom attitude control, and dynamically match relative positions from anchorage to the cruise home port. The algorithm takes the ship's navigation state, relative motion state, environmental disturbance parameters, and surrounding vessel dynamics as inputs, and outputs control commands from the thrusters, side thrusters, and rudder. The model is pre-trained using a virtual testbed based on a port world model, covering various wind and wave levels, tidal current speeds, ship tonnage, and channel congestion conditions, thereby improving the model's generalization ability and resilience.

[0151] In the AI ​​dynamic collaborative matching algorithm based on reinforcement learning, a model predictive control (MPC) rolling optimization mechanism is introduced. Based on the mathematical model of ship relative motion, it predicts the changes in the ship's relative position and attitude in the future time domain. Within the control quantity constraints and safety constraint boundaries, the control commands are rolled and optimized. For berthing scenarios in typhoon weather, the wind-resistant and stable control strategy is optimized to ensure that under wind and wave conditions of level 6 and below, the attitude alignment accuracy meets the following requirements: lateral deviation ≤ 5cm, longitudinal deviation ≤ 10cm, and relative heading deviation ≤ 0.5°, achieving millimeter-level precise docking with a docking time ≤ 10 minutes.

[0152] S43, Dynamic adaptive control of power supply process.

[0153] To address the dynamic fluctuations in the electrical load of powered vessels and the simultaneous power supply demands of multiple vessels, an adaptive fuzzy PID power regulation control algorithm is constructed. Based on the real-time electrical load, voltage-frequency deviation, and charging time requirements of the powered vessels, the PID control parameters are adjusted online to achieve dynamic adaptive regulation of the power output. The power regulation response time is ≤5s, voltage fluctuation is controlled within ±2%, and frequency deviation is controlled within ±0.2Hz, ensuring the stability and reliability of the power supply process.

[0154] A dynamic power allocation strategy for simultaneous power supply to multiple ships can be constructed, which can provide stable power to 1-2 cruise ships at the same time. Based on the power load demand, charging time requirements, and battery charge status of the receiving ships, the output power of each power supply circuit is dynamically adjusted to ensure that the charging time of a single ship is ≤1 hour, while achieving load balancing and stable control of power supply to multiple ships.

[0155] Simultaneously, a grid-connected synchronous control strategy is constructed to achieve seamless grid connection between the green electricity supply ship's output power and the power receiving ship's electrical system, ensuring that the phase difference is ≤3°, the frequency difference is ≤0.2Hz, and the voltage difference is ≤5% during grid connection, thus avoiding damage to the ship's electrical equipment caused by grid connection shocks.

[0156] Optionally, in some embodiments of this application, a Markov decision model is obtained by modeling the charging docking operation, and a dynamic cooperative matching algorithm based on reinforcement learning is used to solve the Markov decision model to output navigation attitude control information. This includes: performing model definition processing on the charging docking operation to obtain the state space, action space, and multi-objective reward function; determining the Markov decision model based on the state space, action space, and multi-objective reward function; and constructing a dynamic cooperative matching algorithm based on reinforcement learning using a multi-agent deep reinforcement learning algorithm to solve the Markov decision model to output navigation attitude control information.

[0157] Among them, model definition processing refers to the modeling operation of defining variables and constraints for the decision model; state space refers to the set of all possible operating states of the ship; action space refers to the set of all control actions that the ship can perform; and multi-objective reward function refers to a multi-index evaluation function for evaluating the merits of control strategies.

[0158] Among them, the multi-agent deep reinforcement learning algorithm, namely the multi-agent deep reinforcement learning algorithm MADDPG, is an algorithm for multi-ship collaborative optimization.

[0159] In this embodiment, decision modeling and reinforcement learning are used to improve the accuracy of ship docking, increase operational efficiency, enhance adaptability to wind and waves, and facilitate precise and stable docking.

[0160] Optionally, in some embodiments of this application, the method further includes: constructing an adaptive power regulation control algorithm, a power dynamic allocation strategy for simultaneous power supply from multiple ships, and a grid-connected synchronization control strategy; and controlling the execution of the charging docking operation based on the adaptive power regulation control algorithm, the power dynamic allocation strategy for simultaneous power supply from multiple ships, the grid-connected synchronization control strategy, and the navigation attitude control information.

[0161] Among them, the adaptive power regulation control algorithm, also known as the adaptive fuzzy PID power regulation control algorithm, refers to an algorithm that automatically adjusts the power supply according to the load.

[0162] Among them, the dynamic power distribution strategy for multiple ships supplying power simultaneously refers to the balanced distribution scheme of power supply loads from multiple ships; the grid-connected synchronous control strategy refers to the phase and frequency control scheme for achieving stable grid connection of power sources.

[0163] In this embodiment, the stability of power output and the balance of power supply across multiple ships are improved through the coordinated use of multiple control strategies.

[0164] The technical research process and other technical details of this application are described below with reference to a specific embodiment.

