An offshore new energy station ship contact accident emergency resource scheduling method

Abstract

This application relates to the technical field of emergency management and discloses a method for emergency resource scheduling in the event of a collision between a vessel and a marine new energy power station. The method includes: multi-source data acquisition and standardized processing to construct a three-dimensional data system of accident-resource-environment and complete data fusion; intelligent identification and severity assessment of accident scenarios, identifying accident scenarios and outputting accident severity and emergency demand priorities; dynamic pooling and precise matching of emergency resources, constructing a dynamic resource pool and solving for the optimal resource combination through a multi-objective optimization model; dynamic adaptive path planning for the environment, generating and dynamically correcting navigation paths and planning resource arrival times; and scheduling execution and closed-loop dynamic control, synchronizing scheduling instructions and dynamically adjusting the plan for abnormal situations. This invention achieves closed-loop management of the entire accident emergency response process, significantly improving emergency response efficiency and reliability.

Description

Technical Field

[0001] This application relates to the technical field of emergency management, and in particular to an emergency resource dispatching method for ship collision accidents at offshore new energy power stations. Background Technology

[0002] Offshore wind power and offshore photovoltaic power plants have become core carriers for clean energy development, and the installed capacity of offshore new energy continues to grow rapidly. At the same time, nearshore waters are busy with shipping routes and have complex navigation environments, resulting in a significant risk of ship collisions caused by deviations from navigation channels or operational errors. Accidents involving ships colliding with offshore new energy power plants not only cause damage to the plant structures and interruptions in energy supply, leading to huge economic losses, but also easily result in casualties, marine environmental pollution from fuel spills, and even secondary disasters, making emergency response extremely difficult.

[0003] Existing emergency resource dispatch technologies for maritime accidents, when applied to collision incidents involving ships at offshore new energy power stations, largely rely on single data sources such as Automatic Identification Systems (AIS) and basic meteorological data. Furthermore, their ability to fuse and process heterogeneous multi-source data is insufficient, failing to provide comprehensive and accurate data support for emergency decision-making. Secondly, at the accident assessment level, existing technologies largely depend on manual assessment of accident severity and demand analysis. They can only achieve basic identification of single accident scenarios and cannot handle complex scenarios involving ship collisions coupled with extreme weather, oil spills, and personnel distress. They also cannot predict the evolution of accidents, easily leading to inaccurate judgments of emergency needs and unreasonable prioritization of responses. At the resource scheduling level, existing solutions are mostly limited to the local operation and maintenance resources of offshore new energy power stations, and cannot achieve cross-unit, cross-industry, and cross-regional emergency resource coordination. At the navigation path planning level, most solutions generate static paths based on the shortest voyage, which can only passively avoid severe sea conditions and cannot adapt to the dynamic changes of extreme sea conditions such as typhoons and strong storm surges. The robustness and safety of the path planning are insufficient, and the scheduling plan is prone to being out of touch with the actual situation on site. Summary of the Invention

[0004] To address the aforementioned technical problems, this application provides an emergency resource dispatching method for ship collision accidents at offshore new energy power stations, employing the following technical solution: A method for emergency resource dispatching in the event of a collision between a vessel and a marine renewable energy power station includes the following steps: S1. Synchronously collect accident data of involved vessels and stations, emergency resource data of the entire region, and marine meteorological, hydrological and navigation environment data; perform noise filtering, spatiotemporal synchronization and heterogeneous semantic fusion on the collected multi-source heterogeneous data, establish data association index, and generate standardized association dataset; S2. Based on historical and real-time navigation data of ships and marine environmental data, predict the navigation trajectory of ships within a set time period through a spatiotemporal prediction model, and calculate the probability of collision risk between ships and offshore new energy station structures; when the probability of collision risk exceeds the preset threshold, trigger risk warning and emergency resource pre-dispatch instructions, and dispatch nearby emergency resources to stand by in the vicinity of the warning area. S3. After a ship collision accident occurs, based on the standardized associated dataset, the accident scene is classified and multi-factor coupled accident scene is identified. The core feature parameters of the accident are extracted. Through the pre-trained accident judgment model, the accident level, emergency demand priority and future evolution trend of the accident are output simultaneously, and the emergency response target is dynamically updated. S4. Based on emergency response objectives, accident evolution trends, and real-time marine environmental data, with the optimization objectives of optimal response time, highest resource adaptability, lowest scheduling cost, and lowest navigation safety risk, a multi-objective optimization model coupling emergency resource matching and navigation path planning is constructed to solve for the optimal combination of emergency resources, the adaptive navigation path of each resource, and the arrival time sequence. S5. Synchronize dispatch instructions and navigation path information to each emergency resource execution terminal, and track the resource operation status and task execution progress in real time; trigger the dynamic update process of the dispatch plan in response to sudden abnormal situations during the execution process.

[0005] Optionally, the objective function of the multi-objective optimization model is: ; in, The solution is a combined approach of emergency resource combinations and corresponding navigation paths. To comprehensively optimize the objectives; , , , These are the weighting coefficients for response time, adaptability, scheduling cost, and security risk, respectively. The weighting coefficients are dynamically adjusted based on the accident level and emergency response objectives. The dynamic response time function is weighted by priority, taking into account the dynamic impact of real-time sea conditions on speed, the priority weight of emergency needs, the advantages of pre-scheduled resources, and the constraints of the timing of handling. The hierarchical dynamic adaptability function comprehensively considers the matching of core requirements at different levels, the collaborative adaptation of resource combinations, the capability redundancy of accident evolution requirements, and the compatibility of cross-domain resource operations. It is a comprehensive cost function for the entire life cycle, which comprehensively considers explicit direct costs, implicit loss costs, cross-domain collaboration costs, opportunity costs, and accident evolution risk costs. This is a spatiotemporally dynamic cumulative safety risk function that comprehensively considers the ship's own seaworthiness, dynamic sea state risk, navigation collision risk, and secondary risks in environmentally sensitive areas, representing the cumulative risk across the entire path.

[0006] Optionally, the priority-weighted dynamic response time function The expression is: Where n is the total number of emergency resources included in the emergency resource combination; The priority weight of the disposal task corresponding to the i-th resource is (0,1], with higher weights for core emergency needs, and the weight normalization constraint of all tasks under the same accident is satisfied. The total voyage distance for the planned navigation path corresponding to the i-th resource is determined in real time by the path planning stage; The dynamic equivalent speed of the i-th resource on the planned path is calculated based on the nominal speed, real-time wave height, wind speed, ocean current direction and speed, and tidal state of the resource, rather than the static nominal speed. Let be the pre-scheduled state coefficient of the i-th resource. If the resource is a pre-scheduled standby resource, Use a preset value of 0.3 to 0.6. If the resource is scheduled after an incident, Set to 0; Let be the timing delay for the i-th resource, calculated based on the pre- and post-task constraints of emergency response. If this resource is a core priority task resource... If the value is 0, and the process needs to be completed before entering the stage, the waiting time is calculated based on the estimated completion time of the preceding task.

