A method and system for remote monitoring and management of offshore wind farms

Abstract

The application discloses a kind of offshore wind farm remote monitoring management method and system, it is related to wind farm supervision field, which comprises the following steps: based on the seabed topography information of target offshore wind farm, determine a plurality of target detection points;Control unmanned underwater vehicle to reach each target detection point, obtain the corresponding underwater foundation image information of each target detection point;Based on the corresponding underwater foundation image information of each target detection point, determine the corresponding underwater foundation fault comprehensive score of wind turbine of each target detection point;Based on the corresponding underwater foundation fault comprehensive score of wind turbine of each target detection point, determine the target offshore wind farm's underwater foundation health status distribution map.

Description

TECHNICAL FIELD

[0001] The present specification relates to the field of wind farm supervision, and more particularly, a method and system for remote monitoring and management of offshore wind farms. BACKGROUND

[0002] With the increasing demand for clean energy worldwide, offshore wind farms, as an important way to obtain renewable energy, have been widely used. However, offshore wind farms are located in the marine environment, and factors such as seawater corrosion, wave impact, and seabed subsidence can affect the foundation structure of the wind turbine. Long-term exposure to the marine environment may cause corrosion, cracking, loosening, and subsidence of the foundation structure, leading to a decrease in safety and stability, and in severe cases, even causing the wind turbine to collapse. Therefore, health monitoring and evaluation of offshore wind turbine foundation structures have become critical to ensure their stable operation.

[0003] Traditional offshore wind power foundation monitoring methods usually rely on manual inspection or fixed monitoring equipment. These methods have the following main shortcomings:

[0004] Manual inspection: manual divers or remote-controlled equipment inspection is time-consuming and labor-intensive, with high costs and high risks, especially in adverse sea conditions.

[0005] Fixed monitoring equipment: fixed sensors installed on the wind turbine foundation may fail due to marine corrosion or equipment aging, making it difficult to fully cover all monitoring areas and respond to complex seabed changes.

[0006] Data limitations: traditional methods usually cannot obtain complete images and multi-angle information of the underwater foundation structure in real time and in detail, limiting accurate assessment of the health status of the wind turbine foundation. SUMMARY

[0007] In the summary section, a series of simplified concepts are introduced, which will be further described in detail in the specific embodiments section. The summary section of the present application does not mean to attempt to limit the key features and necessary technical features of the claimed technical solutions, nor does it attempt to determine the protection scope of the claimed technical solutions.

[0008] In a first aspect, the present application proposes a method for remote monitoring and management of offshore wind farms, comprising:

[0009] Based on the seabed topographic information of the target offshore wind farm, a plurality of target detection points are determined;

[0010] Control the unmanned underwater explorer to reach each of the above target detection points to obtain the corresponding underwater foundation image information of each of the above target detection points;

[0011] Determine the subsea foundation fault comprehensive score of the wind turbine corresponding to each target detection point based on the subsea foundation image information corresponding to each target detection point.

[0012] Determine the subsea foundation health state distribution map of the target offshore wind farm based on the subsea foundation fault comprehensive score of the wind turbine corresponding to each target detection point.

[0013] In a feasible implementation, the determination of the plurality of target detection points based on the seabed topographic information of the target offshore wind farm comprises:

[0014] According to the seabed topographic information, perform analysis and classification operations based on a topographic convolutional neural network model to extract a topographic feature vector of each detection point, wherein the topographic feature vector comprises slope information, geological density information, and sediment loose degree information; and the seabed topographic information is generated based on the detection data;

[0015] According to the topographic feature vector, perform clustering analysis based on a DBSCAN method to obtain a topographic clustering analysis result, wherein the topographic clustering analysis result comprises detection point position information and clustering label information;

[0016] Perform risk degree calculation based on the topographic clustering analysis result and the topographic feature vector, and determine the preliminary detection points according to the risk degree;

[0017] According to the spatial coverage information and the resource limitation information, perform distribution optimization operations on the preliminary detection points by using a greedy algorithm to obtain the target detection points.

[0018] In a feasible implementation, the risk degree F is calculated based on the following formula:

[0019]

[0020] wherein S is the slope information, D is the geological density information, L is the sediment loose degree information, R is other environmental factor information, w1 is a first weight coefficient corresponding to the slope information, w2 is a second weight coefficient corresponding to the geological density information, w3 is a third weight coefficient corresponding to the sediment loose degree information, a is a first adjustment parameter, β is a second adjustment parameter, γ is a third adjustment parameter, δ is a fourth adjustment parameter, η is a fifth adjustment parameter, f(C) is a clustering label function, g(d) is a position distance information function, κ is a sixth adjustment parameter, λ is a seventh adjustment parameter, C represents the clustering label information, and d represents the detection point position information.

[0021] In an embodiment, the distribution optimization of the preliminary detection points based on the spatial coverage information and the resource constraint information is performed by a greedy algorithm to obtain the target detection points, including:

[0022] Performing initial parameter setting of the greedy algorithm based on the preliminary detection points, the spatial coverage information, and the resource constraint information to obtain an initialized parameter list;

[0023] According to the monitoring radius and the position information corresponding to the preliminary detection points, the detection point coverage range information corresponding to each preliminary detection point is calculated;

[0024] Based on all the detection point coverage range information and the current uncovered area, the spatial coverage contribution value corresponding to each preliminary detection point is calculated;

[0025] According to the spatial coverage contribution value corresponding to all preliminary detection points and the resource constraint information, a greedy selection operation is performed to obtain an updated uncovered area and remaining preliminary detection points;

[0026] Based on the updated uncovered area and the remaining preliminary detection points, spatial coverage detection and resource constraint detection are performed to determine whether the termination condition is met;

[0027] If the termination condition is not met, the uncovered area is updated, the selected detection points are removed, and the coverage contribution value is recalculated, and the greedy selection operation is repeated;

[0028] If the termination condition is met, the current remaining preliminary detection points are determined as the target detection points.