[0165] In traditional technologies, the shipping industry, as a significant source of global carbon emissions (accounting for approximately 3% of global anthropogenic carbon emissions), is inevitably undergoing a green transformation. Using shore power instead of traditional diesel generators while ships are in port is a core measure to reduce emissions of sulfur oxides, nitrogen oxides, particulate matter, and greenhouse gases from ships in ports. Fixed shore power systems are limited by the difficulty of dock renovations, inability to cover anchored vessels, and poor compatibility with various ship types. Green electricity supply vessels, with their advantages of flexible deployment, full coverage, and plug-and-play functionality, have become a core supplementary equipment for zero-emission power supply systems in ports, and their application scenarios continue to expand.

[0166] Taking Xiamen Port as an example, its cruise homeport receives over 100 cruise ships annually. Peak charging demand is high during peak tourist berthing periods, and the short berthing window places stringent requirements on charging efficiency and connection time. Simultaneously, the port's anchorage sees over 50 ships waiting to berth daily, which cannot be covered by fixed shore power. The continuous operation of diesel auxiliary engines on waiting ships causes severe carbon emissions and noise pollution, necessitating a mobile green electricity supply solution covering the entire anchorage-terminal scenario. With the global intelligent transformation of ports, shore power technology has evolved from traditional manual operation to "automated connection + intelligent monitoring." Automated terminals such as Shanghai Yangshan Port and Shenzhen Mawan Port have achieved seamless communication between shore power systems and ship IPMS systems, reducing connection time to less than 8 minutes and increasing the ship connection success rate to over 99%. However, existing technologies still have many limitations for dynamic collaborative scenarios between mobile green electricity supply vessels and receiving vessels, especially in typical scenarios such as peak loads at cruise homeports, low-load power replenishment at anchorages, extreme conditions during typhoons, and collaborative operations during busy port periods. The core pain points are concentrated in five main areas:

[0167] Firstly, the real-time performance and compatibility of ship-to-shore communication protocols are insufficient, and the barriers to collaborative control information exchange are prominent. Existing ship-to-shore communication protocols (such as Modbus and DNP3.0) are prone to transmission delays and data packet loss in the complex environment of ports with high humidity and strong electromagnetic interference. They cannot cover stable communication across the entire voyage from the port anchorage to the cruise home port. During peak hours, communication congestion can lead to delays in collaborative control commands, failing to meet the millisecond-level response requirements of power supply and receiving collaborative control. At the same time, the communication interfaces and data protocol standards of ship power management systems of different ship types are not uniform, and there is a lack of a unified power supply and receiving collaborative control protocol that is compatible with multiple ship types. This results in poor synchronization of commands between the power supply and receiving parties, low parameter matching efficiency, and an inability to achieve seamless collaboration.

[0168] Secondly, the ability to fuse multi-source heterogeneous sensor data is insufficient, resulting in poor accuracy and robustness of intelligent sensing under complex operating conditions. Existing green electricity supply and receiving monitoring systems mostly use single electrical parameter monitoring and have not achieved deep fusion of multi-modal data such as interface temperature, current fluctuations, contact pressure, ambient humidity, and ship attitude. Existing AI sensing models are insufficient in extracting subtle fault features such as large fluctuations in ship attitude, decreased insulation in high humidity environments, early signs of poor contact, and hidden leakage under complex operating conditions such as typhoons, high tides, and busy waterways in Xiamen Port. This leads to a significant decrease in fault identification accuracy and makes it impossible to provide early warning of potential hazards under all operating conditions.

[0169] Third, the dynamic coordination control precision between the green electricity supply vessel and the receiving vessel is insufficient, resulting in poor stability in docking and power supply adaptation. The navigation path planning of the green electricity supply vessel from the anchorage to the cruise home port, the attitude alignment with the receiving vessel, and the interface docking process are greatly affected by multiple sources of disturbance such as wind, waves, tides, currents, and channel congestion. Traditional algorithms such as PID and sliding mode control are difficult to achieve millimeter-level attitude precision alignment and cannot adapt to stable docking under the disturbances of wind and waves during typhoons and high tide changes in Xiamen Port. Problems such as interface docking deviation, mechanical wear, and docking failure are prone to occur. At the same time, in response to the efficiency requirements of multiple vessels being powered simultaneously and charging time ≤1 hour in the cruise home port, the existing power regulation algorithm has a lag in response and is prone to voltage fluctuations and frequency deviations, making it impossible to achieve dynamic adaptation and stable output in the power supply process.

[0170] Fourth, the real-time performance, fault tolerance, and interpretability of the safety control system are insufficient, making it difficult to meet the high reliability requirements of ship electrical safety. Existing power supply and receiving emergency protection mechanisms are mostly post-event triggered overcurrent and overtemperature protections, with response times generally exceeding 100ms, which cannot cope with sudden severe faults such as short circuits and arcing. Existing fault diagnosis models are mostly black-box structures, only able to output fault warnings, unable to explain the reasoning logic of fault root causes, and prone to misjudgment and missed judgment. At the same time, there is a lack of risk simulation mechanisms for extreme operating conditions, as well as a graded safety control system of "early warning-degradation-emergency disconnection", resulting in insufficient fault tolerance and emergency response capabilities.

