A method and system for risk assessment and graded early warning of offshore wind power vessels

Abstract

This invention discloses a method and system for risk assessment and graded early warning of offshore wind farm vessels. The method involves collecting and fusing multi-source sensing data; performing spatiotemporal alignment and fusion processing on the fused multi-source sensing data to generate a bird's-eye view feature representation containing spatiotemporal characteristic information of the vessel; based on the bird's-eye view feature representation, employing a fusion prediction model combining a vessel motion physics model and a time series prediction algorithm, calculating the motion reachability set of the target vessel within the predicted time period; conducting a risk assessment based on the spatial relationship between the motion reachability set and a preset safe area of ​​the wind farm, and triggering graded early warnings based on the assessment results. This invention solves the problems of blind spots in single-source monitoring and high false alarm rates in traditional early warning systems, achieving accurate and dynamic assessment and proactive early warning of vessel collision risks in offshore wind farm waters, significantly improving regulatory safety and efficiency.

Description

Technical Field

[0001] This application relates to the field of maritime traffic safety and intelligent supervision technology, and in particular to a method and system for risk assessment and graded early warning of offshore wind power vessels. Background Technology

[0002] With the rapid development of the offshore wind power industry, maritime traffic in wind farm waters is becoming increasingly busy, significantly increasing the risk of collisions between ships and wind turbines. However, traditional maritime traffic safety monitoring methods have significant shortcomings. On the one hand, existing systems often rely on a single data source, such as radar or Automatic Identification System (AIS) for monitoring, each with its own limitations: some ships may disable AIS, making them untrackable, while radar is easily affected by wind turbine obstruction in the complex environment of wind farms, generating numerous false echoes and detection blind spots. On the other hand, traditional risk assessment and early warning methods (such as fixed-distance electronic fences) are often based solely on static distance thresholds, failing to fully consider the dynamic motion performance of ships, the pilot's maneuvering intentions, and the impact of environmental factors such as wind, waves, and currents. This results in low accuracy and a high false alarm rate, failing to meet the urgent need for refined and intelligent safety monitoring of offshore wind farms.

[0003] Therefore, how to achieve accurate and dynamic assessment and proactive early warning of ship collision risks in offshore wind farm waters, and avoid the problems of blind spots in monitoring from a single data source and the high false alarm rate of traditional early warning systems, is a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0004] This application provides a method and system for risk assessment and graded early warning of offshore wind power vessels, which can solve the problems of blind spots in monitoring from a single data source and the high false alarm rate of traditional early warning, thereby improving regulatory safety and efficiency.

[0005] The first aspect of this application provides a method for assessing and classifying navigation risks of offshore wind power vessels, including:

[0006] Collect and integrate multi-source sensing data, which includes at least offshore wind farm location and range data, radar data, ship automatic identification system data, video data, and meteorological data;

[0007] The fused multi-source sensing data is spatiotemporally aligned and fused to generate a bird's-eye view feature representation containing ship spatiotemporal feature information.

[0008] Based on the feature representation of the bird's-eye view, a fusion prediction model combining the ship motion physics model and time series prediction algorithm is adopted to calculate the motion reach set of the target ship within the prediction time period;

[0009] Risk assessment is conducted based on the spatial relationship between the mobility accessibility set and the pre-defined safe zone of the wind farm, and graded early warnings are triggered based on the assessment results.

[0010] Optionally, multi-source sensing data are collected and fused, including:

[0011] Obtain location and extent data of offshore wind farms through a geographic information system;

[0012] The radar system collects real-time information on the target vessel's position, speed, heading, and relative distance to the wind turbine.

[0013] Receive dynamic and static information of the target vessel through the Automatic Identification System (AIS);

[0014] Visual feature data of the target vessel is acquired using video acquisition equipment that includes a multispectral imager and a marine camera.

[0015] Wind speed, wind direction, and visibility information are obtained through meteorological monitoring stations or satellite data sources;

[0016] Location and range data, radar data, Automatic Identification System (AIS) data, video data, and meteorological data are transmitted to a central processing platform for preprocessing, including noise reduction, format standardization, and timestamp alignment, in order to complete data fusion.

[0017] Optionally, the fused multi-source sensing data undergoes spatiotemporal alignment and fusion processing to generate a bird's-eye view feature representation containing ship spatiotemporal feature information, including:

[0018] Visual features are extracted from video data to construct a semantic representation of the ship's appearance;

[0019] The semantic representation of the ship's appearance is then fused with data from the Automatic Identification System (AIS) and radar data.

[0020] The weight ratio of visual features, Automatic Identification System (AIS) data, and radar data in the fusion process is dynamically adjusted based on an attention mechanism.

[0021] By utilizing the spatial cross-attention mechanism, the weighted multi-source perception data are aligned in the spatiotemporal dimension and projected onto a unified bird's-eye view coordinate system to generate a bird's-eye view feature representation.

[0022] Optionally, based on the feature representation of the bird's-eye view, a fusion prediction model combining a ship motion physics model and a time series prediction algorithm is used to calculate the motion reachable set of the target ship within the prediction time period, including:

[0023] Establish a ship maneuvering motion model that characterizes the ship's motion characteristics in multiple degrees of freedom;

[0024] The feature representation of the bird's-eye view is input into a time series prediction network composed of a convolutional neural network, a bidirectional long short-term memory network, and an attention mechanism to extract the spatial correlation features and temporal dependence features of ship motion.

[0025] By using a ship maneuvering motion model to provide physical constraints for a time series prediction network, the future trajectory of a target ship can be predicted.

[0026] Based on the future trajectory, the target ship's current speed, and its maximum maneuverability, a motion reachability set is calculated and generated.