[0007] Optionally, the hierarchical dynamic adaptation function The expression is: in, , , These are the weight coefficients for core requirement adaptation, combined collaborative adaptation, and evolutionary redundancy adaptation, respectively. The weights are dynamically adjusted based on the accident level and the risk of evolution. For the core requirement level adaptability, the expression is: in, This represents the total number of emergency needs. Let k be the priority weight of the emergency demand. For resource combination For the first The single-item matching degree of each requirement, with a value range of [0,1], is calculated based on the matching degree between resource function parameters, scenario adaptation tags and requirements; The expression for resource combination and collaborative adaptability is: in, For the first in the resource portfolio The and the first The collaboration adaptation coefficient of each resource is in the range of (0,1]. It is determined based on the operation timing compatibility, communication system compatibility, disposal space compatibility and the collaboration mechanism of the affiliated unit. When there is no collaboration obstacle, it is taken as 1. When there is collaboration constraint, it is adjusted down according to the degree of constraint. The redundancy fit for evolutionary requirements is expressed as: in, The redundancy coefficient of the i-th resource in response to the needs of accident evolution is denoted by [0,1]. Based on the future evolution trend output by the accident assessment model, the degree to which the performance parameter of the resource exceeds the current needs and covers the potential future needs is calculated. The value is 1 when the potential evolution needs are fully covered.

[0008] Optionally, the total lifecycle cost function The cost is determined by the sum of the explicit direct costs of resource scheduling, the implicit equipment wear and tear costs, the cross-domain collaboration costs, the opportunity costs of resource allocation, and the costs of accident evolution risks.

[0009] Optionally, the spatiotemporal dynamic cumulative security risk function The expression is: Where n is the total number of emergency resources included in the emergency resource combination; The total number of path points after discretization of the planned navigation path for the i-th resource; This represents the proportion of the segmented flight distance corresponding to the p-th path point on the i-th resource path to the total flight distance. The comprehensive dynamic risk value is the comprehensive dynamic risk value of the i-th resource at the p-th path point. The assessment values ​​are obtained based on environmental risk, navigation risk, and secondary environmental risk.

[0010] Optionally, in steps S1-3: The accident data includes ship AIS data, vibration sensor data of station structures, high-definition on-site image data, personnel positioning and vital signs data, satellite remote sensing data, and ship sonar detection data. The emergency resource data includes the real-time location, operating status, performance parameters, and scenario adaptation attributes of rescue vessels, rescue helicopters, medical emergency equipment, station repair equipment, and oil spill prevention equipment. The environmental data includes wave height, wind speed, visibility, tides, ocean currents, as well as data on the distribution of surrounding vessels, waterway control, and congestion. The spatiotemporal prediction model adopts a spatiotemporal graph convolutional network or a spatiotemporal Transformer model, and the set duration is 10-30 minutes; the emergency resource pre-scheduling instruction is generated based on the collision risk level, pre-handling requirements and surrounding resource status, and clarifies the standby location, standby requirements and emergency preparation matters of the pre-scheduled resources. The multi-factor coupled accident scenarios include scenarios where a ship touches a station structure and extreme weather, or where a ship collision is combined with an oil spill and people falling into the water. The accident assessment model adopts a deep learning model that integrates a time-series prediction module. The time-series prediction module is built based on LSTM or Transformer-XL networks and is used to combine the accident evolution law with dynamic environmental data to predict the future evolution trend of the accident scenario and update the priority of emergency needs and emergency response objectives simultaneously.

[0011] Optionally, step S4 includes: A dynamically updated global emergency resource pool is pre-built. This pool is functionally categorized into personnel rescue, medical emergency, station repair, oil spill prevention, and emergency support resources. The operational status, scenario adaptation tags, and location information of each resource are updated in real time. The pool also integrates cross-domain emergency resources, including social emergency resources, maritime regulatory resources, and fisheries emergency resources. Collaborative matching of cross-domain emergency resources employs a federated learning framework. Each cross-domain resource manager acts as a participating node, completing resource data processing and sub-model training locally. Only model parameters are uploaded to the central node for global model aggregation and updates. This allows for the joint training and solution of the cross-domain resource matching model without sharing sensitive local data. The multi-objective optimization model employs the Alternating Direction Multiplier Method (ADMM) for distributed solution. For navigation path planning and optimization, DQN or PPO reinforcement learning algorithms are used. The state space of the reinforcement learning algorithm includes real-time sea conditions, navigation environment, and ship navigation parameters, while the action space includes adjustments to heading, speed, and standby point selection. The reward function is set with navigation safety arrival rate and operational timeliness as its core. For extreme sea state scenarios, short-term meteorological and hydrological forecast data is combined to identify sea state avoidance windows. A segmented navigation path including avoidance standby points and accident operation points is planned. Emergency resources are dispatched to avoidance standby points first, and then proceed to the accident point to perform tasks after entering the operation window.

[0012] Optionally, in step S5, the sudden abnormal situation includes emergency resource failure, marine environment deterioration, accident level escalation, and sudden changes in the navigation environment; the dynamic update process of the scheduling scheme is as follows: based on the real-time data of the abnormal situation, the accident assessment and resource-path coupling optimization steps are re-executed to generate an updated scheduling scheme and synchronize it to the execution end; for local abnormal situations, the pre-configured backup resources and backup paths are directly invoked.

[0013] In summary, this application includes at least one of the following beneficial technical effects: This application shifts the emergency control focus from post-incident handling to pre-incident early warning, constructing a full-process emergency dispatch system encompassing pre-incident risk prediction, dynamic in-process handling, and closed-loop adjustment. It identifies collision risks in advance through spatiotemporal prediction of vessel trajectories, triggering pre-deployment of resources and significantly reducing emergency response time after an accident. The core of this system is a multi-objective optimization model coupling emergency resource matching and navigation path planning. Through a global emergency resource pool and a federated learning framework, it breaks down the limitations of local resources at the facility, achieving cross-unit, cross-industry, and cross-regional global emergency resource collaboration. Simultaneously, it completes joint training and optimal solution of the resource matching model without disclosing sensitive data of all participants, enabling real-time tracking of emergency resource execution status. Overall, this significantly improves the intelligence level, response efficiency, and safety reliability of emergency response to vessel collision accidents at offshore new energy facilities, effectively reducing various losses and environmental risks caused by accidents. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating the emergency resource dispatching method for ship collision accidents at offshore new energy power stations in this application. Detailed Implementation

[0015] The embodiments of this application are described in detail below, and examples of the embodiments are shown in the accompanying drawings.