[0171] Fifth, there is insufficient closed-loop management and system integration throughout the entire process, lacking a closed-loop mechanism for scenario-based verification and continuous iterative optimization. Existing green electricity supply and receiving systems are mostly discrete architectures, with barriers to data interaction between sensing, communication, control, protection, and operation and maintenance modules, failing to form a closed-loop management system covering the entire process of navigation, berthing, docking, power supply, and operation and maintenance. Furthermore, there is a lack of a full-process verification system for typical port scenarios, and no iterative mechanism of "world model pre-simulation + field testing + AI optimization" has been established. This makes it impossible to continuously optimize model performance based on scenario-based test data, resulting in insufficient cross-scenario adaptability and long-term operational stability, making it difficult to form replicable, large-scale application standards.

[0172] The development of technologies such as multimodal large-scale models, edge computing, digital twin world models, and deep reinforcement learning has provided technical pathways to solve the aforementioned problems. However, existing technologies have not yet achieved deep integration with the power supply and receiving scenarios of green electricity supply vessels, and have not formed a complete technical system covering collaborative communication, intelligent sensing, docking control, safety control, scenario verification, and closed-loop optimization. Therefore, it is urgent to construct a dynamic collaboration and safety control method for green electricity supply vessels and power receiving vessels, comprehensively overcome the core pain points of existing technologies, adapt to the multi-scenario operation needs of ports, and achieve highly safe, highly reliable, and highly efficient collaborative operation of the entire green electricity supply and receiving process.

[0173] To address the core pain points of existing technologies and to specifically tackle the key technical, theoretical, and scientific issues related to the dynamic coordination and safety control of green energy supply vessels and power receiving vessels, this application provides a method for dynamic coordination and safety control of green energy supply vessels and power receiving vessels. The method uses multimodal sensing data from the entire power supply and receiving process of green energy supply vessels and power receiving vessels, physical models of the vessel's power system, port environment and berthing condition data, power supply and receiving safety standards, and expert operation and maintenance experience as core processing objects to construct a complete process system of "unified communication protocol foundation - multimodal intelligent sensing - hierarchical safety control - dynamic collaborative control". By establishing a communication foundation through low-latency ship-shore communication and standardized collaborative control protocols, a two-tiered safety control architecture of "large-scale global cognitive reasoning + small-scale real-time closed-loop management" is designed. A multi-modal fusion AI multi-dimensional perception model and hidden danger identification mechanism are constructed, and dynamic collaborative matching and adaptive docking control algorithms for ships under wind and wave disturbances are developed. The four core challenges of communication collaboration, intelligent perception, docking control, and safety control are comprehensively overcome. The system is adapted to the operational needs of typical scenarios such as the Xiamen Port cruise home port and anchorage, and overcomes core challenges such as anti-interference under extreme working conditions, multi-ship collaborative power supply, and efficient fast charging. The system achieves high-safety, high-reliability, and high-efficiency collaborative operation of the green electricity supply ship and the power receiving ship throughout the entire power supply and receiving process.

[0174] The method for dynamic coordination and safety control between green electricity supply vessels and power receiving vessels proposed in this application belongs to the cutting-edge technology field of deep intersection of marine electrical engineering, intelligent shipping control, artificial intelligence and multi-sensor fusion, ship-shore communication and port operation management. This method deeply integrates ship-shore low-latency communication and standardized protocol adaptation, multi-modal sensor data fusion, AI real-time monitoring and large model cognitive reasoning, adaptive collaborative control, and full-process closed-loop safety management technology. It takes multi-source heterogeneous data of the entire process of power supply and receiving between green electricity supply vessels and power receiving vessels, physical models of ship power systems, port hydrological and meteorological data and berthing condition data, ship power supply and receiving safety specifications and expert operation and maintenance experience as the core processing objects, and constructs a green electricity supply and receiving collaborative control, intelligent perception, safety control, and full-process closed-loop management system adapted to multiple scenarios such as Xiamen Port cruise home port, port anchorage, and inland river terminal.

[0175] The proposed method for dynamic coordination and safety control between green electricity supply vessels and receiving vessels is applicable to various port scenarios (such as Xiamen Port), including centralized power supply for cruise ships under high load at cruise homeports, low-load power replenishment for vessels waiting at anchor in port anchorages, mobile green electricity supply for container ships and bulk carriers during berthing, ship-shore coordinated power supply and receiving management, intelligent power supply safety control, and zero-emission port operations. It can support high-quality development of green shipping, intelligent upgrading of port shore power systems, and safe, efficient, and zero-emission operation assurance for ship berthing. The training data used in this method covers ship power supply and receiving design specifications, port shore power operation standards, the International Maritime Organization (IMO) Code for Electrical Safety of Ships, port hydrological, meteorological, and waterway geographic data, historical power supply and receiving condition data, electrical fault cases, hazard handling cases, and expert operation and maintenance experience data. Through the structured fusion of multimodal data and knowledge in the shipping electrical field, it provides comprehensive and accurate knowledge and data support for large-scale model training and intelligent control strategy optimization.

[0176] It should be noted that the dynamic coordination and safety control method for green electricity supply vessels and power receiving vessels in this application can be applied to the dynamic coordination and safety control of various ports. The example in this embodiment is Xiamen Port.

[0177] Figure 2 Here is an overall flowchart of one embodiment, referred to below. Figure 2 The steps of this application are systematically described.