[0027] Optionally, a risk assessment is conducted based on the spatial relationship between the motion accessibility set and the preset wind farm safety zone, and a tiered early warning is triggered based on the assessment results, including:

[0028] Calculate the overlap area between the motion reachability set and the safe zone of the wind farm;

[0029] When the area of ​​the overlapping region is zero, it is determined that there is no risk of collision;

[0030] When the area of ​​the overlapping region is not zero, the risk level is quantified based on the ratio of the area of ​​the overlapping region to the area of ​​the reachable set. The risk level includes at least low risk, medium risk and high risk.

[0031] Based on the quantified risk level, corresponding early warning response measures are triggered. These measures include audible and visual alarms, radio warnings, and automatic activation of emergency plans. The emergency plans include providing navigable route suggestions to the target vessel via VHF radio.

[0032] Optionally, the method also includes:

[0033] If the overlapping area is irregular in shape, the ratio is calculated using an image grayscale segmentation algorithm.

[0034] Optionally, the ship maneuvering motion model is a mathematical model based on the ship's mass, moment of inertia, velocity and acceleration in the plane, and forces and torques acting on the ship, used to describe the ship's sway, pitch and roll motions in the plane.

[0035] The second aspect of this application provides a system for assessing and classifying navigation risks for offshore wind power vessels, including:

[0036] The multi-source sensing data acquisition and fusion module is used to collect and fuse multi-source sensing data from radar, automatic identification system for ships, video and meteorological monitoring equipment. The multi-source sensing data includes wind farm environmental information and target ship motion status information.

[0037] The spatiotemporal feature generation module is used to perform spatiotemporal alignment and fusion of multi-source sensing data to generate a bird's-eye view feature representation containing ship spatiotemporal feature information;

[0038] The trajectory prediction and reachability set calculation module is used to predict the trajectory of a target ship based on the feature representation of the bird's-eye view and combined with the ship motion physics model, and calculate the reachability set of the target ship within the prediction time period.

[0039] The risk assessment and early warning response module is used to calculate the overlapping area between the motion reachable set and the preset safety area of ​​the wind farm, classify the risk level according to the risk degree of the overlapping area, and trigger the corresponding graded early warning response.

[0040] The third aspect of this application provides a device for assessing and classifying navigation risks of offshore wind power vessels, including:

[0041] One or more processors;

[0042] A memory on which one or more programs are stored;

[0043] When the one or more programs are executed by the one or more processors, the one or more processors implement the offshore wind power vessel navigation risk assessment and classification early warning method as described in any of the above.

[0044] The fourth aspect of this application provides a computer storage medium for storing a program, which, when executed, is used to implement the method for assessing and classifying navigation risks of offshore wind power vessels as described in any of the preceding claims.

[0045] Compared with the prior art, the present invention has the following significant advantages:

[0046] (1) Improved the comprehensiveness and robustness of monitoring. By integrating multi-source sensing data such as radar, AIS, video (including multispectral) and meteorology, and using spatiotemporal attention mechanism for deep alignment and fusion, the blind spots and defects of single data sources are effectively compensated, and stable and accurate tracking and feature extraction of ships (including ships without AIS) are achieved in complex sea conditions (such as fog and night).

[0047] (2) Dynamic and accurate risk assessment was achieved. By combining ship maneuvering motion models with deep learning prediction algorithms, a "motion reachability set" based on the ship's physical motion limits was calculated, replacing the traditional fixed distance threshold. The dynamic risk quantification method based on the overlap area between the reachability set and the safe zone more realistically reflects the ship's collision probability and significantly reduces the false alarm rate.

[0048] (3) An intelligent and hierarchical early warning and response system has been established. Differentiated early warning measures are automatically triggered based on the quantified risk level, ranging from initial audible and visual warnings to proactive VHF radio warnings, and finally to automatically providing navigable route suggestions in high-risk situations. This system achieves an upgrade from passive alarm to proactive intervention, making the transmission of early warning information more timely and effective, and improving emergency response capabilities and overall regulatory efficiency. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 A flowchart illustrating a method for assessing and classifying navigation risks for offshore wind power vessels, provided as an embodiment of this application;

[0051] Figure 2 This application provides a schematic diagram of a multi-source sensing data processing process as an embodiment of the present application.

[0052] Figure 3 A schematic diagram of a bird's-eye view spatial mapping module based on spatiotemporal Transformer provided in this application embodiment;

[0053] Figure 4 A flowchart of a CNN-BiLSTM-Attention time series prediction algorithm provided in this application embodiment;

[0054] Figure 5 A schematic diagram of a graded early warning platform for ship conflicts in offshore wind power waters provided in this application embodiment;

[0055] Figure 6 A schematic diagram of a marine wind power vessel navigation risk assessment and classification early warning system provided in this application embodiment;

[0056] Figure 7 This is a schematic diagram of the structure of a device provided in an embodiment of this application. Detailed Implementation

[0057] This application provides a method and system for risk assessment and graded early warning of offshore wind power vessels, which can solve the problems of blind spots in monitoring from a single data source and the high false alarm rate of traditional early warning, thereby improving regulatory safety and efficiency.

[0058] See Figure 1The figure is a flowchart illustrating a method for assessing and classifying navigation risks for offshore wind power vessels according to an embodiment of this application. The method for assessing and classifying navigation risks for offshore wind power vessels provided in this embodiment can be implemented, for example, through the following steps S101-104.

[0059] S101: Collect and fuse multi-source sensing data.