[0016] In the description of this specification, the references to "certain embodiments," "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples" refer to specific features, structures, materials, or characteristics described in connection with the described embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0017] This application discloses an emergency resource dispatching method for ship collision accidents at offshore new energy power stations, referring to... Figure 1 It includes the following steps: S1. Synchronously collect accident data of involved vessels and stations, emergency resource data of the entire region, and marine meteorological, hydrological and navigation environment data; perform noise filtering, spatiotemporal synchronization and heterogeneous semantic fusion on the collected multi-source heterogeneous data, establish data association index, and generate standardized association dataset; S2. Based on historical and real-time navigation data of ships and marine environmental data, predict the navigation trajectory of ships within a set time period through a spatiotemporal prediction model, and calculate the probability of collision risk between ships and offshore new energy station structures; when the probability of collision risk exceeds the preset threshold, trigger risk warning and emergency resource pre-dispatch instructions, and dispatch nearby emergency resources to stand by in the vicinity of the warning area. S3. After a ship collision accident occurs, based on the standardized associated dataset, the accident scene is classified and multi-factor coupled accident scene is identified. The core feature parameters of the accident are extracted. Through the pre-trained accident judgment model, the accident level, emergency demand priority and future evolution trend of the accident are output simultaneously, and the emergency response target is dynamically updated. S4. Based on emergency response objectives, accident evolution trends, and real-time marine environmental data, with the optimization objectives of optimal response time, highest resource adaptability, lowest scheduling cost, and lowest navigation safety risk, a multi-objective optimization model coupling emergency resource matching and navigation path planning is constructed to solve for the optimal combination of emergency resources, the adaptive navigation path of each resource, and the arrival time sequence. S5. Synchronize dispatch instructions and navigation path information to each emergency resource execution terminal, and track the resource operation status and task execution progress in real time; trigger the dynamic update process of the dispatch plan in response to sudden abnormal situations during the execution process.

[0018] Specifically, step S1 involves multi-source data acquisition and standardized fusion processing. First, a three-dimensional data access system covering the entire dimensions of air, sea, people, and machines for accidents, resources, and the environment is constructed to achieve synchronous acquisition and standardized fusion of data across all scenarios, providing accurate and comprehensive data support for subsequent full-process emergency dispatch.

[0019] During the data acquisition phase, three main categories of core data are collected simultaneously: The first category is accident-related data, which specifically includes: Data collected via ship AIS terminals includes vessel type, tonnage, speed, course, location coordinates, and vessel identification. Vibration sensors deployed on offshore new energy power plant structures such as wind turbine foundations, towers, and offshore substations collect vibration amplitude and frequency data to assess damage after collisions. High-definition cameras on board and high-definition monitoring equipment deployed around the power plant collect image data of the accident scene, collision locations, personnel distress, and oil spills. Positioning wristbands and vital sign monitoring devices worn by on-site personnel and crew collect location, heart rate, and blood pressure data. Satellite remote sensing data is integrated to obtain information on sea conditions and oil spill spread over a wide area. Ship sonar data is also integrated to obtain collision damage data on underwater structures of vessels and underwater foundations of power plant structures.

[0020] The second category is comprehensive emergency resource data, which specifically includes: Relying on the BeiDou satellite navigation system and IoT sensors, real-time data of various emergency resources is collected, covering rescue ships, rescue helicopters, medical emergency teams and equipment, site repair teams and equipment, oil spill prevention equipment and teams, and emergency support materials. The specific data collected includes the real-time location, operating status (idle / occupied / faulty), core performance parameters (such as the speed, wind resistance level, and range of rescue ships, the applicable scenarios of medical equipment, and the types of structures that repair equipment is compatible with), scenario adaptation tags, affiliated management units, and dispatch contact information of each resource, so as to achieve real-time perception of the status of emergency resources across the entire region.

[0021] The third category is marine environment-related data, specifically including: Meteorological and hydrological data such as wave height, wind speed, wind direction, visibility, tides, and ocean current speed were collected from meteorological and hydrological stations around the accident site and surrounding sea areas; navigation environment data such as the distribution of surrounding vessels, channel congestion, channel control information, and the distribution of prohibited and obstructed areas were collected from the maritime department's regional navigation monitoring system.

[0022] In the data preprocessing and fusion stage, the collected multi-source heterogeneous data are standardized and fused. First, a Kalman filter algorithm is used to filter noise from damage data collected by vibration sensors, positioning data collected by BeiDou and AIS, and dynamic environmental data collected by meteorological and hydrological data, thereby eliminating random interference and measurement errors and improving data accuracy. The Kalman filter is implemented through preset system state equations and observation equations. The system state equations describe the dynamic change law of the corresponding data, and the observation equations describe the mapping relationship between the sensor-collected data and the actual state. Those skilled in the art can set appropriate equation parameters according to the type of collected data to complete the noise filtering.

[0023] Secondly, based on a unified UTC timestamp, all collected multi-source data are spatiotemporally synchronized to eliminate time discrepancies between different devices and data sources, ensuring consistency of all data in the time dimension. Then, for unstructured data such as satellite remote sensing images and high-resolution on-site images, the VisionTransformer (ViT) algorithm is used for feature extraction. Specifically, the input image is divided into multiple fixed-size image blocks, each block is converted into a corresponding embedding vector, and after adding location encoding information, it is input into the Transformer encoder. Through a self-attention mechanism, the global and local features of the image are learned, and a feature vector containing key information such as collision damage, oil spill area, and personnel location is output, completing the structured transformation of unstructured data. Next, a graph neural network (GNN) is used to construct a correlation graph of multi-source data. Various types of data from different data sources are used as nodes in the graph, and spatiotemporal and business correlations between data are used as edges. Deep learning of node features and correlations is performed through graph convolutional layers to complete semantic alignment and feature fusion of heterogeneous data from different sources and of different types, eliminating semantic discrepancies between data and achieving deep fusion of multi-source data.

[0024] Finally, all processed data is converted into a standardized structured format, each data point is assigned a unique identifier, and a correlation index is established according to the three-dimensional dimensions of "accident ID-resource ID-collection time". Data of the same accident, the same resource, and the same time dimension are linked and bound together to generate a standardized correlation dataset, providing a unified data foundation for subsequent early warning, analysis, and scheduling.

[0025] Step S2 is pre-collision risk warning and resource pre-scheduling, which moves the emergency control checkpoint forward, realizes the early prediction of collision risks and the pre-deployment of emergency resources, and reduces the emergency response time after an accident occurs from the source.

[0026] The specific implementation process is as follows: First, we continuously collect historical and real-time AIS navigation data of passing ships in the sea area, as well as corresponding marine meteorological and hydrological data and ocean current data, as the basic data for trajectory prediction.

[0027] Then, the pre-trained spatiotemporal prediction model is used to predict the ship's navigation trajectory within the next 10-30 minutes. In this embodiment, the spatiotemporal prediction model adopts the spatiotemporal Transformer model, which can effectively capture the spatiotemporal dependence characteristics of the ship's navigation trajectory, and at the same time integrate the influence of environmental factors such as weather and ocean currents on the ship's navigation, so as to achieve accurate prediction of the ship's future trajectory. The input of the spatiotemporal prediction model is the ship's historical navigation sequence data, real-time navigation data, and environmental data, and the output is the trajectory coordinate sequence of the ship within the future set time period.

[0028] Next, based on the ship's future navigation trajectory output by the model, and combined with the coordinate range and safety protection range of each structure of the offshore new energy station, the collision risk probability between the ship and the station structure is calculated. The collision risk probability can be calculated by comprehensively considering parameters such as the closest distance between the ship's predicted trajectory and the structure, the ship's heading deviation, and its speed.