[0178] Step S1: Construction of power supply and receiving collaborative control protocol and communication architecture based on ship-shore low-latency communication.

[0179] This step can be implemented sequentially as S11, S12, and S13. The core objective is to complete the construction of a low-latency redundant communication system for the entire shipping route of each port (such as Xiamen Port) and develop a unified power supply and receiving collaborative control protocol, so as to provide a stable communication foundation for subsequent collaborative control and security prevention.

[0180] S11 Ship-Shore Collaborative Communication System Selection and Redundancy Deployment. The implementation process is as follows:

[0181] 1. Complete the selection, joint calibration and deployment of a multi-link redundant communication system for green electricity supply ships, powered cruise ships and shore-based control centers in various port scenarios (such as Xiamen Port). The main link adopts 5G+TSN fiber optic communication, and is equipped with 5G base station supplementation and optimization along the channel from the anchorage of each port (such as Xiamen Port) to the cruise home port. The backup link adopts a maritime satellite communication enhancement scheme to achieve full-area communication coverage without dead zones in the anchorage, channel and cruise home port operation area of ​​each port (such as Xiamen Port).

[0182] 2. Deploy the IEEE 1588 PTP high-precision time synchronization protocol to complete the time synchronization calibration of ship-shore equipment, achieve a data sampling time deviation of ≤0.8ms between ship and shore, and ensure the time synchronization of collaborative control commands; carry out electromagnetic shielding and moisture-proof and salt spray-resistant hardware modifications on all communication equipment to adapt to the port operation environment of high humidity, high salt spray and strong electromagnetic interference in various ports (such as Xiamen Port).

[0183] 3. Develop unified standards for multi-disciplinary data in green electricity supply and reception, clarify the storage format, time granularity, field naming conventions, and data verification rules for time-series electrical and sensor data, unstructured text data, power system model data, and hydrological and meteorological data, and achieve standardized definitions for multi-source heterogeneous data.

[0184] S12 Unified Power Supply and Receiving Coordination Control Protocol Development. The implementation process is as follows:

[0185] 1. Based on the ISO 18131 international standard for ship shore power communication, develop a unified power supply and receiving coordination control protocol with a publish-subscribe architecture, clearly define five core interaction logics: pre-docking parameter matching, navigation coordination scheduling, docking action coordination, real-time power supply control, and emergency protection interaction, and complete the development and testing of the protocol stack.

[0186] 2. Based on the protocol adaptive adaptation model described in the invention, the development and deployment of the protocol conversion gateway are completed, enabling adaptive conversion of communication protocols and data formats of IPMS systems for different ship types such as mainstream cruise ship brands and container ships docked at various ports (such as Xiamen Port), achieving plug-and-play compatibility.

[0187] 3. The compatibility and reliability tests of the protocol were completed. The test results showed that the protocol adaptation success rate was 100%, the command interaction synchronization error was ≤2ms, and it met the collaborative control requirements of various ports (such as Xiamen Port).

[0188] S13 Ship-to-shore communication low-latency and high-reliability optimization. The implementation process is as follows:

[0189] 1. Based on the communication delay optimization model described in the invention, the parameters of the communication link are optimized and the performance is tuned. By adopting edge computing node local data processing, protocol frame lightweight compression, and redundant link seamless switching technology, a dynamic scheduling mechanism for communication resources during busy periods in ports is designed to ensure priority transmission of core instructions and further reduce transmission delay.

[0190] 2. The performance test of the communication link across the entire flight segment was completed. The test results showed that the total end-to-end delay of the collaborative control command was ≤8ms, the transmission delay of the emergency protection command was ≤3ms, and the packet loss rate was ≤0.0005%, which fully met the performance targets set in the invention and was suitable for the communication needs of the entire operation process from the anchorage to the cruise home port of various ports (such as Xiamen Port).

[0191] Step 2: Construction of a multimodal fusion AI multidimensional perception and real-time monitoring system for power supply and receiving hazards.

[0192] This step can be implemented sequentially in the order of S21, S22, and S23. The core objective is to complete the deployment of the multimodal sensing system and the construction of the AI ​​hazard identification model, so as to realize the real-time monitoring and accurate identification of hazards in the entire power supply and receiving process under all operating conditions in each port.

[0193] S21 Power Supply and Reception Full-Process Multimodal Sensing System Deployment and Dataset Construction. The implementation process is as follows:

[0194] 1. Complete the selection, calibration, and deployment of a multimodal sensing system for the entire power supply and receiving process of the green electricity supply vessel, including: electrical parameter monitoring modules (voltage, current, power, frequency, and phase difference sensors, sampling frequency 100Hz), interface status monitoring modules (contact pressure sensors, infrared temperature sensors, stroke sensors, and wear monitoring sensors), environmental status monitoring modules (humidity, salt spray, wind speed and direction, and visibility sensors), and ship status monitoring modules (BeiDou differential positioning system, IMU inertial measurement unit, and ship electrical load monitoring module), to achieve full coverage data collection for all operating conditions during navigation, berthing, docking, and power supply.