[0060] In this embodiment, the multi-source sensing data includes at least offshore wind farm location and extent data, radar data, Automatic Identification System (AIS) data, video data, and meteorological data. The location and extent data of the offshore wind farm are acquired through a Geographic Information System (GIS); real-time location, speed, heading, and relative distance to the wind turbines of target vessels are collected through a radar system; dynamic and static information of target vessels is received through the AIS; visual feature data of target vessels is acquired through video acquisition equipment including a multispectral imager and a marine camera; wind speed, wind direction, and visibility information are obtained through meteorological monitoring stations or satellite data sources; the location and extent data, radar data, AIS data, video data, and meteorological data are transmitted to a central processing platform for preprocessing, including noise reduction, format unification, and timestamp alignment, to complete data fusion.

[0061] Specifically, such as Figure 2 As shown, Figure 2 This application provides a flowchart illustrating the acquisition, transmission, and preprocessing of multi-source sensing data. The sensing layer acquires radar data, AIS data, video data, and weather information from multiple sensors such as radar, cameras, multispectral imagers, and weather stations. The network layer transmits data via 5G, BeiDou satellite, and other methods. The data layer performs preprocessing operations such as noise reduction, format unification, and timestamp alignment, as well as data storage, providing a high-quality data foundation for subsequent fusion analysis. The presentation layer displays the reachability set and ship information obtained from the data layer after inputting into the early warning model and identification / positioning model on the platform.

[0062] Collect location and size data, radar data, AIS data, video data, and weather information for offshore wind farms. Location and size data includes the latitude and longitude coordinates of the wind farm boundaries and the radius of the wind turbine distribution.

[0063] Specifically, multi-source sensing data is collected as follows: Location and size data of offshore wind farms are obtained from a Geographic Information System (GIS) database, updated daily; radar data is collected through shore-based or shipboard radar systems. The radar system interprets the echoes from radar targets to obtain the ship's coordinates, speed, heading, and distance to the wind turbines, with a collection interval of every 30 seconds; AIS data includes the ship's dynamic information (position, speed, heading) and static information (ship length, type, draft, destination), received through AIS base stations, with an update interval of every 2-6 seconds to ensure real-time performance; visual... The frequency data is acquired through an improved multispectral imager and a marine camera. The multispectral imager uses the visible to near-infrared band (400-900nm) to capture the surface texture and thermal radiation characteristics of ships, while the marine camera acquires 1080P resolution video streams at a frame rate of 30fps and a monitoring distance of more than 5 kilometers. Weather data, including wind speed, wind direction, temperature, humidity, and visibility, is obtained from meteorological monitoring stations or satellite data sources and is updated every 5 minutes. All data is transmitted to the central processing platform via satellite network, terrestrial network, or direct intranet connection and undergoes data preprocessing (including noise reduction, format standardization, and timestamp alignment).

[0064] S102: Perform spatiotemporal alignment and fusion processing on the fused multi-source sensing data to generate a bird's-eye view feature representation containing spatiotemporal feature information of ships.

[0065] In this embodiment, visual features are extracted from video data to construct a semantic representation of the ship's appearance; the semantic representation of the ship's appearance is then fused with data from the Automatic Identification System (AIS) and radar data; the weight ratios of visual features, AIS data, and radar data during the fusion process are dynamically adjusted based on an attention mechanism; and the multi-source perception data, after weight adjustment, are aligned in the spatiotemporal dimension using a spatial cross-attention mechanism and projected onto a unified bird's-eye view coordinate system to generate a bird's-eye view feature representation.

[0066] Specifically, a bird's-eye view (BEV) spatial mapping module based on spatiotemporal Transformer is designed based on the obtained multimodal data. For example... Figure 3 As shown, Figure 3A schematic diagram of a bird's-eye view spatial mapping module based on spatiotemporal Transformer provided in this application embodiment illustrates the alignment and fusion process of multi-source sensing data in BEV space, to better solve the problem that a ship may be obscured by another ship when the camera cannot capture it, and also to provide more reference information for the real-time position and speed of the ship. That is, at the feature fusion level, a bird's-eye view (BEV) spatial mapping module based on spatiotemporal Transformer is designed, and a spatial cross-attention mechanism is used to realize the spatiotemporal alignment of multi-source sensing data. First, spatial feature alignment is performed based on pixel depth information, and multi-view images are projected onto a unified bird's-eye view coordinate system using query vectors. Combined with the spatial cross-attention module, the feature extraction capability of ship targets under complex sea conditions is enhanced, effectively capturing the ship motion modal features in scenarios such as surges and fog. In the multimodal data processing stage, pre-fusion and post-fusion are adopted, that is, first constructing the semantic representation of the ship appearance through visual features, and then performing post-fusion processing with the real-time position data of AIS and radar point cloud information. (b) Spatial Cross-Attention: Spatial Cross-Attention Mechanism

[0067] This invention enables BEV queries to extract the desired spatial features from multi-camera features through an attention mechanism. It employs a sparse attention mechanism based on Deformable Attention to allow the sum of each BEV query and a portion of the image region to interact.

[0068] Specifically, for each BEV feature located at (x, y), its corresponding real-world coordinates x', y' can be calculated. Then, a lift operation is performed on the BEV query to obtain multiple 3D points on the z-axis. With these 3D points, the projection points of these points onto the view plane can be obtained using camera intrinsic and extrinsic parameters. Due to limitations of camera parameters, each BEV query typically only has valid projection points in 1-2 views. Based on deformable attention, these projection points are used as reference points to sample features in the surrounding area. The BEV query is updated using weighted sampled features, thus completing feature aggregation in spatial space.

[0069] This invention treats BEV features as similar to memory capable of conveying sequence information. Each generated BEV feature obtains the necessary temporal information from the BEV features of the previous time step, ensuring dynamic acquisition of the required temporal features, unlike stacking BEV features from different time steps which only acquires a fixed length of temporal information. Specifically, given the BEV features of the previous time step, this invention first aligns the previous BEV features with the current time step based on ego motion, ensuring that features at the same location correspond to the same location in the real world.