[0029] In an embodiment, the control of the unmanned underwater explorer to reach each target detection point to obtain the underwater basic image information corresponding to each target detection point includes:

[0030] Based on the position information, environmental information, and seabed obstacle information corresponding to each target detection point, a target path is obtained based on a PPO algorithm;

[0031] Based on the target path, the unmanned underwater explorer is controlled to reach each target detection point;

[0032] When the unmanned underwater explorer reaches the target detection point, the base type information of the wind turbine corresponding to the current target detection point is determined;

[0033] Based on the base type information, target shooting position information is determined;

[0034] Based on the target shooting position information, the underwater basic image information is shot.

[0035] In an embodiment, the subsea foundation fault comprehensive score of each target detection point is determined based on the subsea foundation image information corresponding to each target detection point, including:

[0036] The fault weight coefficients corresponding to the subsea foundation images at different positions are determined according to the base type information;

[0037] The subsea foundation fault scores at different positions are determined based on the subsea foundation images at different positions and the fault detection model;

[0038] The subsea foundation fault comprehensive score is determined according to the base type, the fault weight coefficients corresponding to the subsea foundation images at different positions, and the subsea foundation fault scores at different positions.

[0039] In an embodiment, when the base type information is a single pile foundation, the subsea foundation fault comprehensive score E1 is determined based on the following formula:

[0040]

[0041] In the formula, F1 is the fault score of the outer wall of the pile body, F2 is the fault score of the connection between the pile bottom and the seabed, F3 is the fault score of the intertidal zone, W4 is the weight coefficient corresponding to the fault score of the outer wall of the pile body, W5 is the weight coefficient corresponding to the fault score of the connection between the pile bottom and the seabed, W6 is the weight coefficient corresponding to the fault score of the intertidal zone, and ε, θ, ξ, and τ are adjustment parameters.

[0042] 8. The offshore wind farm remote monitoring and management method according to claim 1, characterized in that when the base type information is a gravity foundation, the subsea foundation fault comprehensive score E2 is determined based on the following formula:

[0043]

[0044] In the formula, F4 is the fault score of the base edge, F5 is the fault score of the tower connection, F6 is the fault score of the surrounding seabed, W7 is the weight coefficient corresponding to the fault score of the base edge, W8 is the weight coefficient corresponding to the fault score of the tower connection, W9 is the weight coefficient corresponding to the fault score of the surrounding seabed, and A, B, C, M, and N are adjustment parameters.

[0045] In an embodiment, the subsea foundation health status distribution map of the target offshore wind farm is determined based on the subsea foundation fault comprehensive score of each target detection point, including:

[0046] mapping the underwater foundation fault comprehensive score of the wind turbine corresponding to each of the target detection points to a geographical space diagram of the target area to construct a foundation health state scatter plot distribution diagram;

[0047] performing interpolation operation and smoothing operation based on the foundation health state scatter plot distribution diagram to obtain the underwater foundation health state distribution diagram.

[0048] In a second aspect, the application discloses a remote monitoring and management system for offshore wind farms, comprising:

[0049] a first determination unit configured to determine a plurality of target detection points based on seabed topographic information of a target offshore wind farm;

[0050] an acquisition unit configured to control an unmanned underwater probe to reach each of the target detection points and acquire underwater foundation image information corresponding to each of the target detection points;

[0051] a second determination unit configured to determine an underwater foundation fault comprehensive score of a wind turbine corresponding to each of the target detection points based on the underwater foundation image information corresponding to each of the target detection points;

[0052] a third determination unit configured to determine an underwater foundation health state distribution diagram of the target offshore wind farm based on the underwater foundation fault comprehensive score of the wind turbine corresponding to each of the target detection points.

[0053] In summary, the present application intelligently selects multiple high-risk target detection points by analyzing the seabed topography information of the target offshore wind farm, so that the monitoring range covers the key areas of the seabed topography features. Compared with the limitations of fixed monitoring equipment, the detection points of the present application can be flexibly adjusted to ensure that the high-risk areas of the seabed are fully covered. The present application uses an unmanned underwater probe to generate a navigation path according to the location of the target detection point and environmental factors, and realizes automatic navigation to each detection point and collection of underwater basic image information. This method avoids the safety risks of manual diving detection and can also stably perform monitoring tasks when the sea conditions are complex, greatly improving the detection efficiency and safety. The present application uses underwater cameras, sonars and other sensing devices to obtain underwater basic image information of each target detection point, and calculates an underwater basic fault comprehensive score based on a fault detection model. This image information-based evaluation method can capture various fault information such as corrosion, cracks and settlement in real time, and has higher accuracy and comprehensiveness than single sensor monitoring, providing accurate evaluation results for the health status of each detection point. The present application draws a health status distribution map of the underwater basic fault comprehensive score of each detection point and performs smoothing and interpolation processing, making the monitoring results more intuitive. The distribution map can show the health status of all wind turbine foundations in the wind farm, helping operation and maintenance personnel quickly identify potential risk areas. Compared with the separate monitoring method in the traditional method, the health status distribution map provides an overall perspective, which is convenient for developing a comprehensive maintenance plan. Based on the health status distribution map provided by the present application, operation and maintenance personnel can identify high-risk detection points in advance to realize early warning of faults and reduce the risk of wind turbine foundation structure failure. In addition, the flexible inspection method of the unmanned underwater probe can optimize the detection frequency, reduce unnecessary manual inspection and maintenance work, and reduce the overall maintenance cost. In summary, the present application solves the problems of coverage limitations, low efficiency and inaccurate evaluation in the traditional monitoring method through intelligent inspection and image information analysis of the unmanned underwater probe, combined with fault comprehensive score and health status distribution map, and provides an efficient, accurate and intuitive offshore wind farm foundation health monitoring method, which provides a reliable guarantee for the long-term stable operation of the wind farm.