[0195] 2. Collect the full life cycle operating data of cruise ship shore power supply and reception at each port from 2022 to 2024, as well as the hydrological, meteorological and waterway geographical data of each port and the historical data of port traffic flow. This includes multi-source sensor data, cruise ship power system operation data, safety specification text data, historical fault and hidden danger handling cases, and expert operation and maintenance experience data to construct the original dataset.

[0196] 3. Perform full-process processing including data cleaning, spatiotemporal alignment of multi-source data, and enhancement of small sample extreme working condition data. Through the data fusion model described in the invention, optimize the dataset and fuse features to finally construct a standardized green electricity supply and receiving multimodal training dataset, which is divided into training set, validation set and test set in a ratio of 8:1:1 for subsequent model training and performance verification.

[0197] S22 Multimodal Feature Fusion and AI Hazard Identification Model Construction. The implementation process is as follows:

[0198] 1. Based on the multimodal joint embedding model described in the invention, feature extraction and fusion of different modal data are completed. Through adaptive weight adjustment, deep fusion of multimodal features and structured embedding of domain knowledge graph is achieved. Feature extraction weights are optimized for extreme scenarios such as typhoon days and high humidity, generating feature vectors of a unified dimension, and realizing unified representation of cross-domain heterogeneous data.

[0199] 2. Based on unified embedding features, an improved VIT AI multidimensional perception hazard identification model with an attention mechanism is constructed. The model is trained, validated, and tested based on the training dataset. Transfer learning is used to improve the model's identification performance under extreme working conditions with small samples.

[0200] 3. The model performance test was completed. The test results showed that the model's accuracy in identifying potential hazards such as short circuits, leakage, poor contact, and interface overheating was ≥99.7%, and the lead time for identifying early signs of poor contact was ≥35s, which met the performance targets set in the invention.

[0201] S23 Hazard Classification, Early Warning, and Emergency Trigger Mechanism Design. The implementation process is as follows:

[0202] 1. Based on the hazard identification results, a three-level early warning and triggering mechanism of "Level 1 warning - Level 2 downgrade - Level 3 emergency disconnection" is constructed, clarifying the judgment threshold, triggering conditions and execution actions of each level of early warning, and adapting to the safety management and control needs of different working conditions in various ports.

[0203] 2. Complete the simulation test and actual ship joint commissioning of the graded early warning mechanism, realize the audible and visual prompts and operation and maintenance push of the first-level early warning, the power derating and load limiting of the second-level degradation, and the millisecond-level circuit disconnection of the third-level emergency disconnection, so as to ensure operational safety.

[0204] Step 3: Construction of a full-process security control system of "AI real-time monitoring + large model inference".

[0205] This step can be implemented sequentially in the order of S31, S32, and S33. The core objective is to build a large-scale model for power supply and receiving, a full-scenario world model for each port, and a two-level collaborative safety control architecture to achieve fault root cause reasoning and interpretable operation and maintenance decision generation.

[0206] S31 Large-scale model construction and RAG retrieval enhancement mechanism for power supply and receiving domains. The implementation process is as follows:

[0207] 1. Based on the constructed multimodal dataset of green electricity supply and reception and the domain knowledge graph, a large-scale model of the shipping vertical domain is used for pre-training and fine-tuning to build a large-scale model dedicated to green electricity supply and reception. The training and verification of three core capabilities are completed: fault root cause reasoning, operation and maintenance decision optimization, and global risk assessment. Based on the geography, hydrology and meteorology, port operation rules, and historical ship operation data of each port, a full-scenario world model of green electricity supply and reception for each port is constructed. The model parameter calibration is completed, and the model simulation and the matching degree of actual ship working conditions are ≥98%.

[0208] 2. Introduce the RAG retrieval enhancement mechanism to build a dedicated knowledge base for the green electricity supply and reception field, covering ship electrical safety specifications, port shore power operation standards, scenario-based operation plans for various ports, fault handling cases, expert operation and maintenance experience, etc., to realize real-time knowledge retrieval and matching in the large model reasoning process, suppress the illusion of large models, and improve the accuracy and compliance of reasoning results.

[0209] 3. Completed large model performance testing. The test results showed that the accuracy of fault root cause reasoning was ≥99%, the compliance of reasoning results was 100%, and there was no hallucination output.

[0210] S32 two-tier collaborative security architecture design and deployment. The implementation process is as follows:

[0211] 1. Based on the two-level collaborative security control architecture described in the invention, the development, deployment and joint debugging of the global cognitive reasoning layer and the real-time control layer are completed, and the core functions, input and output boundaries and collaborative logic of the two-layer architecture are clarified.

[0212] 2. The global cognitive reasoning layer deploys a dual-drive architecture of "large model of power supply and receiving field + world model of port full scenario", and completes the deployment and verification of four core functions: operation process pre-drilling, navigation status prediction, risk simulation pre-verification, and global safety constraint output; the real-time control layer deploys a lightweight AI small model after distillation, and realizes millisecond-level real-time hidden danger monitoring, graded protection triggering, and emergency disconnection control.

[0213] 3. Design a layered triggering and updating mechanism. The global cognitive reasoning layer is updated once every 20 seconds. Emergency updates are triggered when there are sudden changes in operating conditions, escalation of hidden dangers, or sudden changes in the environment. The real-time control layer has a control frequency of 100Hz to achieve continuous closed-loop control.