[0070] For a BEV query located at (x, y) at the current time step, the object it represents may be static or dynamic. However, it is known that the object it represents will appear within a certain range around (x, y) at the previous time step. Therefore, deformable attention is used again to sample features with (x, y) as a reference point.

[0071] This invention does not explicitly design a forget gate, but instead uses attention wights in the attention mechanism to balance the fusion process of historical temporal features and current BEV features. Each BEV query aggregates spatial features in spatial space through spatial cross-attention, and also aggregates temporal features through temporal self-attention. This process is repeated multiple times to ensure that spatiotemporal features can promote each other and achieve more accurate feature fusion.

[0072] Further steps include the following:

[0073] In the multimodal data processing stage, pre-fusion and post-fusion are adopted. First, a semantic representation of the ship's appearance is constructed through visual features, and then it is fused with real-time position data from AIS and radar point cloud information. A specially designed adaptive weight fusion module establishes associations between multi-source perception data based on a self-attention mechanism. This module is used to construct an association matrix between visual features, AIS data, and radar data, and dynamically adjusts the weight ratio of each data source to achieve deep fusion of multi-source perception data.

[0074] At the feature fusion level, a bird's-eye view (BEV) spatial mapping module based on spatiotemporal Transformer is designed, which utilizes a spatial interactive attention mechanism to achieve spatiotemporal alignment of multi-source perception data. First, spatial feature alignment is performed based on pixel depth information. Then, query vectors are used to project multi-view images onto a unified bird's-eye view coordinate system. Combined with a spatial cross-attention module, the feature extraction capability of ship targets under complex sea conditions is enhanced, effectively capturing the modal features of ship motion in scenarios such as waves, fog, and haze.

[0075] S103: Based on the feature representation of the bird's-eye view, a fusion prediction model combining the ship motion physics model and the time series prediction algorithm is adopted to calculate the motion reach set of the target ship within the prediction time period.

[0076] In this embodiment, a ship maneuvering motion model is established to characterize the ship's motion characteristics in multiple degrees of freedom. The bird's-eye view feature representation is input into a time-series prediction network composed of a convolutional neural network, a bidirectional long short-term memory network, and an attention mechanism to extract the spatial correlation and temporal dependence features of the ship's motion. The ship maneuvering motion model provides physical constraints for the time-series prediction network to predict the future trajectory of the target ship. Based on the future trajectory, the target ship's current speed, and its ultimate maneuverability, a motion reachability set is calculated and generated. The ship maneuvering motion model is a mathematical model based on the ship's mass, moment of inertia, velocity and acceleration in the plane, and forces and moments acting on the ship, used to describe the ship's sway, pitch, and bow motions in the plane.

[0077] Specifically, based on the obtained bird's-eye view, the CNN-BiLSTM-Attention time series prediction algorithm is used to predict the real-time trajectory of ships, thereby obtaining the ship reachability set; such as Figure 4 As shown, Figure 4 This application provides a flowchart of a CNN-BiLSTM-Attention time series prediction algorithm, illustrating the algorithm structure for ship trajectory prediction based on the CNN-BiLSTM-Attention time series prediction algorithm: CNN extracts spatial features, BiLSTM captures forward and backward temporal dependencies, the Attention mechanism weights key information, and finally outputs the ship reachability set. Specifically, it includes the following operations:

[0078] First, a Maneuvering Model Group (MMG) model is established, and a physical representation of ship motion is constructed through refined parameter calibration. The model uses the time parameter matrix t and the rudder angle range matrix u. i Using the propeller speed matrix u0 as the input vector, based on fluid dynamics equations and rigid body kinematics principles, the motion characteristics of the ship in six degrees of freedom—swell, sway, heave, roll, pitch, and bow—are accurately characterized, providing physical-level dynamic constraints for trajectory prediction.

[0079] Then, the ship's motion reachability set is determined based on the ship's speed, limit rudder angle, and reachability set prediction time t. Spatial correlation features of ship motion parameters are extracted using multi-layer convolutional kernels of a convolutional neural network (CNN); a bidirectional long short-term memory network (BiLSTM) utilizes forward and backward hidden layer states to simultaneously mine historical and future tracks in the time series; and an attention mechanism weights ship features at different locations to avoid interference from redundant information.

[0080] The ship reachability set contains all the locations the ship will reach within the predicted time t, where t is a positive number.

[0081] The formula used in the MMG model is as follows:

[0082]

[0083] Where G represents the location of the ship's center of gravity. For ship quality, and These are the ship's accelerations along the x-axis and y-axis, respectively. For the ship's turning angular acceleration, and These represent the ship's speeds along the x-axis and y-axis, respectively. For the ship's turning angular velocity, , These are the lateral and longitudinal forces acting on the ship's center of gravity, respectively. The primary torque acting on the center of gravity of a ship.

[0084] Select a vessel on the nautical chart, and click on the historical trajectory in the information pop-up to display the vessel's historical trajectory on the chart. Simultaneously, you can define the vessel's prediction duration, and the system will automatically provide predictions of the vessel's movement trajectory for the next 2, 5, 10, and 20 minutes.

[0085] S104: Conduct risk assessment based on the spatial relationship between the motion reach set and the preset wind farm safety zone, and trigger graded early warning based on the assessment results.

[0086] In this embodiment, the overlapping area between the reachable set and the safe area of ​​the wind farm is calculated; when the area of ​​the overlapping area is zero, it is determined that there is no collision risk; when the area of ​​the overlapping area is not zero, the risk level is quantified according to the ratio of the area of ​​the overlapping area to the area of ​​the reachable set, and the risk level includes at least low risk, medium risk and high risk; the corresponding early warning response measures are triggered according to the quantified risk level, and the early warning response measures include audible and visual alarms, radio warnings and automatic activation of emergency plans, and the emergency plans include providing navigable route suggestions to the target vessel via VHF radio.