[0029] When the calculated collision risk probability exceeds a preset safety threshold, a collision risk warning is automatically triggered for the vessel involved, the management of the offshore new energy station, and the maritime regulatory department, reminding the vessel to correct its course in a timely manner to avoid collision risks. Simultaneously, an emergency resource pre-dispatch instruction is automatically generated. Based on the collision risk level, anticipated pre-response needs, and the real-time status of surrounding emergency resources, the nearest emergency resources suitable for pre-response needs are selected, and pre-dispatch instructions are issued to their execution terminals. The instructions clearly specify the standby location, standby requirements, and emergency preparations, dispatching emergency resources to safe locations near the warning area in advance to be on standby for emergency response. Once a subsequent vessel collision occurs, the standby emergency resources can arrive at the accident scene in the shortest possible time to carry out response, significantly reducing emergency response time and preventing the expansion of accident losses.

[0030] Step S3 involves intelligent identification and dynamic evolution analysis of the accident scenario. Once a ship collision accident has occurred, this step uses a standardized associated dataset to accurately identify the accident scenario and dynamically analyze the accident evolution trend, providing precise targets and needs for subsequent resource scheduling.

[0031] The specific implementation process is divided into two stages: accident scene classification and identification, and accident level and evolution trend analysis. In the accident scene classification and identification stage, based on the standardized association dataset obtained in step S1, the collision location coordinates, the AIS identifier of the involved vessel, and the GIS map data of the station facilities are extracted. The collision location coordinates are matched with the coordinate range of structures such as wind turbines, offshore substations, and photovoltaic platforms. Combined with the vessel's AIS data and on-site image feature data, the accident scene is classified and identified using a pre-constructed decision tree algorithm.

[0032] The specific decision tree judgment logic is as follows: First, it is determined whether the collision location coordinates fall within the coordinate range of the station structure. If so, it is determined to be the core scenario of "ship-station structure contact". Then, combined with the damage data of vibration sensors and image feature data, it is further subdivided into sub-scenarios such as wind turbine foundation contact, tower contact, offshore substation contact, and photovoltaic platform contact. If the collision location coordinates are not within the range of the station structure and the collision object is the AIS identifier of another ship, it is determined to be the scenario of "ship-to-ship collision". At the same time, combined with real-time environmental data, oil spill data from image recognition, and personnel vital signs data, it completes the identification of multi-factor coupled accident scenarios, such as "ship contact with wind turbine + typhoon approach", "ship collision + fuel leak + personnel falling into the water", and "structure damage + secondary capsizing risk" and other complex coupled scenarios, to achieve comprehensive and accurate identification of accident scenarios.

[0033] In the accident severity and evolution trend assessment phase, core characteristic parameters are first extracted from a standardized associated dataset. These parameters include the number of casualties and distressed persons, vital signs of personnel, extent of damage to facilities and structures, damage to vessels involved, presence and volume of oil spill, sea conditions and navigation environment in the surrounding waters, and the risk of secondary disasters. The extracted core characteristic parameters are then input into a pre-trained accident assessment model. The model simultaneously outputs the current accident severity level, emergency response priority, and future evolution trend of the accident.

[0034] Accident levels are classified into four categories—particularly serious, serious, relatively serious, and general—according to relevant national standards for maritime emergency management. Emergency demand priorities are used to clarify the core sequence of emergency response, such as prioritizing personnel rescue, oil spill prevention, hazard control, and structure repair, while also specifying the priority of various emergency resources. The accident assessment model employs a deep learning model integrating a time-series prediction module. The main body uses a fully connected neural network architecture, and the time-series prediction module is built on a Transformer-XL network, effectively capturing the temporal dependencies of long sequences. Combining historical accident evolution patterns, current accident characteristics, and real-time environmental dynamic data, it predicts the future evolution trend of the accident scenario, such as the extent and speed of oil spill spread, the risk of increased damage to structures, the impact of extreme weather on accident response, and the probability of secondary disasters. Based on the predicted evolution trend, the model synchronously updates the emergency demand priorities and emergency response objectives, achieving a linked assessment of "current scenario identification - future trend prediction - dynamic update of response objectives," avoiding the problem of static assessment being disconnected from the dynamic development of the accident.

[0035] The specific steps for pre-training the accident assessment model are as follows: First, historical data on maritime ship collision accidents and emergency response cases at offshore new energy power stations were collected, covering a full range of cases with different accident scenarios, accident levels, environmental conditions, and response effects. Then, the collected case data was standardized and labeled, including accident scenario classification, accident level, core emergency needs, need priority, accident evolution process, environmental change data, and response effects, constructing a standardized training dataset. Next, the training dataset was divided into training, validation, and test sets in an 8:1:1 ratio. Model training hyperparameters, including learning rate, iteration count, batch size, and loss function, were set. Supervised training of the model was conducted using the training set, and the model's generalization ability was validated using the validation set. Model parameters were adjusted to avoid overfitting. Finally, the performance of the trained model was tested using the test set to ensure that the model's scene recognition accuracy, accident level determination accuracy, and evolution trend prediction accuracy meet the requirements of emergency response, completing the model's pre-training. In practical applications, the model can undergo incremental learning and continuous optimization based on newly added accident response cases, continuously improving its adaptability and accuracy.

[0036] Step S4 is a multi-objective optimization scheduling that couples resource matching and route planning. It breaks the rigid logic of existing technologies that select resources first and then plan routes, and constructs a multi-objective optimization model that couples emergency resource matching and navigation route planning. This achieves joint optimization of resources and routes, ensuring that the scheduling scheme meets both emergency response needs and adapts to complex maritime environmental conditions.

[0037] First, a comprehensive emergency resource pool is pre-built and dynamically maintained. Based on the functional attributes of emergency resources, all resources are categorized into five main types: personnel rescue, medical emergency, station repair, oil spill prevention, and emergency support. Each emergency resource in the pool is labeled with corresponding basic attributes and scenario adaptation tags. Basic attributes include resource location, performance parameters, affiliated unit, and dispatch contact information. Scenario adaptation tags include the applicable accident scenarios, response tasks, and environmental adaptability (such as wind resistance level and sea state suitability). Simultaneously, relying on the resource data collected in real-time in step S1, the resource status in the pool is updated in real-time, marking the idle, occupied, and faulty states of resources and automatically removing unusable resources to ensure the real-time nature and accuracy of the resource pool data. To overcome the limitations of local station resources, this embodiment's emergency resource pool also integrates cross-domain emergency resources, including social emergency rescue resources from coastal ports, civilian medical and rescue forces, patrol and law enforcement vessels from maritime departments, coast guard rescue resources, and dispatchable fishing vessels in the surrounding area, constructing a "government-enterprise collaborative, comprehensive coverage" emergency resource network and expanding the dispatch boundaries of emergency resources.