[0214] 4. Completed architecture performance testing. The test results show that the emergency protection response time of the real-time control layer is ≤40ms, which meets the performance target set in the invention.

[0215] S33 Fault Root Cause Reasoning and Interpretable Operation and Maintenance Decision Generation. The implementation process is as follows:

[0216] 1. Introduce Explainable AI (XAI) technology and adopt the SHAP value method to generate an interpretability report with fault reasoning logic, risk basis, and handling priority, thereby solving the interpretability problem of black box models.

[0217] 2. Construct an AI-powered operation and maintenance decision-making model. Based on the root cause reasoning results of faults, combined with domain knowledge and historical port handling cases, push precise operation and maintenance handling suggestions and work instructions to achieve a closed loop of the entire process of "hazard identification - root cause reasoning - handling suggestions".

[0218] 3. Completed real-ship scenario verification, with 100% feasibility of operation and maintenance suggestions and an improvement of more than 60% in fault handling efficiency.

[0219] Step 4: Dynamic collaborative matching and adaptive control between green electric ships and powered ships under wind and wave disturbances.

[0220] This step can be implemented sequentially in the order of S41, S42, and S43. The core objective is to build a dynamic collaborative matching algorithm for ships to achieve autonomous navigation between port anchorage and home port, precise docking under wind and wave disturbances, and dynamic adaptation of the power supply process.

[0221] S41 Markov decision process modeling of ship relative motion and docking process. The implementation process is as follows:

[0222] 1. Model the entire process of green electricity supply ship's navigation from port anchorage to home port, attitude alignment with the powered cruise ship, interface docking, and dynamic power supply adaptation as a Markov decision process (MDP). Complete the state space definition covering the ship's navigation status, relative position, attitude, electrical parameters, operation mode, and dynamics of surrounding ships, as well as the action space definition covering the propulsion system, docking mechanism, power system control quantities, and multi-ship power distribution coefficients.

[0223] 2. Based on Pareto optimality theory, a multi-objective reward function is designed to balance six core objectives: navigation efficiency, docking accuracy, attitude stability, power supply stability, operational efficiency, and operational safety. The adaptive adjustment rules of weight coefficients under different operational modes are clarified, providing a foundation for the optimization of reinforcement learning control strategies.

[0224] The S42 AI dynamic collaborative matching algorithm was constructed and validated. The implementation process is as follows:

[0225] 1. The MADDPG multi-agent deep reinforcement learning algorithm is adopted to construct an AI dynamic collaborative matching algorithm. The model is pre-trained and optimized based on the virtual test field of the port world model. It covers the full-scene working conditions of 0-6 level wind and waves, different tidal current speeds, ships of different tonnages, and busy and congested waterways in various ports, thereby improving the model's anti-disturbance ability and generalization ability.

[0226] 2. A model predictive control (MPC) rolling optimization mechanism is introduced. Based on the mathematical model of the ship's relative motion, the control commands are optimized in the rolling time domain. The wind-resistant stability control strategy is optimized for berthing scenarios in typhoon days, so as to achieve precise control of the ship's attitude.

[0227] 3. The port ship docking test was completed. The test results showed that under normal wind and wave conditions of level 3, the attitude alignment accuracy of the green power supply ship and the power receiving cruise ship was ≤3cm in the lateral deviation, ≤8cm in the longitudinal deviation, and ≤0.3° in the relative heading deviation. The docking time was ≤8 minutes and the docking success rate was 100%. Stable docking can still be completed under level 6 wind and wave conditions. The wind resistance meets the requirements for berthing operations during typhoon days.

[0228] S43 power supply process dynamic adaptive control. The implementation process is as follows:

[0229] 1. Construct an adaptive fuzzy PID power regulation control algorithm, complete the algorithm development, parameter tuning and actual ship deployment, and realize dynamic adaptive adjustment of power output; construct a dynamic power allocation strategy for simultaneous power supply to multiple ships, supporting stable power supply to 1-2 cruise ships at the same time.

[0230] 2. Construct a grid-connected synchronous control strategy to achieve seamless grid connection between the green electricity supply ship's output power and the IPMS system of the receiving cruise ship, controlling the grid connection phase difference ≤2°, frequency difference ≤0.15Hz, and voltage difference ≤3% to avoid grid connection impact.

[0231] 3. Completed the actual ship power supply test. The test results show that the power regulation response time is ≤3s, the voltage fluctuation is controlled within ±1.5%, and the frequency deviation is controlled within ±0.1Hz. When a single cruise ship is fully powered, the charging time is ≤50 minutes, which meets the operation efficiency requirement of ≤1 hour. When two cruise ships are powered at the same time, the load balance is ≥98%, and the power supply stability meets the design requirements.

[0232] In this embodiment, through the sequential implementation of the above steps, the dynamic coordination and safety control of the green electricity supply vessel and the power receiving vessel are fully realized. The adaptability verification of typical scenarios such as cruise home ports and anchorages in various ports has been completed. Core problems such as typhoon-resistant berthing, high-efficiency fast charging, and multi-vessel collaborative power supply have been overcome. The operation efficiency, safety, and stability all meet the design requirements, solve the core pain points of existing technologies, and can fully support the zero-emission operation of ports and the transformation of green shipping.