[0087] Specifically, a risk classification and early warning system for ship navigation conflicts is constructed based on the obtained ship reachability set; further, the risk classification for ship navigation conflicts includes:

[0088] The warning levels are divided into three levels: low risk (500 meters from the wind turbine, audible and visual alarm), medium risk (300 meters from the wind turbine, VHF radio warning), and high risk (100 meters from the wind turbine, automatic triggering of the emergency plan). The emergency plan refers to providing navigational route suggestions to ships within 100 meters of the wind turbine via VHF radio until the ships change course.

[0089] The following methods are used to construct a risk classification and early warning system for ship navigation conflicts based on reachability sets:

[0090] Calculate the overlap area between the safe zone of the wind farm and the reachable set of the ship. When the overlap area is zero, it is determined that the ship does not have a collision conflict with the wind farm at the current moment. When the overlap area is not zero, it is determined that the ship has a collision conflict with the wind farm at the current moment. The degree of collision conflict is represented by the ratio of the area of ​​the overlap area to the area of ​​the reachable set of the ship.

[0091] The degree of collision is calculated using the following formula:

[0092]

[0093] Where P represents the degree of collision. For the ship's position, This refers to the overlapping area between the ship-accessible area and the wind farm safety zone. Let be the reachable set function of the ship over a time period t. and These are the maximum rudder angles that can be used on the port and starboard sides of the vessel, respectively.

[0094] In one implementation of this application, if the overlapping area is irregular in shape, an image grayscale segmentation algorithm is used to calculate the ratio.

[0095] Specifically, when the overlapping region is irregularly shaped, the Otsu grayscale processing algorithm is used to calculate the area ratio, which is represented by the inter-class variance (ICV), as shown in the following formula:

[0096]

[0097] Where A and B represent the regions in the grayscale image where the grayscale value t is less than the preset grayscale threshold T and the regions where the grayscale value t is greater than the preset grayscale threshold T, respectively. and These represent the percentages of pixels contained in regions A and B within the total number of pixels in the entire image, respectively. and These are the average gray values ​​of regions A and B, respectively. The average of all gray values ​​in the entire image. The weights for regions A and B, .

[0098] In one implementation of this application, the weights of risk factors in wind farm waters are assigned based on an expert evaluation system and the mutual information method in information entropy theory. Combining the weights of the risk factors with the degree of collision conflict, an evidence-based reasoning method is used to construct a quantitative model of ship-machine collision risk under the influence of multiple factors, thereby achieving dynamic calculation of ship collision risk in wind farm waters. The risk factors include lookout negligence, high winds and waves, poor visibility, the deployment of navigation aids, and interference from small vessels.

[0099] The formula for calculating the weight of risk factors is as follows:

[0100]

[0101] in, Weights for risk factors and These represent the sample information and the total amount of information for the risk factors, respectively. Let be the probability function of the risk factors;

[0102] The calculation formula for the evidence reasoning method is as follows:

[0103]

[0104] ;

[0105]

[0106] ;

[0107]

[0108] Where L is the number of risk factors, Let be the expert evaluation value of the k-th risk factor. whether), The weight of the k-th risk factor. For basic probability weights, For uncertain probability weights, To integrate the results of the impact of the k-th risk factor and the (k+1)-th risk factor on the ship-machine collision risk, Let k be the normalized state distribution of ship-machine collision risk, where k is a positive integer.

[0109] In one implementation of this application, the risk assessment object and results are displayed in text form on the main interface of the risk quantification and early warning module. Real-time early warning or alarm information can be displayed on the main page in real time, and various effects such as sound and light can be used to provide prompts on the nautical chart screen.

[0110] Specifically, such as Figure 5 As shown, Figure 5 This application provides a schematic diagram of the framework of a graded early warning platform for ship conflicts in offshore wind power waters, illustrating the overall framework of the platform.

[0111] First, the system's cloud-based global task scheduling module distributes and describes tasks, which are mainly divided into three tasks shown in the image.

[0112] The communication unit is responsible for mission one, collecting information from various heterogeneous data sources, including AIS, radar, multispectral imagers and marine cameras that capture visual information, and meteorological monitoring stations that provide environmental data;

[0113] The system edge is responsible for tasks two and three, performing data fusion analysis, preprocessing all raw data, predicting the future trajectory of ships, and calculating their motion reachability set. Based on the ship reachability set, risk quantification is performed, dynamically assessing the degree of collision risk by calculating the overlap area between the ship reachability set and the safe area of ​​the wind farm. Based on the size and location of the overlap area, the risk is divided into three distinct levels: low risk, medium risk, and high risk, with corresponding early warning trigger thresholds and preliminary response measures set for each level.

[0114] The terminal serves as the interface between the system and the user. The results of the risk assessment are transformed into intuitive visual information displayed on the platform's main interface, including vessel location, risk level, and warning information, supplemented by diverse audio and visual prompts. Furthermore, when a high-risk warning is triggered, the system notifies monitoring personnel to intervene via VHF radio or other means.

[0115] Upon receiving a user's instruction to click the early warning information module, the system provides the user with calculation results of dynamic risk quantification for vessels, including but not limited to risk quantification between vessels and between vessels and fixed objects. Based on the vessel's motion status, an reachability set for the vessel is constructed and visualized on the nautical chart. Risk assessments are conducted on vessels entering the early warning area based on the risk quantification results, providing three levels of risk warnings and responses. Furthermore, the assessment objects and results are displayed in text form on the main interface of the risk quantification and early warning module. Real-time warnings or alarms can be displayed on the main page in real time, and accompanied by various audio and visual effects on the nautical chart screen. Upon receiving a user's instruction to click the emergency management module, when an unavoidable collision or other accident that threatens the wind turbine is about to occur or has already occurred, the system will activate the emergency plan and handle it according to the risk accident response procedure. Simultaneously, the emergency response system supports automatic or manual plan queries and provides relevant personnel with emergency alarms, emergency resource allocation and contact services with maritime rescue teams. Personnel only need to click the corresponding button for the desired service and enter the administrator password to complete the corresponding operation.