[0054] The offshore wind farm remote monitoring management method of the present application, other advantages, objectives and features of the present application will be embodied in part through the following description, and will be understood by those skilled in the art through the study and practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0055] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are for purposes of illustration only and are not considered a limitation of the present specification. Moreover, like reference numerals are used to designate like parts throughout the drawings. In the drawings:

[0056] Figure 1 This application provides a flowchart illustrating a remote monitoring and management method for offshore wind farms.

[0057] Figure 2 This application provides a data mining-based inspection system for new energy power stations. Detailed Implementation

[0058] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus. The technical solutions of the embodiments of this application will now be clearly and completely described in conjunction with the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.

[0059] Please see Figure 1 The above is a flowchart illustrating a remote monitoring and management method for offshore wind farms provided in this application, including:

[0060] S110. Based on the seabed topographic information of the target offshore wind farm, determine multiple target detection points.

[0061] For example, multiple target detection points are determined based on the seabed topography information of the target offshore wind farm. By collecting and analyzing the seabed topography data of the wind farm, high-risk locations in different topographic feature areas can be identified, such as areas with steep slopes or areas with loose sediments. Based on this topographic information, appropriate algorithms, such as clustering or intelligent filtering algorithms, are used to identify and filter these areas, ultimately resulting in multiple target detection points suitable for monitoring.

[0062] S120: Control the unmanned underwater vehicle to reach each of the above-mentioned target detection points and acquire the corresponding underwater basic image information of each target detection point.

[0063] Exemplarily, after determining the target detection points, the unmanned underwater probe is controlled to reach each of the above target detection points. In this process, the unmanned underwater probe plans a path and navigates according to the location of the target detection point and environmental factors (such as ocean currents, sea currents, etc.) to smoothly reach the designated detection point. At each target detection point, the underwater probe uses underwater cameras, sonar, and other sensing devices to obtain corresponding underwater foundation image information. This image information contains specific information about the wind turbine foundation, providing data support for subsequent fault assessment.

[0064] S130, based on the above-mentioned underwater foundation image information corresponding to each target detection point, determining the underwater foundation fault comprehensive score of the wind turbine corresponding to each target detection point.

[0065] Exemplarily, after obtaining the underwater foundation image information, based on the image information, the underwater foundation fault comprehensive score of the wind turbine corresponding to each target detection point is determined. Through image analysis technology, specific fault features in the image are identified, such as cracks, corrosion, settlement, etc. Different fault features are applied to the corresponding fault detection model, combined with fault weight coefficients to calculate the score of each location, and finally the underwater foundation fault comprehensive score is generated. The score is used to reflect the health status of the wind turbine foundation corresponding to each target detection point.

[0066] S140, based on the underwater foundation fault comprehensive score of the wind turbine corresponding to each of the above target detection points, determining the underwater foundation health status distribution map of the target offshore wind farm.

[0067] Exemplarily, after obtaining the underwater foundation fault comprehensive score of each target detection point, the underwater foundation health status distribution map of the target offshore wind farm is determined according to the scores. The fault comprehensive score of each detection point is mapped into a spatial distribution map, and a continuous health status distribution map is generated through smoothing and interpolation processing. This distribution map can intuitively display the health status of different detection points (such as healthy, sub-healthy, fault warning, serious fault), thereby helping operation and maintenance personnel to timely discover high-risk areas and potential fault trends, providing important reference for subsequent maintenance and monitoring.

[0068] In summary, the present application intelligently selects multiple high-risk target detection points by analyzing the seabed topography information of the target offshore wind farm, so that the monitoring range covers the key areas of the seabed topography features. Compared with the limitations of fixed monitoring equipment, the detection points of the present application can be flexibly adjusted to ensure that the high-risk areas of the seabed are fully covered. The present application uses an unmanned underwater probe to generate a navigation path based on the location of the target detection point and environmental factors, enabling automatic navigation to each detection point and collecting underwater basic image information. This method avoids the safety risks of manual diving detection and can also perform monitoring tasks stably in complex sea conditions, greatly improving detection efficiency and safety. The present application uses underwater cameras, sonars and other sensing devices to obtain underwater basic images at different positions of each target detection point, and calculates a comprehensive underwater foundation fault score based on different position underwater basic image fault detection models. This image information-based evaluation method can capture a variety of fault information such as corrosion, cracks and subsidence in real time, and has higher accuracy and comprehensiveness than single sensor monitoring, providing accurate evaluation results for the health status of each detection point. The present application draws a health status distribution map of the comprehensive underwater foundation fault score of each detection point and performs smoothing and interpolation processing, making the monitoring results more intuitive. The distribution map can show the health status of all wind turbine foundations in the wind farm, helping operators quickly identify potential risk areas. Compared with the separate monitoring method in the traditional method, the health status distribution map provides an overall perspective, facilitating the development of a comprehensive maintenance plan. Based on the health status distribution map provided by the present application, operators can identify high-risk detection points in advance to achieve early warning of faults and reduce the risk of wind turbine foundation structure failure. In addition, the flexible inspection method of the unmanned underwater probe can optimize the detection frequency, reduce unnecessary manual inspection and maintenance work, and reduce the overall maintenance cost. In summary, the present application solves the problems of coverage limitations, low efficiency and inaccurate evaluation in traditional monitoring methods through intelligent inspection and image information analysis of the unmanned underwater probe, combined with comprehensive fault scores and health status distribution maps, providing an efficient, accurate and intuitive offshore wind farm foundation health monitoring method, which provides a reliable guarantee for the long-term stable operation of the wind farm.

[0069] In some examples, the above determination of a plurality of target detection points based on seabed topography information of a target offshore wind farm includes:

[0070] According to the above seabed topography information, a terrain convolutional neural network model is used for analysis and classification operation to extract a terrain feature vector of each detection point to be detected, wherein the terrain feature vector includes slope information, geological density information and sediment loose degree information; and the seabed topography information is generated based on the detection data.