[0233] It should be noted that any technical feature in any of the above embodiments provided in this application is also applicable to any of the following embodiments provided in this application. For example, the technical features in the relevant embodiments of the above methods are also applicable to the relevant embodiments of the following devices, systems or equipment, and similar details will not be repeated hereafter.

[0234] Figure 3 Here is a structural block diagram of the device in one embodiment, with reference to Figure 3 The dynamic coordination and safety control device between the green electricity supply vessel and the power receiving vessel includes:

[0235] The communication architecture deployment module 301 is used to develop a power supply and receiving coordination control protocol for ship-shore low-latency communication based on a protocol adaptive adaptation model, so as to deploy the communication architecture.

[0236] The multi-dimensional perception and early warning module 302 is used to acquire multi-modal sensing data during the charging docking operation between the supply vessel and the power receiving vessel according to the deployed communication architecture, and to construct a multi-dimensional perception hazard identification model based on the multi-modal sensing data in order to output multi-dimensional perception hazard early warning information related to the charging docking operation.

[0237] The two-level collaborative early warning module 303 is used to construct a large model of the power supply and receiving field based on multimodal sensor data and domain knowledge graph, using a two-level collaborative safety control architecture, so as to output global cognitive reasoning layer hidden danger early warning information and real-time control layer hidden danger early warning information related to charging docking operation.

[0238] The charging docking control module 304 is used to model the charging docking operation to obtain a Markov decision model, and solve the Markov decision model based on a dynamic cooperative matching algorithm of reinforcement learning to output navigation attitude control information; the navigation attitude control information is used to control the execution of the charging docking operation.

[0239] The alarm stop module 305 is used to stop the execution of charging docking operations when the warning level represented by the multi-dimensional perception hazard warning information, the global cognitive reasoning layer hazard warning information, and the real-time control layer hazard warning information meets the target level.

[0240] In this embodiment of the application, based on, as follows Figure 3 The connection relationships between the various modules or units shown in the diagram improve the efficiency and safety of the charging docking operation process through the cooperation between these modules or units.

[0241] In another embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As shown, it includes a processor, memory, input / output interfaces, and a communication interface. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface is connected to the system bus via the input / output interfaces. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores relevant data. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. The computer program can be executed by the processor to implement the various methods described in the above embodiments.

[0242] In yet another embodiment, a computer device is provided, such as a terminal, whose internal structure diagram may be as follows: Figure 5 As shown, it includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. The computer program can be executed by the processor to implement the various methods described in the above embodiments.

[0243] Those skilled in the art will understand that Figure 4 and Figure 5 The structure shown is only a block diagram of a part of the structure related to the present application and does not constitute a limitation on the computer device on which the present application is applied. It may also include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements, in order to realize the function of the computer device.

[0244] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0245] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the systems, devices, equipment, modules or units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0246] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, devices, or methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0247] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0248] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0249] In the above embodiments, the implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, in the form of a computer program product.

[0250] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium, or a semiconductor medium (e.g., a solid-state drive), etc.

[0251] The technical solutions provided by the embodiments of this application have been described in detail above. Specific examples have been used in the embodiments of this application to illustrate the principles and implementation methods of the embodiments of this application. The description of the above embodiments is only for the purpose of helping to understand the methods and core ideas of the embodiments of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the embodiments of this application. Therefore, the content of this specification should not be construed as a limitation on the embodiments of this application.

Claims

1. A method for dynamic coordination and safety control between green electricity supply vessels and power receiving vessels, characterized in that, The method includes: Based on the protocol adaptive adaptation model, a power supply and receiving coordination control protocol for ship-shore low-latency communication is developed for the deployment of the communication architecture. Based on the deployed communication architecture, multimodal sensor data is acquired during the charging docking operation between the supply vessel and the receiving vessel. A multidimensional perception hazard identification model is constructed based on this multimodal sensor data to output multidimensional perception hazard early warning information associated with the charging docking operation. Specifically, this includes: collecting the multimodal sensor data during the charging docking operation; constructing a multimodal training dataset based on the multimodal sensor data and green electricity supply and receiving lifecycle operating condition data; optimizing the multimodal training dataset using a data cleaning model, a feature alignment model, and a few-shot augmentation model to obtain an optimized dataset; constructing a multimodal joint embedding model; transforming the optimized dataset to obtain a unified-dimensional feature vector; and constructing an improved multidimensional perception hazard identification model incorporating an attention mechanism based on the unified-dimensional feature vector. Based on the multimodal sensing data and domain knowledge graph, a two-level collaborative security control architecture is adopted to construct a large model of the power supply and receiving domain, so as to output the global cognitive reasoning layer hidden danger warning information and the real-time control layer hidden danger warning information associated with the charging docking operation; A Markov decision model is obtained by modeling the charging docking operation. The Markov decision model is solved by a dynamic cooperative matching algorithm based on reinforcement learning to output navigation attitude control information. The navigation attitude control information is used to control the execution of the charging docking operation. If the warning level represented by the multi-dimensional perception hazard warning information, the global cognitive reasoning layer hazard warning information, and the real-time control layer hazard warning information meets the target level, the charging docking operation shall be stopped.