[0116] This invention integrates radar, AIS, multispectral imager, and video data, utilizing the BEVFormer model for feature-level fusion to achieve precise positioning and dynamic tracking of vessels. For example, in a pilot project at the Pinghai Bay wind farm in Fujian, the system successfully identified and tracked small fishing boats not equipped with AIS, reducing the positioning error to within 5 meters, an improvement of over 60% in accuracy compared to traditional systems. This effectively solves the problems of false echoes and detection blind spots caused by wind turbine obstruction, providing a high-quality data foundation for risk assessment.

[0117] The proposed hierarchical early warning mechanism, based on ship reachability set calculation, achieves multi-gradient real-time early warning, significantly reducing the false alarm rate. Traditional electronic fence technology can only trigger alarms based on a single threshold, resulting in a high false alarm rate and an inability to differentiate risk levels. This invention predicts ship trajectories using a CNN-BiLSTM-Attention model and simultaneously divides warnings into low, medium, and high levels: the lowest level warning is triggered when a ship enters the monitoring range; the highest level warning is triggered when a ship approaches a wind turbine, based on its obstacle avoidance capabilities. In tests at the Xinghua Bay Wind Farm, the false alarm rate was reduced from 30% in traditional systems to below 5%, and the early warning response time was shortened to the second level, improving regulatory efficiency and safety. The system integrates audible and visual alarms, VHF communication equipment, and automatic message transmission functions to ensure timely delivery of early warning information to ships and the monitoring center.

[0118] This invention enables all-weather, all-time monitoring, overcoming technical limitations under adverse weather conditions. Through the complementary advantages of multispectral imagers and radar, the system can operate stably even under extreme conditions such as nighttime, heavy fog, and torrential rain. The multispectral imager uses the visible to near-infrared bands to capture ship characteristics, while the radar provides all-weather monitoring capabilities, and combined with weather data, ensures continuous monitoring.

[0119] Based on the methods provided in the above embodiments, this application also provides an offshore wind power vessel navigation risk assessment and classification early warning system. The offshore wind power vessel navigation risk assessment and classification early warning system is described below with reference to the accompanying drawings.

[0120] See Figure 6 The figure is a schematic diagram of the structure of a marine wind power vessel navigation risk assessment and classification early warning system provided in an embodiment of this application.

[0121] The offshore wind power vessel navigation risk assessment and classification early warning system 600 provided in this application includes: a multi-source data acquisition and fusion module 601, a spatiotemporal feature generation module 602, a trajectory prediction and reachability set calculation module 603, and a risk assessment and early warning response module 604.

[0122] The multi-source data acquisition and fusion module 601 is used to acquire and fuse multi-source sensing data from radar, automatic identification system for ships, video and meteorological monitoring equipment. The multi-source sensing data includes wind farm environmental information and target ship motion status information.

[0123] The spatiotemporal feature generation module 602 is used to perform spatiotemporal alignment and fusion of multi-source sensing data to generate a bird's-eye view feature representation containing ship spatiotemporal feature information.

[0124] The trajectory prediction and reachability set calculation module 603 is used to predict the trajectory of a target ship based on the feature representation of the bird's-eye view and combined with the ship motion physical model, and calculate the reachability set of the target ship within the prediction time period.

[0125] The risk assessment and early warning response module 604 is used to calculate the overlapping area between the motion reachable set and the preset safety area of ​​the wind farm, classify the risk level according to the risk degree of the overlapping area, and trigger the corresponding graded early warning response.

[0126] In one possible implementation, the multi-source data acquisition and fusion module 601 is specifically used for:

[0127] Obtain location and extent data of offshore wind farms through a geographic information system;

[0128] The radar system collects real-time information on the target vessel's position, speed, heading, and relative distance to the wind turbine.

[0129] Receive dynamic and static information of the target vessel through the Automatic Identification System (AIS);

[0130] Visual feature data of the target vessel is acquired using video acquisition equipment that includes a multispectral imager and a marine camera.

[0131] Wind speed, wind direction, and visibility information are obtained through meteorological monitoring stations or satellite data sources;

[0132] Location and range data, radar data, Automatic Identification System (AIS) data, video data, and meteorological data are transmitted to a central processing platform for preprocessing, including noise reduction, format standardization, and timestamp alignment, in order to complete data fusion.

[0133] In one possible implementation, the spatiotemporal feature generation module 602 is specifically used for:

[0134] Visual features are extracted from video data to construct a semantic representation of the ship's appearance;

[0135] The semantic representation of the ship's appearance is then fused with data from the Automatic Identification System (AIS) and radar data.

[0136] The weight ratio of visual features, Automatic Identification System (AIS) data, and radar data in the fusion process is dynamically adjusted based on an attention mechanism.

[0137] By utilizing the spatial cross-attention mechanism, the weighted multi-source perception data are aligned in the spatiotemporal dimension and projected onto a unified bird's-eye view coordinate system to generate a bird's-eye view feature representation.

[0138] In one possible implementation, the trajectory prediction and reachability set calculation module 603 has functions for:

[0139] Establish a ship maneuvering motion model that characterizes the ship's motion characteristics in multiple degrees of freedom;

[0140] The feature representation of the bird's-eye view is input into a time series prediction network composed of a convolutional neural network, a bidirectional long short-term memory network, and an attention mechanism to extract the spatial correlation features and temporal dependence features of ship motion.