[0038] To address the collaborative matching of cross-domain emergency resources, this embodiment employs a federated learning framework to resolve privacy and security issues related to cross-domain data sharing, achieving comprehensive resource collaborative optimization while protecting privacy. The specific implementation process of the federated learning framework is as follows: The managers of each cross-domain resource in the emergency resource pool are designated as participating nodes in the federated learning process. Examples include site maintenance units, maritime departments, port rescue units, and civilian rescue organizations. Each participating node only possesses the resource data it manages and does not need to share raw data externally. Secondly, each participating node preprocesses its own resource data locally, constructs a local resource matching sub-model, and trains the sub-model based on its local data. Only the trained model parameters, not the original resource data, are encrypted and uploaded to the central node of the federated learning process. The central node receives the encrypted model parameters uploaded by each participating node. A federated averaging algorithm is used to aggregate the model parameters of multiple nodes and update them to obtain a global resource matching model. The central node encrypts and distributes the updated global model parameters to each participating node. Each participating node updates and optimizes its local sub-model based on the global model parameters. Through multiple rounds of iterative processes of local training, parameter uploading, global aggregation, and parameter distribution, the joint training of the cross-domain resource matching model is completed. The final global model can adapt to cross-domain resources across the entire domain, achieving optimal matching of resources across units and industries. At the same time, there is no need to share the original sensitive data of each participating party throughout the process, effectively solving the contradiction between information silos and data privacy protection.

[0039] Based on this, and using the emergency response objectives, emergency demand priorities, and accident evolution trends output in step S3, combined with the real-time marine environmental data collected in step S1, a multi-objective optimization model coupling emergency resource matching and navigation path planning is constructed to achieve joint optimization of resources and paths.

[0040] The objective function of the multi-objective optimization model is: ; The meanings of each parameter are clearly defined as follows: The decision variables to be solved are the combined scheme of emergency resource combinations and corresponding navigation routes, including the types and quantities of various emergency resources selected in the scheme, as well as core information such as the navigation routes and estimated arrival times of each resource. The comprehensive optimization objective function is a comprehensive solution objective for multi-objective optimization. The comprehensive response time function of the resource combination represents the comprehensive value of the response time of each emergency resource in the resource combination from the current location to the accident point. In this embodiment, the weighted comprehensive value of the longest response time of each resource to the accident point and the weighted average response time is adopted. The weight is adjusted based on the priority of emergency needs to ensure that the resources corresponding to the core needs can arrive as quickly as possible. , , , These are the weight coefficients corresponding to the four optimization objectives: response time, resource adaptability, scheduling cost, and security risk, and they satisfy the following conditions: The weighting coefficients are dynamically adjusted based on the accident level and emergency response objectives. For example, the higher the accident level, the more weighting coefficients are applied. and , The higher the value, the more priority is given to ensuring the timeliness of emergency response, the adaptability of resources, and the safety of navigation. When the accident level is low and the handling needs are not urgent, the value can be appropriately increased. The weighting is adjusted while also taking into account the control of scheduling costs.

[0041] The priority-weighted dynamic response time function The expression is: Where n is the total number of emergency resources included in the emergency resource combination; As the priority weight of the disposal task corresponding to the i-th resource, in this embodiment, the weight of the first-level core needs such as personnel rescue and oil spill prevention is 0.8~1.0, the weight of the second-level needs such as site hazard control is 0.4~0.7, and the weight of the third-level needs such as subsequent emergency repair support is 0.1~0.3, and the weight of all tasks is normalized. The total voyage distance for the planned navigation path corresponding to the i-th resource is determined in real time by the path planning stage; Let i be the dynamic equivalent speed of the i-th resource on the planned path: ;in The nominal still water speed of the vessel. The speed reduction coefficient due to wave height is obtained by fitting the ship's wave resistance level with the real-time wave height. The higher the wave height, the smaller the reduction coefficient. The speed reduction coefficient due to wind speed is obtained by fitting the wind speed and the ship's windward area. The speed correction factor for the ocean current is greater than 1 when sailing with the current and less than 1 when sailing against the current. It is calculated based on the angle between the current direction and the ship's course and the current speed. Thus, the dynamic equivalent speed is calculated based on the nominal speed of the resource, real-time wave height, wind speed, current direction and speed, and tidal state, rather than the static nominal speed. Let be the pre-scheduling state coefficient for the i-th resource. In this embodiment, it refers to standby resources that have been pre-scheduled to a 3-nautical-mile radius from the warning area. Take 0.5, take 0.4 for resources that have been pre-scheduled to within 5 nautical miles, and take 0 for resources that have not been pre-scheduled; The waiting time for the timing of the i-th resource is set to 0 in this embodiment. The waiting time for priority tasks such as personnel rescue and oil spill prevention is 0. The site repair resources need to enter the site after personnel rescue, oil spill prevention, and hazard investigation are completed. The waiting time is determined based on the estimated completion time of the preceding tasks. If multiple resources need to work together, the waiting time is matched based on the timing requirements of the collaborative work.

[0042] The comprehensive adaptability function of resource combination and emergency needs, with a value range between 0 and 1. The higher the value, the better the adaptability between resources and emergency response needs. The adaptability is calculated based on the matching degree between resource scenario adaptability tags, performance parameters and emergency needs. The hierarchical dynamic adaptation function The expression is: in, , , These are the weight coefficients for core requirement adaptation, combined collaborative adaptation, and evolutionary redundancy adaptation, respectively. The weighting is dynamically adjusted based on the accident level and evolving risk; for major and above-level accidents, Take 0.6, Take 0.25, Take 0.15, prioritizing the matching of core needs; for general accidents, Take 0.5, Take 0.2, We set it to 0.3 to balance redundancy and adaptability.

[0043] For the core requirement level adaptability, the expression is: in, This represents the total number of emergency needs. Let k be the priority weight of the emergency demand. For resource combination For the first The single-item matching degree of each requirement, with a value range of [0,1], is calculated based on the degree of matching between resource function parameters, scenario adaptation tags, and requirements; in this embodiment, the core requirement adaptation degree... In the calculation, matching degree calculation rules are set for each emergency need: for example, for personnel rescue needs, rescue ships with maritime life-saving qualifications, equipped with life-saving equipment and medical personnel, have a single matching degree. Take 1; for vessels with only basic navigation capabilities and no rescue qualifications, take 0.2; for resources that have no personnel rescue capabilities at all, take 0.

[0044] The expression for resource combination and collaborative adaptability is: in, For the first in the resource portfolio The and the first The collaboration and adaptation coefficient of each resource is as follows: for resources belonging to the same station operation and maintenance unit, the collaboration coefficient is 1; for resources with regular collaboration between the station and the maritime department, the collaboration coefficient is 0.9; for resources with no regular collaboration mechanism across industries, the collaboration coefficient is 0.7; for resources with conflicting workspaces (such as large crane vessels and small rescue vessels operating on the same work surface), the collaboration coefficient is reduced to 0.5~0.8 according to the degree of conflict; for resources with incompatible communication systems, the collaboration coefficient is reduced to below 0.6. The redundancy fit for evolutionary requirements is expressed as: Among them, based on the evolution trend output by the accident assessment model, such as predicting that the typhoon will arrive in the accident area in 3 hours, the redundancy coefficient is calculated if the ship's wind resistance level exceeds the current sea state requirements and can cover the sea state after the typhoon arrives. Take 1; if it just meets the current sea conditions but cannot cope with the typhoon, take 0.2; if it is predicted that the accident may cause oil spill to spread, then the redundancy coefficient of the multi-functional rescue vessel equipped with oil spill prevention equipment is taken as 1, and the redundancy coefficient of the vessel with only personnel rescue capability is taken as 0.3.