2. The method according to claim 1, characterized in that, The aforementioned protocol-based adaptive adaptation model is used to develop a power supply and receiving coordination control protocol for ship-to-shore low-latency communication, for the deployment of the communication architecture, including: A protocol adaptive adaptation model is constructed based on protocol adaptation rules; Based on the sum of the end-to-end total delay of the instruction and the delay of each process, a communication delay optimization model is constructed. Based on the delay summation relationship represented in the delay optimization model as characterized in the protocol adaptive adaptation model, the power supply and receiving coordination control protocol for ship-shore low-latency communication is determined.

3. The method according to claim 1, characterized in that, Based on the multimodal sensing data and domain knowledge graph, a two-level collaborative security control architecture is adopted to construct a large model of the power supply and receiving domain, so as to output the global cognitive reasoning layer hidden danger early warning information and the real-time control layer hidden danger early warning information associated with the charging docking operation, including: Based on multimodal fusion features, prediction duration and world model, a global cognitive reasoning layer sub-model is constructed to output the hidden danger warning information and global security constraints of the global cognitive reasoning layer; A lightweight model after distillation is used to construct a real-time control layer sub-model based on real-time collected high-frequency features and the global security constraints, so as to output the real-time control layer hidden danger early warning information.

4. The method according to claim 1, characterized in that, The method further includes: Based on the hidden danger warning information of the global cognitive reasoning layer and the hidden danger warning information of the real-time control layer, cognitive reasoning ability is performed based on the large model to generate reasoning information of the root cause of the fault. Based on the reasoning information of the root cause of the fault, domain knowledge, and historical handling cases, operation and maintenance handling suggestions are generated based on the operation and maintenance decision model.

5. The method according to claim 1, characterized in that, The charging docking operation is modeled to obtain a Markov decision model. A dynamic cooperative matching algorithm based on reinforcement learning is used to solve the Markov decision model to output navigation attitude control information, including: The charging docking operation is model defined to obtain the state space, action space, and multi-objective reward function; The Markov decision model is determined based on the state space, the action space, and the multi-objective reward function. A dynamic collaborative matching algorithm for reinforcement learning is constructed using a multi-agent deep reinforcement learning algorithm to solve the Markov decision model and output navigation attitude control information.

6. The method according to claim 1, characterized in that, The method further includes: An adaptive power regulation control algorithm, a power dynamic allocation strategy for multiple ships supplying power simultaneously, and a grid-connected synchronization control strategy were constructed. The execution of the charging docking operation is controlled by the adaptive power adjustment control algorithm, the power dynamic allocation strategy for simultaneous power supply from multiple ships, the grid-connected synchronization control strategy, and the navigation attitude control information.

7. A dynamic coordination and safety control device for green electricity supply vessels and power receiving vessels, characterized in that, The device includes: The communication architecture deployment module is used to develop a power supply and receiving coordination control protocol for ship-shore low-latency communication based on a protocol adaptive adaptation model, so as to deploy the communication architecture. A multi-dimensional perception and early warning module is used to acquire multi-modal sensor data during the charging docking operation between the supply vessel and the power receiving vessel, based on a pre-deployed communication architecture. It then constructs a multi-dimensional perception hazard identification model based on this multi-modal sensor data to output multi-dimensional perception hazard early warning information associated with the charging docking operation. Specifically, it is used to: collect the multi-modal sensor data during the charging docking operation; construct a multi-modal training dataset based on the multi-modal sensor data and green electricity supply and receiving lifecycle operating condition data; optimize the multi-modal training dataset by constructing a data cleaning model, a feature alignment model, and a few-shot augmentation model to obtain an optimized dataset; construct a multi-modal joint embedding model; transform the data based on the optimized dataset to obtain a unified-dimensional feature vector; and construct an improved multi-dimensional perception hazard identification model incorporating an attention mechanism based on the unified-dimensional feature vector. The two-level collaborative early warning module is used to construct a large model of the power supply and receiving field based on the multimodal sensing data and the domain knowledge graph, using a two-level collaborative security control architecture, so as to output the global cognitive reasoning layer hidden danger early warning information and the real-time control layer hidden danger early warning information associated with the charging docking operation. The charging docking control module is used to model the charging docking operation to obtain a Markov decision model, and solve the Markov decision model based on a dynamic cooperative matching algorithm of reinforcement learning to output navigation attitude control information; the navigation attitude control information is used to control the execution of the charging docking operation. The alarm stop module is used to stop the execution of the charging docking operation when the warning level represented by the multi-dimensional perception hazard warning information, the global cognitive reasoning layer hazard warning information, and the real-time control layer hazard warning information meets the target level.

8. A computer device, characterized in that, The computer device includes: At least one processor and memory; The memory is used to store program code, and the processor is used to call the program code stored in the memory to execute the method as described in any one of claims 1 to 6.

9. A computer storage medium, characterized in that, It includes instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1 to 6.

Images