[0071] The terrain feature vector is clustered based on the DBSCAN method to obtain terrain clustering analysis results, wherein the terrain clustering analysis results include point position information to be detected and clustering label information.

[0072] Risk degree and terrain feature vector calculation are performed based on the terrain clustering analysis results, and preliminary detection points are determined according to the risk degree.

[0073] The preliminary detection points are distributed and optimized by a greedy algorithm based on spatial coverage information and resource limitation information to obtain the target detection points.

[0074] For example, in order to fully understand the seabed terrain of the target offshore wind farm, detection data of the target offshore wind farm is obtained. The detection data includes but is not limited to sonar scanning, topographic mapping and other methods to obtain three-dimensional structural information of the seabed terrain.

[0075] After obtaining the detection data, the seabed terrain information is generated through a data processing step. This step includes preprocessing of the detection data, such as denoising, coordinate standardization and data integrity checking, to ensure the accuracy of the data. Finally, the generated seabed terrain information can truly reflect the undulating morphology of the seabed, the distribution of seabed soil and other characteristics, providing a clear topographic description for subsequent analysis.

[0076] After obtaining the seabed terrain information, a terrain convolutional neural network model is used for analysis and classification to extract terrain feature vectors of different detection points. The terrain convolutional neural network model can automatically identify and extract terrain features related to the health of the wind farm foundation structure. By analyzing the slope, geological density and sediment loose degree of different regions, the terrain feature vector is extracted, which includes slope information, geological density information and sediment loose degree information. The slope information (S) can be calculated by using the seabed terrain data obtained by high-resolution sonar scanning. The slope information (S) is obtained by the following formula:

[0077] S = tan(θ) (where θ is the slope angle, calculated based on sonar data)

[0078] The geological density information (D) can be provided by seabed sediment sampling and laboratory analysis data, by measuring the density of the soil around each detection point. The specific sampling method includes using drilling equipment to extract samples from the seabed and measuring the mass.

[0079] The sediment loose degree information (L) is calculated based on known geological survey data and image analysis results of loose areas of the seabed.

[0080] Based on the above terrain feature vector, the application performs clustering analysis by DBSCAN (Density-Based Spatial Clustering of Applications with Noise) method to obtain terrain clustering analysis results. The above terrain clustering analysis results include position information and clustering label information of the to-be-detected points, and the clustering label identifies the feature type of each region, providing classification basis for further screening of high-risk detection points.

[0081] After obtaining the terrain clustering analysis results, risk degree and terrain feature vector calculation are performed based on the above terrain clustering analysis results to obtain preliminary detection points. Risk degree calculation is to evaluate the risk degree of possible faults according to the terrain features of each to-be-detected point. Specifically, the application assigns a corresponding risk score according to different combinations of slope information, geological density information and sediment loose degree information of the to-be-detected points. The to-be-detected points with higher scores are identified as preliminary detection points, which usually have higher potential risks.

[0082] After obtaining the preliminary detection points, the application optimizes them according to the spatial coverage information and resource limitation information. A greedy algorithm is used to optimize the distribution of the preliminary detection points to ensure that the distribution of the target detection points can maximize the spatial coverage rate and meet the resource limitation conditions (such as the maximum number of detection points, the number of detection equipment, etc.). The greedy algorithm selects the detection points with the highest coverage rate step by step, so that the final target detection points can effectively cover the high-risk areas, thereby ensuring the comprehensiveness and effectiveness of the monitoring.

[0083] In summary, by obtaining the detection data of the target offshore wind farm, generating seabed terrain information, extracting terrain feature vectors using a terrain convolutional neural network model, performing clustering analysis based on DBSCAN method, calculating risk degree to obtain preliminary detection points, and finally performing distribution optimization by a greedy algorithm, the target detection points that meet the spatial coverage information and resource limitation information are obtained.

[0084] In some examples, the above risk degree F is calculated based on the following formula:

[0085]

[0086] wherein S is the slope information, D is the geological density information, L is the sediment loose degree information, R is the other environmental factor information, w1 is the first weight coefficient corresponding to the slope information, w2 is the second weight coefficient corresponding to the geological density information, w3 is the third weight coefficient corresponding to the sediment loose degree information, a is the first adjustment parameter, b is the second adjustment parameter, g is the third adjustment parameter, d is the fourth adjustment parameter, h is the fifth adjustment parameter, f(C) is the clustering label function, g(d) is the position distance information function, the sixth adjustment parameter, and C represents the clustering label information and d represents the position information of the to-be-detected point.

[0087] For example, the terrain feature vector includes the slope information S, the geological density information D, and the sediment loose degree information L, and the seabed terrain information is generated based on the detection data. The slope information S is used to reflect the steepness of the terrain of the detection point. The geological density information D represents the seabed geological density of the target detection point, and can evaluate the stability and impact resistance of the seabed. The sediment loose degree information L reflects the seabed sediment loose degree of the target detection point, and evaluates the structural integrity of the soil in the region. The other environmental factor information E can include but is not limited to environmental factors such as ocean currents and tides, and has an additional impact on the overall risk of the detection point.

[0088] The first weight coefficient w1 is used to adjust the weight of the slope information in the risk degree calculation, and the higher the value, the greater the influence of the slope information on the overall risk. The second weight coefficient w2 is used to adjust the weight of the geological density information, and reflects the relative importance of the geological density in the risk degree. The third weight coefficient w3 is used to adjust the weight of the sediment loose degree information in the risk degree, so that the influence of the sediment loose degree on the overall risk is properly reflected.

[0089] In the clustering label information term Kxf(C), C represents the clustering label information, and represents the label assigned to the detection point in the clustering analysis. f(C) is a clustering label function that converts different clustering labels into different risk levels. For example, certain clustering labels may represent higher-risk geological regions. K is the sixth adjustment parameter, which controls the contribution of the clustering label to the risk degree. This term reflects the influence of the risk category in the clustering analysis on the risk degree. Detection points with different labels may be in different risk regions, and therefore are assigned different risk levels.