[0045] The comprehensive scheduling cost function of the resource combination represents the total scheduling cost corresponding to the resource combination scheme, including the comprehensive value of labor cost, fuel cost, equipment operation and maintenance cost, material consumption cost, etc. Specifically: In the formula: The explicit direct costs of resource allocation include fuel costs, labor costs, material consumption costs, and equipment leasing costs, calculated based on resource range, operation duration, personnel allocation, and material consumption. The hidden equipment wear and tear costs are calculated by fitting the sea condition severity of the navigation route and the wear and tear characteristics of the ship's equipment. High-load navigation under extreme sea conditions corresponds to higher wear and tear costs, while the basic wear and tear value is taken under calm sea conditions. For cross-domain collaboration costs, the calculation is based on the cross-domain level of the resource's owner, the complexity of the coordination process, and the cost of cross-system integration. Local site resources are set to 0, while cross-unit and cross-industry resources have corresponding costs added according to the difficulty of collaboration. For example, local site operation and maintenance resources are set to 0, cross-regional resources within the same group are set to basic collaboration costs, government departments such as maritime and coast guard are set to administrative coordination costs, and non-governmental social resources are set to business integration costs, which are added level by level according to the cross-domain level. The opportunity cost of resource allocation is calculated based on the risk cost caused by the gap in emergency support capabilities and the decrease in redundancy of backup resources in the area of ​​responsibility after the resource is allocated. The opportunity cost of core backup resources is higher, while that of non-core redundant resources is lower. For core backup rescue vessels at the station, allocation will lead to a decrease in the emergency support capabilities of the entire station area, and the opportunity cost is 50% to 100% of the daily emergency support value of the vessel. For redundant backup resources in the port and dispatchable fishing vessels in the surrounding area, the opportunity cost is lower or 0.

[0046] The cost of accident evolution risk is expressed as: in, The probability of incident escalation is calculated based on the response time and adaptability of resource combinations. The longer the response time and the lower the adaptability of core requirements, the higher the probability of incident escalation. To calculate the total additional handling costs after an accident escalates, based on historical accident case data, the average additional handling cost corresponding to each escalation level is determined, thus achieving a quantitative calculation of risk costs.

[0047] The comprehensive safety risk function of the navigation path represents the comprehensive safety risk of the navigation path corresponding to each resource in the resource combination, including the comprehensive value of sea state risk, ship collision risk, and regulatory constraint risk. The higher the value, the greater the risk. The spatiotemporal dynamic cumulative security risk function The expression is: Where n is the total number of emergency resources included in the emergency resource combination; The total number of path points after discretization of the planned navigation path for the i-th resource; This represents the proportion of the segmented flight distance corresponding to the p-th path point on the i-th resource path to the total flight distance. The comprehensive dynamic risk value is the comprehensive dynamic risk value of the i-th resource at the p-th path point. The assessment values ​​are obtained based on environmental risk, navigation risk, and secondary environmental risk.

[0048] ; in: , , These are the weighting coefficients for environmental risk, navigation risk, and secondary environmental risk, respectively, which can be adjusted based on the characteristics of the navigation area. The dynamic sea state environmental risk value is calculated by fitting the real-time wave height, wind speed, and ocean current data of the waypoints and the ship seaworthiness parameters (wind resistance level, stability, and wave resistance) of the i-th resource. The value range is [0,1], and it is taken as 1 when it exceeds the ship seaworthiness limit. The collision risk value is calculated based on the distribution of ships around the waypoint, real-time speed, predicted trajectory, and combined with the ship domain model. The value range is [0,1]. The secondary risk value is set for environmentally sensitive areas. If the path point is located in an environmentally sensitive area such as a marine ecological protection area, aquaculture area, or drinking water source protection area, the corresponding preset high-risk value is used. For non-sensitive areas, the value is 0.

[0049] To solve the aforementioned multi-objective optimization model, this embodiment employs the Alternating Direction Multiplier Method (ADMM) distributed optimization algorithm. This algorithm can decompose large-scale joint optimization problems into multiple small-scale sub-problems for distributed solution, significantly improving the solution speed. It is particularly suitable for large-scale optimization scenarios involving global and cross-domain resources, and can quickly converge to obtain the Pareto optimal solution, namely the optimal combination of emergency resources, the adaptive navigation path of each resource, and the arrival sequence. At the same time, it automatically excludes resources that exceed the performance adaptation range or cannot meet the safe navigation requirements, ensuring the rationality and feasibility of the scheduling scheme.

[0050] For refined planning and dynamic optimization of navigation paths, this embodiment employs the PPO reinforcement learning algorithm. This algorithm exhibits stable training performance and excellent convergence in scenarios with continuous action spaces, and can adapt to dynamically changing maritime environmental conditions, enabling real-time adaptive optimization of navigation paths. The specific settings of the PPO reinforcement learning algorithm are as follows: State space S: Defined as the total state information that the algorithm can observe, including the current position, course, speed, and performance parameters of the emergency vessel, the real-time sea state parameters and navigation environment parameters of the surrounding sea area, the location of fixed constraint areas (no navigation zone, obstacle zone), the location of the accident point, the time requirements for emergency response, etc., to achieve a comprehensive perception of the navigation environment and its own state. Action Space A: Defined as the navigation adjustment actions that the algorithm can execute, including continuous and discrete actions such as ship course adjustment, speed adjustment, route inflection point adjustment, and selection of hazard avoidance standby point, covering the core dimensions of ship navigation control; The reward function R is set around the safe arrival rate and operational efficiency of navigation. The specific rules are as follows: a high positive reward is given when the planned route completely avoids dangerous areas, restricted areas, and navigation conflicts, and can safely reach the accident point within the target time; a corresponding positive reward is given when the route has shorter navigation time and lower energy consumption; a high negative reward is given when the route enters a restricted area, a sea state exceeding the ship's seaworthiness, or poses a collision risk with other ships; and a corresponding negative reward is given when the accident point cannot be reached within the specified time. Through the setting of the reward function, the algorithm is guided to learn the optimal navigation route that balances safety, efficiency, and economy.

[0051] For extreme sea state scenarios such as typhoons and strong storm surges, this embodiment also implements segmented route planning under extreme sea conditions. The specific implementation process is as follows: Based on short-term meteorological and hydrological forecast data issued by meteorological departments, the trend of sea state changes over a future period is accurately identified, and suitable safe-haven periods for ship navigation and emergency operations are delineated. Based on the safe-haven periods and emergency response needs, a segmented navigation route is planned, including safe-haven standby points and accident operation points. The safe-haven standby points are selected in relatively stable sea areas where wind and waves can be safely avoided. Emergency resources are dispatched to the safe-haven standby points to wait. Once the sea conditions ease and the operation window opens, they are quickly dispatched from the standby points to the accident points to carry out emergency response tasks. Through segmented route planning, an upgrade from "passively avoiding severe sea conditions" to "actively planning safe-haven and operation windows" is achieved, significantly improving the feasibility and safety of emergency dispatch under extreme sea conditions and preventing safety accidents from occurring while emergency resources are navigating in extreme sea conditions.