[0141] By using a ship maneuvering motion model to provide physical constraints for a time series prediction network, the future trajectory of a target ship can be predicted.

[0142] Based on the future trajectory, the target ship's current speed, and its maximum maneuverability, a motion reachability set is calculated and generated.

[0143] In one possible implementation, the risk assessment and early warning response module 604 has the following functions:

[0144] Calculate the overlap area between the motion reachability set and the safe zone of the wind farm;

[0145] When the area of ​​the overlapping region is zero, it is determined that there is no risk of collision;

[0146] When the area of ​​the overlapping region is not zero, the risk level is quantified based on the ratio of the area of ​​the overlapping region to the area of ​​the reachable set. The risk level includes at least low risk, medium risk and high risk.

[0147] Based on the quantified risk level, corresponding early warning response measures are triggered. These measures include audible and visual alarms, radio warnings, and automatic activation of emergency plans. The emergency plans include providing navigable route suggestions to the target vessel via VHF radio.

[0148] In one possible implementation, the risk assessment and early warning response module 604 is also used for:

[0149] If the overlapping area is irregular in shape, the ratio is calculated using an image grayscale segmentation algorithm.

[0150] In one possible implementation, the ship maneuvering motion model is a mathematical model based on the ship's mass, moment of inertia, velocity and acceleration in the plane, and forces and torques acting on the ship, used to describe the ship's sway, pitch, and bow motions in the plane.

[0151] Since the offshore wind power vessel navigation risk assessment and classification early warning system 600 is a system corresponding to the offshore wind power vessel navigation risk assessment and classification early warning method provided in the above method embodiments, the specific implementation of each module of the offshore wind power vessel navigation risk assessment and classification early warning system 600 is based on the same concept as in the above method embodiments. Therefore, for the specific implementation of each module of the offshore wind power vessel navigation risk assessment and classification early warning system 600, please refer to the description of the offshore wind power vessel navigation risk assessment and classification early warning method in the above method embodiments, and it will not be repeated here.

[0152] This application embodiment also provides an offshore wind power vessel navigation risk assessment and classification early warning device, the device including: a processor and a memory;

[0153] The memory is used to store instructions;

[0154] The processor is used to execute the instructions in the memory to perform the offshore wind power vessel navigation risk assessment and graded early warning method mentioned in the above embodiments.

[0155] It should be noted that the hardware structure of the offshore wind power vessel navigation risk assessment and classification early warning equipment provided in this application embodiment can be as follows: Figure 7 The structure shown, Figure 7 This is a schematic diagram of the structure of a device provided in an embodiment of this application.

[0156] Please see Figure 7 As shown, device 700 includes: a processor 710, a communication interface 720, and a memory 730. The number of processors 710 in device 700 can be one or more. Figure 7 Taking a processor as an example, in this embodiment, the processor 710, communication interface 720, and memory 730 can be connected via a bus system or other means. Figure 7 Taking the connection between China and Israel via the 740 bus system as an example.

[0157] Processor 710 may be a central processing unit (CPU), a network processor (NP), or a combination of a CPU and an NP. Processor 710 may further include hardware chips. These hardware chips may be application-specific integrated circuits (ASICs), programmable logic devices (PLDs), or combinations thereof. The PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), generic array logic (GAL), or any combination thereof.

[0158] The memory 730 may include volatile memory, such as random-access memory (RAM); the memory 730 may also include non-volatile memory, such as flash memory, hard disk drive (HDD) or solid-state drive (SSD); the memory 730 may also include a combination of the above types of memory.

[0159] Optionally, the memory 730 stores an operating system and programs, executable modules, or data structures, or subsets thereof, or extended sets thereof. The programs may include various operation instructions for implementing various operations. The operating system may include various system programs for implementing various basic services and handling hardware-based tasks. The processor 710 can read the programs in the memory 730 to implement the offshore wind power vessel navigation risk assessment and classification early warning method provided in this application embodiment.

[0160] The bus system 740 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus system 740 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 7 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0161] This application also provides a computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the offshore wind power vessel navigation risk assessment and classification early warning method mentioned in the above embodiments.

[0162] This application also provides a computer program product containing instructions that, when run on a computer, causes the computer to execute the offshore wind power vessel navigation risk assessment and classification early warning method mentioned in the above embodiments.

[0163] Although the invention has been specifically shown and described in conjunction with preferred embodiments, those skilled in the art should understand that various changes in form and detail may be made to the invention without departing from the spirit and scope of the invention as defined in the appended claims, all of which shall be within the scope of protection of the invention.

Claims

1. A method for assessing and classifying navigation risks of offshore wind power vessels, characterized in that, The method includes: Collect and fuse multi-source sensing data, which includes at least offshore wind farm location range data, radar data, automatic identification system data, video data, and meteorological data; The fused multi-source sensing data undergoes spatiotemporal alignment and fusion processing to generate a bird's-eye view feature representation containing ship spatiotemporal characteristic information. Specifically, this includes: extracting visual features from the video data to construct a semantic representation of the ship's appearance; performing post-fusion processing on the semantic representation of the ship's appearance with the ship automatic identification system data and the radar data; dynamically adjusting the weight ratios of the visual features, the ship automatic identification system data, and the radar data in the fusion process based on an attention mechanism; and aligning the weighted multi-source sensing data in the spatiotemporal dimension using a spatial cross-attention mechanism and projecting it onto a unified bird's-eye view coordinate system to generate the bird's-eye view feature representation. Specifically, the spatial cross-attention mechanism is used to align multi-source sensing data with adjusted weights in the spatiotemporal dimension and project them onto a unified bird's-eye view coordinate system to generate the bird's-eye view feature representation. This includes: performing spatial feature alignment based on pixel depth information; projecting multi-view images onto a unified bird's-eye view coordinate system using query vectors; enhancing the feature extraction capability of ship targets under complex sea conditions by combining the spatial cross-attention module; and using a temporal self-attention mechanism to dynamically obtain temporal information from the bird's-eye view features aligned at the previous time step, so as to balance the fusion of historical temporal features and current bird's-eye view features. Based on the bird's-eye view feature representation, a fusion prediction model combining a ship motion physics model and a time series prediction algorithm is used to calculate the motion reach set of the target ship within the prediction time period; Risk assessment is performed based on the spatial relationship between the motion reachable set and the preset wind farm safety area, and graded early warnings are triggered based on the assessment results.