[0090] In the position distance information term l xg(d), d represents the clustering label information, and d represents the position information of the to-be-detected point, and represents the distance between the detection point and other key points (such as risk points, historical accident points, etc.). g(d) is a distance function, which is usually The closer the distance, the higher the risk (where ∈ is a small constant to avoid the case of d = 0). λ is the seventh adjustment parameter, used to adjust the weight of distance information in the risk degree.

[0091] Effect: This item considers the impact of location distance on risk. The closer the detection point is to the high-risk area, the higher the risk degree.

[0092] In some examples, the above-mentioned distribution optimization operation of the above-mentioned preliminary detection points according to the spatial coverage information and the resource limitation information to obtain the above-mentioned target detection points includes:

[0093] Based on the above-mentioned preliminary detection points, the above-mentioned spatial coverage information and the above-mentioned resource limitation information, an initial parameter setting operation of the greedy algorithm is performed to obtain an initialized parameter list.

[0094] According to the monitoring radius and the position information corresponding to the above-mentioned preliminary detection points, the coverage range information of the detection points corresponding to each of the above-mentioned preliminary detection points is calculated.

[0095] Based on the coverage range information of all the above-mentioned detection points and the current uncovered area, the spatial coverage contribution value corresponding to each of the above-mentioned preliminary detection points is calculated.

[0096] According to the spatial coverage contribution value corresponding to all the preliminary detection points and the above-mentioned resource limitation information, a greedy selection operation is performed to obtain an updated uncovered area and remaining preliminary detection points.

[0097] Based on the above-mentioned updated uncovered area and the above-mentioned remaining preliminary detection points, a spatial coverage detection operation and a resource limitation detection operation are performed to determine whether the termination condition is met.

[0098] In the case where the termination condition is not met, the above-mentioned uncovered area is updated, the selected detection points are removed, and the coverage contribution value is recalculated, and the above-mentioned greedy selection operation is repeated.

[0099] In the case where the above-mentioned termination condition is met, the current above-mentioned remaining preliminary detection points are determined as the above-mentioned target detection points.

[0100] For example, based on the preliminary detection points, the spatial coverage information and the resource limitation information, the initial parameter setting operation of the greedy algorithm is performed. This step is used to define the parameter list of the greedy algorithm, including the preliminary detection point set, the spatial coverage target and the resource limitation condition (such as the maximum number of detection points). Through this step, the initialized parameter list is obtained, providing a basis for the subsequent operation of the greedy algorithm.

[0101] After the parameter initialization is completed, the detection point coverage range information corresponding to each preliminary detection point is calculated according to the position information and the monitoring radius of each preliminary detection point. The detection point coverage range information is used to determine the spatial region that can be covered by each detection point in its surrounding region. This step provides spatial distribution basis for subsequent spatial coverage contribution value calculation.

[0102] According to all the detection point coverage range information and the current uncovered region, the spatial coverage contribution value corresponding to each preliminary detection point is calculated. The spatial coverage contribution value represents the new coverage effect of the detection point on the uncovered region. This step obtains the contribution size of the detection point to the spatial coverage rate by evaluating the coincidence degree of the coverage range of each preliminary detection point and the uncovered region, and provides a sorting basis for the greedy selection operation.

[0103] After the spatial coverage contribution value of each preliminary detection point is calculated, the greedy selection operation is performed based on the contribution value size and the resource limitation information. Specifically, starting from the detection point with the highest spatial coverage contribution value, detection points are gradually selected to join the target detection point set until the resource limitation condition is reached. After each detection point is selected, the uncovered region and the remaining preliminary detection point set are updated to provide the current spatial coverage information for the next greedy selection operation.

[0104] After each round of greedy selection operation, the spatial coverage rate detection operation and the resource limitation detection operation are performed based on the updated uncovered region and the remaining preliminary detection point. The spatial coverage rate detection operation is used to check whether the current target detection point set has reached the spatial coverage rate requirement, and the resource limitation detection operation is used to confirm whether the remaining detection points are within the resource limitation range. If both meet the termination condition, it is considered to meet the termination condition.

[0105] If the spatial coverage rate detection operation or the resource limitation detection operation does not meet the termination condition, the greedy selection operation is repeated to continue selecting the detection point with the highest spatial coverage contribution value until the termination condition is met.

[0106] When the spatial coverage rate detection operation and the resource limitation detection operation both meet the termination condition, the current remaining preliminary detection point is determined as the target detection point. This step marks the completion of the distribution optimization operation, and finally obtains the target detection point set that meets the spatial coverage rate information and the resource limitation information.

[0107] The embodiment of the present application effectively utilizes the greedy algorithm to optimize the distribution of the preliminary detection point under the resource limitation condition, and realizes the maximization of the spatial coverage rate.

[0108] In some examples, the above control unmanned underwater explorer to reach each of the above target detection points, and obtain the corresponding underwater basic image information of each of the above target detection points, including:

[0109] Based on the position information, environmental information, and seabed obstacle information corresponding to each of the above target detection points, the target path is obtained based on the PPO algorithm.

[0110] Based on the above target path, the unmanned underwater explorer is controlled to reach each of the above target detection points.

[0111] In the case where the unmanned underwater explorer reaches the target detection point, the base type information of the wind turbine corresponding to the current target detection point is determined.

[0112] Based on the above base type information, the target shooting position information is determined.

[0113] Based on the above target shooting position information, the underwater foundation image information is shot.