[0052] Simultaneously, based on the emergency demand priority output in step S3, and combined with the navigation paths and estimated arrival times of various emergency resources, the arrival sequence of multiple resources is planned to ensure smooth coordination of on-site emergency response. For example, for accident scenarios with the highest priority of personnel rescue, the routes and arrival times of rescue vessels and medical emergency resources are planned first to ensure they arrive at the accident site first to carry out personnel rescue; for scenarios with oil spill risk, while ensuring personnel rescue, oil spill control resources are dispatched to arrive first to deploy oil booms and prevent the spread of oil spills; for scenarios involving damage to site structures, repair resources are dispatched to arrive after the hazard is controlled to carry out repair operations, achieving orderly arrival and efficient coordination of various resources, and avoiding problems such as resource congestion and poor coordination of response processes.

[0053] Step S5 enables the issuance of emergency dispatch instructions, tracking of execution status, and dynamic adjustment of sudden abnormal situations, completing a closed-loop management of the entire process from pre-warning to in-process handling to dynamic adjustment, ensuring the continuity and reliability of emergency response.

[0054] The specific implementation process is divided into two stages: synchronization of dispatch instructions and status feedback, and dynamic adjustment in the event of emergencies. In the stage of synchronization of dispatch instructions and status feedback, relying on multi-channel maritime communication technologies, including maritime satellite communication, maritime 4G / 5G communication, and microwave communication, dispatch instructions are synchronously issued to the selected execution terminals of each emergency resource. These instructions clearly define core information such as task assignments, handling objectives, accident location information, planned navigation routes, arrival time requirements, and coordination requirements, ensuring that each execution terminal can clearly and comprehensively grasp the emergency task requirements. Simultaneously, through the BeiDou satellite navigation system and IoT sensors, the operational status, real-time location, speed, heading, and task execution progress of each emergency resource are tracked in real time at a set high frequency. A status feedback mechanism is established, with each execution terminal periodically transmitting information on task execution, on-site accident handling progress, and changes in the on-site environment, ensuring that the dispatch and command center can grasp the entire process of emergency response in real time, providing data support for dynamic adjustments.

[0055] In the dynamic adjustment phase for sudden anomalies, a full-process anomaly monitoring mechanism is established to monitor the operational status of emergency resources, changes in the marine environment, the handling of accident sites, and changes in the navigation environment in real time. When a sudden anomaly is identified, the dynamic update process of the dispatch plan is automatically triggered. These sudden anomalies include: emergency resources malfunctioning and unable to continue their mission; deterioration of the marine environment, such as wind speed and wave height exceeding the safe range for ship navigation, and a significant decrease in visibility; escalation of the accident site, such as increased damage to structures, expansion of the oil spill area, new personnel distress situations, and increased risk of secondary disasters; and sudden changes in the navigation environment, such as temporary channel closures or new navigation obstacles.

[0056] The specific dynamic update process of the scheduling scheme is as follows: Upon identifying a sudden anomaly, real-time full data corresponding to the anomaly is immediately collected. The updated real-time data is then input into the accident analysis model in step S3 to re-evaluate the accident scenario, accident level, emergency needs, and evolution trend, updating the emergency response objectives. Based on the updated emergency response objectives, the resource-path coupling optimization step in step S4 is re-executed to generate updated emergency resource combination schemes, navigation paths, and arrival sequences. The updated scheduling instructions are immediately and synchronously sent to the corresponding execution terminals via multi-channel communication technology, enabling rapid dynamic adjustment of the scheduling scheme. For local anomalies such as single resource failures or single path impassability, pre-configured backup resources and backup paths can be directly invoked without re-solving the entire scheme, further shortening the adjustment response time and ensuring the continuity and stability of emergency response. Through this closed-loop dynamic control mechanism, various unexpected situations during emergency response can be effectively addressed, ensuring that the scheduling scheme always adapts to the actual situation on-site, significantly improving the reliability and resilience of emergency scheduling.

[0057] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for emergency resource dispatching in the event of a ship collision at a marine new energy power station, characterized in that, Includes the following steps: S1. Synchronously collect accident data of involved vessels and stations, emergency resource data of the entire region, and marine meteorological, hydrological and navigation environment data; perform noise filtering, spatiotemporal synchronization and heterogeneous semantic fusion on the collected multi-source heterogeneous data, establish data association index, and generate standardized association dataset; S2. Based on historical and real-time navigation data of ships and marine environmental data, predict the navigation trajectory of ships within a set time period through a spatiotemporal prediction model, and calculate the probability of collision risk between ships and offshore new energy station structures; when the probability of collision risk exceeds the preset threshold, trigger risk warning and emergency resource pre-dispatch instructions, and dispatch nearby emergency resources to stand by in the vicinity of the warning area. S3. After a ship collision accident occurs, based on the standardized associated dataset, the accident scene is classified and multi-factor coupled accident scene is identified. The core feature parameters of the accident are extracted. Through the pre-trained accident judgment model, the accident level, emergency demand priority and future evolution trend of the accident are output simultaneously, and the emergency response target is dynamically updated. S4. Based on emergency response objectives, accident evolution trends, and real-time marine environmental data, with the optimization objectives of optimal response time, highest resource adaptability, lowest scheduling cost, and lowest navigation safety risk, a multi-objective optimization model coupling emergency resource matching and navigation path planning is constructed to solve for the optimal combination of emergency resources, the adaptive navigation path of each resource, and the arrival time sequence. S5. Synchronize dispatch instructions and navigation path information to each emergency resource execution terminal, and track resource operation status and task execution progress in real time; trigger the dynamic update process of dispatch plan in response to sudden abnormal situations during execution. The objective function of the multi-objective optimization model is: ； in, The solution is a combined approach of emergency resource combinations and corresponding navigation paths. To comprehensively optimize the objectives; , , , These are the weighting coefficients for response time, adaptability, scheduling cost, and security risk, respectively. The weighting coefficients are dynamically adjusted based on the accident level and emergency response objectives. The dynamic response time function is weighted by priority, taking into account the dynamic impact of real-time sea conditions on speed, the priority weight of emergency needs, the advantages of pre-scheduled resources, and the constraints of the timing of handling. The hierarchical dynamic adaptability function comprehensively considers the matching of core requirements at different levels, the collaborative adaptation of resource combinations, the capability redundancy of accident evolution requirements, and the compatibility of cross-domain resource operations. It is a comprehensive cost function for the entire life cycle, which comprehensively considers explicit direct costs, implicit loss costs, cross-domain collaboration costs, opportunity costs, and accident evolution risk costs. This is a spatiotemporally dynamic cumulative safety risk function that comprehensively considers the ship's own seaworthiness, dynamic sea state risk, navigation collision risk, and secondary risks in environmentally sensitive areas, representing the cumulative risk across the entire path.