2. The method according to claim 1, characterized in that, The collection and fusion of multi-source sensing data includes: The location and extent data of the offshore wind farm were obtained through a geographic information system. The radar system collects real-time information on the target vessel's position, speed, heading, and relative distance to the wind turbine. The system receives dynamic and static information about the target vessel through an Automatic Identification System (AIS). Visual feature data of the target vessel are acquired using video acquisition equipment that includes a multispectral imager and a marine camera; Wind speed, wind direction, and visibility information are obtained through meteorological monitoring stations or satellite data sources; The location and range data, radar data, automatic identification system data, video data, and meteorological data are transmitted to the central processing platform for preprocessing such as noise reduction, format unification, and timestamp alignment to complete data fusion.

3. The method according to claim 1, characterized in that, Based on the bird's-eye view feature representation, a fusion prediction model combining a ship motion physics model and a time series prediction algorithm is used to calculate the motion reachable set of the target ship within the prediction time period, including: Establish a ship maneuvering motion model that characterizes the ship's motion characteristics in multiple degrees of freedom; The bird's-eye view feature representation is input into a time series prediction network composed of a convolutional neural network, a bidirectional long short-term memory network, and an attention mechanism to extract the spatial correlation features and temporal dependence features of ship motion. The ship maneuvering motion model is used to provide physical constraints for the time series prediction network to predict the future trajectory of the target ship. Based on the future trajectory, the target vessel's current speed, and its maximum maneuverability, the motion reachability set is calculated and generated.

4. The method according to claim 1, characterized in that, The risk assessment based on the spatial relationship between the motion reachability set and the preset wind farm safety zone, and the triggering of graded early warnings based on the assessment results, includes: Calculate the overlap area between the motion reachable set and the safe zone of the wind farm; When the area of ​​the overlapping region is zero, it is determined that there is no risk of collision. When the area of ​​the overlapping region is not zero, the risk level is quantified based on the ratio of the area of ​​the overlapping region to the area of ​​the motion reachable set, and the risk level includes at least low risk, medium risk and high risk; Based on the quantified risk level, corresponding early warning response measures are triggered. These measures include audible and visual alarms, radio warnings, and automatic activation of emergency plans. The emergency plans include providing navigable route suggestions to the target vessel via VHF radio.

5. The method according to claim 4, characterized in that, The method further includes: If the overlapping region is irregular in shape, the ratio is calculated using an image grayscale segmentation algorithm.

6. The method according to claim 3, characterized in that, The ship maneuvering motion model is a mathematical model based on the ship's mass, moment of inertia, velocity and acceleration in the plane, and forces and torques acting on the ship. It is used to describe the ship's sway, pitch, and bow motions in the plane.

7. A risk assessment and grading early warning system for offshore wind power vessels, characterized in that, The system includes: The multi-source sensing data acquisition and fusion module is used to acquire and fuse multi-source sensing data from radar, automatic identification system for ships, video and meteorological monitoring equipment. The multi-source sensing data includes wind farm environmental information and target ship motion status information. The spatiotemporal feature generation module is used to perform spatiotemporal alignment and fusion of the multi-source sensing data to generate a bird's-eye view feature representation containing ship spatiotemporal feature information; The spatiotemporal feature generation module is specifically used to extract visual features from the video data and construct a semantic representation of the ship's appearance; perform post-fusion processing on the semantic representation of the ship's appearance with the data from the automatic ship identification system and the radar data; dynamically adjust the weight ratio of the visual features, the data from the automatic ship identification system, and the radar data in the fusion process based on an attention mechanism; and use a spatial cross-attention mechanism to align the weighted multi-source perception data in the spatiotemporal dimension and project it onto a unified bird's-eye view coordinate system to generate the bird's-eye view feature representation. Specifically, the spatial cross-attention mechanism is used to align multi-source sensing data with adjusted weights in the spatiotemporal dimension and project them onto a unified bird's-eye view coordinate system to generate the bird's-eye view feature representation. This includes: performing spatial feature alignment based on pixel depth information; projecting multi-view images onto a unified bird's-eye view coordinate system using query vectors; enhancing the feature extraction capability of ship targets under complex sea conditions by combining the spatial cross-attention module; and using a temporal self-attention mechanism to dynamically obtain temporal information from the bird's-eye view features aligned at the previous time step, so as to balance the fusion of historical temporal features and current bird's-eye view features. The trajectory prediction and reachability set calculation module is used to predict the trajectory of the target ship based on the bird's-eye view feature representation and combined with the ship motion physics model, and calculate the reachability set of the target ship within the prediction time period; The risk assessment and early warning response module is used to calculate the overlapping area between the motion reachable set and the preset safety area of ​​the wind farm, classify the risk level according to the risk degree of the overlapping area, and trigger the corresponding graded early warning response.

8. An electronic device, characterized in that, The device includes: a processor and a memory; The memory is used to store instructions; The processor is configured to execute the instructions in the memory to perform the method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, Including instructions that, when run on a computer, cause the computer to perform the method described in any one of claims 1-6 above.

Images