[0114] For example, based on the position information, environmental information (such as ocean currents, sea currents) and seabed obstacle information of each target detection point, the target path of the unmanned underwater explorer is obtained through the PPO algorithm. The position information of the target detection point is determined by analyzing the seabed topographic information, extracting the topographic feature vector based on the topographic convolutional neural network model, and dividing multiple high-risk areas through the DBSCAN clustering algorithm. The geographic position information of each detection point includes latitude, longitude and depth coordinates, which can be obtained through a multi-beam sonar survey system or an inertial navigation system (INS) combined with GPS, forming accurate spatial coordinate markers. Environmental information includes sea current speed and direction, water temperature, water pressure, visibility and other parameters, which are mainly collected in real time through ocean observation equipment such as offshore buoy systems, tidal radar sites and CTD (conductivity-temperature-depth) profilers, and historical and predicted data from regional ocean databases or weather platforms can also be called. Seabed obstacle information is obtained by detecting the seabed area with the front-looking sonar system, side-scan sonar or underwater laser scanner on the unmanned underwater vehicle, and combined with image recognition algorithms for obstacle recognition and modeling, finally forming a seabed obstacle spatial distribution map.

[0115] PPO algorithm (Proximal Policy Optimization) is a reinforcement learning algorithm suitable for optimizing path planning to ensure that the unmanned underwater explorer can avoid obstacles and efficiently and safely reach the target detection point in a complex seabed environment.

[0116] After obtaining the target path, the navigation of the unmanned underwater explorer is controlled based on the path, so that it reaches each target detection point along the optimal path. By precisely controlling the movement of the unmanned underwater explorer, it is ensured that it can successfully reach the target detection point position according to the plan.

[0117] When the unmanned underwater probe reaches the target detection point, the base type information of the wind turbine corresponding to the current detection point is determined. Different base types of wind turbines (such as single pile foundation, gravity foundation, etc.) affect the structural characteristics of their underwater foundations, so the subsequent image acquisition strategy needs to be adjusted according to the specific base type information.

[0118] After determining the base type information of the wind turbine, the most suitable target shooting position information is determined according to the information. The key monitoring positions of different base types are different, for example, the single pile foundation focuses on the connection between the pile body outer wall and the seabed, while the gravity foundation focuses on the edge of the base and the connection point area. The target shooting position information is used to guide the unmanned underwater probe to collect image information at the key positions of each base type, ensuring the comprehensiveness and effectiveness of the data.

[0119] Finally, based on the target shooting position information, the unmanned underwater probe is controlled to collect underwater foundation image information at each detection point. The image information taken by the probe at the key positions is used for subsequent fault detection and evaluation. These image data contain detailed information of the underwater foundation, which helps to identify the damage or abnormalities of the foundation structure in a timely manner.

[0120] Through the above steps, this embodiment uses PPO algorithm to plan the path of the unmanned underwater probe, accurately controls its arrival at the target detection point, and optimizes the shooting position based on the base type information, finally collects detailed underwater foundation image information, providing reliable data support for the health monitoring of wind turbine foundations.

[0121] In some examples, based on the underwater foundation image information corresponding to each of the above target detection points, a comprehensive fault score of the underwater foundation of the wind turbine corresponding to each target detection point is determined, including:

[0122] According to the base type information, determine the fault weight coefficients corresponding to the underwater foundation images at different positions.

[0123] Based on the underwater foundation images at different positions and the fault detection model, determine the fault scores of the underwater foundation at different positions.

[0124] According to the base type, the fault weight coefficients corresponding to the underwater foundation images at different positions, and the fault scores of the underwater foundation at different positions, determine the comprehensive fault score of the underwater foundation.

[0125] Exemplarily, according to the foundation type information of the wind turbine corresponding to the target detection point, the fault weight coefficients of the underwater foundation images of different positions are determined. Different foundation types (such as single pile foundation, gravity foundation, etc.) have structural differences, and different positions have different influences on the stability of the overall structure. The underwater foundation images of each position are assigned corresponding fault weight coefficients to reflect the criticality of the position. The higher the weight coefficient, the greater the influence of the fault of the position on the overall score.

[0126] After determining the fault weight coefficients of each position, the underwater foundation fault score of each position is calculated based on the underwater foundation image information and the fault detection model of each position. The fault detection model detects the fault features (such as cracks, corrosion, settlement, etc.) in the image using image analysis algorithms and outputs the corresponding score. The score is used to reflect the fault condition detected in the image, and the higher the score, the greater the fault risk of the position.

[0127] Finally, according to the fault weight coefficients and fault scores of different positions, the underwater foundation fault comprehensive score of the target detection point is calculated. Specifically, the fault score of each position is multiplied by its corresponding weight coefficient, and then the scores of all positions are added to obtain the underwater foundation fault comprehensive score. The comprehensive score reflects the overall health status of the detection point, and the higher the score, the greater the fault risk.

[0128] The embodiment realizes the fault risk assessment of the target detection point by assigning fault weight coefficients based on foundation type information, calculating fault scores using a fault detection model, and finally obtaining an underwater foundation fault comprehensive score, which provides a quantitative basis for health monitoring of wind turbine foundations.

[0129] In some examples, in the case of the above foundation type information being a single pile foundation, the above underwater foundation fault comprehensive score E1 is determined based on the following formula:

[0130]

[0131] In the formula, F1 is the fault score of the outer wall of the pile body, F2 is the fault score of the connection between the pile bottom and the seabed, F3 is the fault score of the intertidal zone, W4 is the weight coefficient corresponding to the fault score of the outer wall of the pile body, W5 is the weight coefficient corresponding to the fault score of the connection between the pile bottom and the seabed, W6 is the weight coefficient corresponding to the fault score of the intertidal zone, and ε, θ, ξ, τ are adjustment parameters.

[0132] Exemplarily, in a feasible implementation, when it is detected that the target wind turbine foundation type is a monopile foundation, the system first arrives at the set target detection point by an autonomous underwater vehicle (AUV), and obtains a structural image around the pile body by using a high-resolution underwater camera carried by the AUV. According to the typical structure of a monopile foundation, the system divides the image into three key regions: a pile body outer wall region, a pile bottom and seabed connection region, and a tidal zone region. The system pre-processes the images of the regions, including image enhancement, angle correction, edge extraction and the like, to improve the subsequent recognition accuracy.