2. The emergency resource dispatching method for ship collision accidents at offshore new energy power stations according to claim 1, characterized in that, The priority-weighted dynamic response time function The expression is: Where n is the total number of emergency resources included in the emergency resource combination; The priority weight of the disposal task corresponding to the i-th resource is (0,1], with higher weights for core emergency needs, and the weight normalization constraint of all tasks under the same accident is satisfied. The total voyage distance for the planned navigation path corresponding to the i-th resource is determined in real time by the path planning stage; The dynamic equivalent speed of the i-th resource on the planned path is calculated based on the nominal speed, real-time wave height, wind speed, ocean current direction and speed, and tidal state of the resource, rather than the static nominal speed. Let be the pre-scheduled state coefficient of the i-th resource. If the resource is a pre-scheduled standby resource, Use a preset value of 0.3 to 0.

6. If the resource is scheduled after an incident, Set to 0; Let be the timing delay for the i-th resource, calculated based on the pre- and post-task constraints of emergency response. If this resource is a core priority task resource... If the value is 0, and the process needs to be completed before entering the stage, the waiting time is calculated based on the estimated completion time of the preceding task.

3. The emergency resource dispatching method for ship collision accidents at offshore new energy power stations according to claim 1, characterized in that, The hierarchical dynamic adaptation function The expression is: in, , , These are the weight coefficients for core requirement adaptation, combined collaborative adaptation, and evolutionary redundancy adaptation, respectively. The weights are dynamically adjusted based on the accident level and the risk of evolution. For the core requirement level adaptability, the expression is: in, This represents the total number of emergency needs. Let k be the priority weight of the emergency demand. For resource combination For the first The single-item matching degree of each requirement, with a value range of [0,1], is calculated based on the matching degree between resource function parameters, scenario adaptation tags and requirements; The expression for resource combination and collaborative adaptability is: in, For the first in the resource portfolio The and the first The collaboration adaptation coefficient of each resource is in the range of (0,1]. It is determined based on the operation timing compatibility, communication system compatibility, disposal space compatibility and the collaboration mechanism of the affiliated unit. When there is no collaboration obstacle, it is taken as 1. When there is collaboration constraint, it is adjusted down according to the degree of constraint. The redundancy fit for evolutionary requirements is expressed as: in, The redundancy coefficient of the i-th resource in response to the needs of accident evolution is denoted by [0,1]. Based on the future evolution trend output by the accident assessment model, the degree to which the performance parameter of the resource exceeds the current needs and covers the potential future needs is calculated. The value is 1 when the potential evolution needs are fully covered.

4. The emergency resource dispatching method for ship collision accidents at offshore new energy power stations according to claim 1, characterized in that, The comprehensive life-cycle cost function The cost is determined by the sum of the explicit direct costs of resource scheduling, the implicit equipment wear and tear costs, the cross-domain collaboration costs, the opportunity costs of resource allocation, and the costs of accident evolution risks.

5. The emergency resource dispatching method for ship collision accidents at offshore new energy power stations according to claim 1, characterized in that, The spatiotemporal dynamic cumulative security risk function The expression is: Where n is the total number of emergency resources included in the emergency resource combination; The total number of path points after discretization of the planned navigation path for the i-th resource; This represents the proportion of the segmented flight distance corresponding to the p-th path point on the i-th resource path to the total flight distance. The comprehensive dynamic risk value is the comprehensive dynamic risk value of the i-th resource at the p-th path point. The assessment values ​​are obtained based on environmental risk, navigation risk, and secondary environmental risk.

6. The emergency resource dispatching method for ship collision accidents at offshore new energy power stations according to claim 1, characterized in that, In steps S1-3: The accident data includes ship AIS data, vibration sensor data of station structures, high-definition on-site image data, personnel positioning and vital signs data, satellite remote sensing data, and ship sonar detection data. The emergency resource data includes the real-time location, operating status, performance parameters, and scenario adaptation attributes of rescue vessels, rescue helicopters, medical emergency equipment, station repair equipment, and oil spill prevention equipment. The environmental data includes wave height, wind speed, visibility, tides, ocean currents, as well as data on the distribution of surrounding vessels, waterway control, and congestion. The spatiotemporal prediction model adopts a spatiotemporal graph convolutional network or a spatiotemporal Transformer model, and the set duration is 10-30 minutes; the emergency resource pre-scheduling instruction is generated based on the collision risk level, pre-handling requirements and surrounding resource status, and clarifies the standby location, standby requirements and emergency preparation matters of the pre-scheduled resources. The multi-factor coupled accident scenarios include scenarios where a ship touches a station structure and extreme weather, or where a ship collision is combined with an oil spill and people falling into the water. The accident assessment model adopts a deep learning model that integrates a time-series prediction module. The time-series prediction module is built based on LSTM or Transformer-XL networks and is used to combine the accident evolution law with dynamic environmental data to predict the future evolution trend of the accident scenario and update the priority of emergency needs and emergency response objectives simultaneously.

7. The emergency resource dispatching method for ship collision accidents at offshore new energy power stations according to claim 1, characterized in that, Step S4 includes: A dynamically updated global emergency resource pool is pre-built. This pool is functionally categorized into personnel rescue, medical emergency, station repair, oil spill prevention, and emergency support resources. The operational status, scenario adaptation tags, and location information of each resource are updated in real time. The pool also integrates cross-domain emergency resources, including social emergency resources, maritime regulatory resources, and fisheries emergency resources. Collaborative matching of cross-domain emergency resources employs a federated learning framework. Each cross-domain resource manager acts as a participating node, completing resource data processing and sub-model training locally. Only model parameters are uploaded to the central node for global model aggregation and updates. This allows for the joint training and solution of the cross-domain resource matching model without sharing sensitive local data.

8. The emergency resource dispatching method for ship collision accidents at offshore new energy power stations according to claim 7, characterized in that, The multi-objective optimization model employs the Alternating Direction Multiplier Method (ADMM) for distributed solution. For navigation path planning and optimization, DQN or PPO reinforcement learning algorithms are used. The state space of the reinforcement learning algorithm includes real-time sea conditions, navigation environment, and ship navigation parameters, while the action space includes adjustments to heading, speed, and standby point selection. The reward function is set with navigation safety arrival rate and operational timeliness as its core. For extreme sea state scenarios, short-term meteorological and hydrological forecast data is combined to identify sea state avoidance windows. A segmented navigation path including avoidance standby points and accident operation points is planned. Emergency resources are dispatched to avoidance standby points first, and then proceed to the accident point to perform tasks after entering the operation window.

9. The emergency resource dispatching method for ship collision accidents at offshore new energy power stations according to claim 1, characterized in that, In step S5, the sudden abnormal situation includes emergency resource failure, marine environment deterioration, accident level escalation, and sudden changes in the navigation environment; the dynamic update process of the scheduling scheme is as follows: based on the real-time data of the abnormal situation, the accident assessment and resource-path coupling optimization steps are re-executed to generate an updated scheduling scheme and synchronize it to the execution end; for local abnormal situations, the pre-configured backup resources and backup paths are directly called.

Images