[0133] The processed images are input into a defect recognition model based on deep learning. The model is composed of a convolutional neural network (CNN) and is trained in combination with an expert annotated image set, and can automatically identify typical underwater foundation defects such as corrosion, cracks, deformation and settlement in each structural region. The system respectively calculates a pile body outer wall failure score F1, a pile bottom and seabed connection failure score F2, and a tidal zone region failure score F3, and all scores are normalized to the interval [0, 1], and the higher the score, the more serious the failure.

[0134] In order to improve the fitting ability to the actual structure failure risk, the system adopts the following complex nonlinear combination formula to fuse the above three scores to generate an underwater foundation failure comprehensive score E1:

[0135]

[0136] Wherein, W4, W5, W6 are the score weight coefficients of the pile body outer wall, the pile bottom connection and the tidal zone region, respectively, for adjusting the contribution of each region in the scoring result; ε, θ, ξ, τ, ψ are adjustment parameters for controlling the nonlinear response degree of the model to the failure scores of each region.

[0137] The first term in the formula adopts an exponential amplification mechanism, so that when the pile body outer wall has a higher risk, the score rapidly rises, thereby enhancing the sensitivity of the model to the corrosion of the external structure; the second term corresponds to the risk of the pile bottom and seabed connection region, and the third term The third term W6·(F3+θ·F1·F3) represents the linkage between the intertidal zone and the outer wall defects; the fourth and fifth terms are adjustment terms that express the nonlinear coupling relationship between regions and the influence of periodic disturbances, respectively, through the quantity and trigonometric function structures. All parameters in the above formula are calibrated and fitted through training samples and engineering experience data to ensure their stability and discriminability in actual environments. The score result E1 is finally used to construct the health atlas of the wind farm subsea foundation and support the automatic early warning and priority ranking of the inspection task of abnormal structures.

[0138] Specifically, to ensure the accuracy and practicality of the comprehensive scoring model, all weight coefficients (such as W4, W5, W6) and adjustment parameters (such as ε, θ, ξ, τ, ψ) need to be determined through a systematic training and verification process.

[0139] First, the system establishes a training sample set composed of a large number of historical inspection data, manual repair records, and expert scoring results. Each sample data corresponds to a set of regional score values (F1, F2, F3) and a structure health label, which is usually derived from manual assessment grades or known failure states of subsea structures.

[0140] Next, a multi-objective regression method is used to fit the parameters of the above formula. The training goal is to minimize the mean square error (MSE) between the model output score E1 and the manual score label, as follows:

[0141]

[0142] where N is the number of training samples, is the manual score label of the ith sample, is the model calculation result.

[0143] The parameter optimization process uses a genetic algorithm (GA) combined with a particle swarm optimization global search algorithm and a local gradient descent algorithm to avoid getting stuck in local optima. During training, a regularization term is set to constrain the parameter amplitude to prevent model overfitting. After training, the model is evaluated in an independent validation set in terms of accuracy, mean square error, and robustness indicators to ensure its generalization ability and stability.

[0144] When the subsea foundation failure comprehensive score E1 of all target detection points is generated, the system maps the score values of each detection point to a two-dimensional geographical spatial coordinate map of the wind farm. This process includes the following steps:

[0145] 1. Scatter plot generation: The latitude and longitude coordinates of each target detection point and the corresponding score value form a three-tuple (x, y, E1). The scatter plot is generated by plotting all three-tuples on the two-dimensional geographical spatial coordinate map of the wind farm.i y i ,E 1,i ), which constitutes the initial scatter plot distribution of the basic health state.

[0146] 2. Spatial interpolation processing: the system uses the Kriging interpolation method to perform spatial interpolation on the scoring points, fills in the scoring blank areas between the detection points, and generates a complete and continuous health score field.

[0147] 3. Smoothing and normalization: Gaussian kernel smoothing operation is performed on the interpolation results to eliminate local noise points, and the scoring interval is standardized to make the results more suitable for visual expression.

[0148] 4. Heat map rendering: finally, the scoring value is mapped to a color gradient in a pseudo-color mapping manner, and the closer the color is to red, the higher the structural risk is, and the closer the color is to green, the better the health state is. The system output is a health state heat map, which can be integrated into the operation and maintenance platform or the dispatch command center large screen to realize real-time visual inspection state perception.

[0149] In some examples, in the case where the above foundation type information is a gravity foundation, the above underwater foundation fault comprehensive score E2 is determined based on the following formula:

[0150]

[0151] In the formula, F4 is the fault score of the foundation edge, F5 is the fault score of the tower connection, F6 is the fault score of the surrounding seabed, W7 is the weight coefficient corresponding to the fault score of the foundation edge, W8 is the weight coefficient corresponding to the fault score of the tower connection, W9 is the weight coefficient corresponding to the fault score of the surrounding seabed, A, B, C, M and N are adjustment parameters.

[0152] For example, when the system detects that the foundation type information of the target wind turbine is a gravity foundation, the system will perform partition detection according to the distribution characteristics and failure sensitive areas of this structure type, and comprehensively calculate the underwater foundation fault comprehensive score E2 to realize fine structure health evaluation.

[0153] Specifically, the system first reaches the set detection point by an autonomous underwater vehicle (AUV), and uses high-resolution multi-angle underwater camera equipment to collect images of the key areas of the gravity foundation structure. The system divides the collected images into three areas according to the structure layout of the gravity foundation: the foundation edge area, the tower connection area, and the surrounding seabed area. The images in each area are preprocessed, including defogging, distortion correction, illumination normalization, and edge enhancement, to ensure the quality of the input data for the subsequent recognition model.

[0154] Subsequently, the system uses a deep learning-based feature recognition model to automatically analyze the above three types of images. This model is built using a convolutional neural network (CNN) and trained based on a large number of structural failure samples, and can extract the following structural area features:

[0155] For the base edge area, the system mainly identifies features such as changes in the profile ratio, concave, structural asymmetry, etc., and outputs a failure score F4;

[0156] For the tower connection area, the system evaluates the interface connection quality based on image crack detection and connection anchor identification, and outputs a score F5;

[0157] For the surrounding seabed area, the system calls on sonar ripple image and settlement modeling data to identify surrounding sedimentation trends, local subsidence, uneven support, etc. and generate a score F6.

[0158] The above three score items are normalized to the [0, 1] interval, which represents the risk level of each area. The higher the score, the greater the risk.

[0159] Next, the system weights and combines the above three scores based on the following nonlinear combination formula to generate a comprehensive score E2 for the offshore foundation failure:

[0160]

[0161] Where F4, F5, F6 are the failure scores of the base edge, tower connection and surrounding seabed, respectively, and W7, W8, W9 are the regional weight coefficients corresponding to the three types of scores; A, B, C, M, N are multiple adjustment parameters used to control the non-linear growth rate of the score and the strength of the structural coupling term.

[0162] The first term in the above formula uses the arctangent function tan -1 (AF4) to simulate the gradual amplification of the base edge failure on the system score, enhancing the model's sensitivity in the early stages of mild failure; the second term is a weighted ratio structure used to express the risk trade-off relationship between the tower connection and the base edge; the third term combines the square root function and the weighted sum to jointly model the base area and seabed subsidence risk; the fourth term is a nonlinear cross-enhancement term that will cause the score to rise rapidly when the tower connection and base edge both fail; the fifth term uses a multivariate natural logarithm structure to integrate the joint uncertainty information between the three score variables.

[0163] All parameters are obtained by a training sample driven structure regression method. The system first constructs a labeled sample library containing real fault records and inspection evaluation results, records the expert judgment risk level corresponding to each set of structure image pair. Then, a hybrid optimization algorithm such as network search algorithm (grid search) and Bayesian optimization or genetic-gradient hybrid algorithm is used to determine the optimal parameter combination, so that the mean square error between the model prediction value and the expert labeled value is minimized:

[0164]

[0165] Wherein, N is the number of training samples, is the artificial label score, is the model calculation value. To avoid overfitting, the system introduces a regularization term in the training stage, and at the same time, physical constraint checks are made on all score items and parameters. For example, if F6 represents the amount of seabed subsidence, the model will constrain its physically interpretable interval to avoid non-real output.

[0166] The final score result E2 will be used to construct the structural health map of the wind power plant, and can be presented on the platform interface through the health heat map, realizing automatic task scheduling, inspection priority sorting and early warning response strategy formulation based on risk level.

[0167] The embodiment formulates different comprehensive score formulas according to the base type information, so that the single pile foundation and the gravity foundation can be adjusted in scoring calculation, improving the applicability and accuracy of the score. Through the fault weight coefficient and the adjustment parameter, the risk weight of different positions and the interaction between positions can be fully considered, so that the risk calculation is more realistic and scientific. The health state distribution map generated by the fault comprehensive score can quickly and intuitively show the structural health state of the entire offshore wind farm, effectively helping the maintenance personnel to find high-risk areas and timely develop maintenance plans. The score formula design enables the comprehensive use of multi-position information collected by the unmanned underwater vehicle, realizes the complete process from detection to health state evaluation, and improves the accuracy and efficiency of monitoring.

[0168] In some examples, the above-mentioned seabed foundation fault comprehensive score of the wind turbine corresponding to each of the above-mentioned target detection points is used to determine the seabed foundation health state distribution map of the target offshore wind farm, including:

[0169] Mapping the seabed foundation fault comprehensive score of the wind turbine corresponding to each of the above-mentioned target detection points to the geographic space map of the target area to construct a foundation health state scatter plot.

[0170] Performing interpolation and smoothing operations based on the above-mentioned foundation health state scatter plot to obtain the above-mentioned seabed foundation health state distribution map.

[0171] Exemplarily, the underwater foundation fault comprehensive score of each target detection point is correspondingly mapped with the geospatial position of the target area. By positioning the score data of each detection point on the geospatial map of the wind farm, a scatter plot of the foundation health status is constructed. The scatter plot shows the health status of each detection point, and the color or intensity represents the high or low of the fault comprehensive score, so that the operation and maintenance personnel can quickly identify the health status of each detection point in the wind farm.

[0172] After the scatter plot of the foundation health status is generated, the discrete health status scores are converted into a continuous health status distribution map through interpolation operation and smoothing operation. Specifically, the interpolation operation can use algorithms such as Kriging interpolation and bilinear interpolation to supplement the health status distribution of the uncovered area to obtain a continuous state distribution. Subsequently, the smoothing operation is applied to eliminate local mutation areas, so that the health status distribution map is more natural and continuous, facilitating the operation and maintenance personnel to evaluate the health status of the entire wind farm.

[0173] By constructing the scatter plot and performing interpolation and smoothing processing, the underwater foundation health status distribution map of the target offshore wind farm is generated. The distribution map not only can intuitively show the health status of each detection point, but also can reflect the health status of the entire wind farm area through continuous color or intensity distribution, effectively helping the operation and maintenance personnel to identify potential risk areas and analyze health status trends, thereby providing scientific decision support.

[0174] In a second aspect, the present application further provides a remote monitoring and management system for offshore wind farms, comprising:

[0175] A first determination unit 21 is configured to determine a plurality of target detection points based on the seabed topographic information of the target offshore wind farm.

[0176] An acquisition unit 22 is configured to control the unmanned underwater explorer to reach each of the target detection points and acquire the corresponding underwater foundation image information of each of the target detection points.

[0177] A second determination unit 23 is configured to determine the underwater foundation fault comprehensive score of the wind turbine corresponding to each of the target detection points based on the corresponding underwater foundation image information of each of the target detection points.

[0178] A third determination unit 24 is configured to determine the underwater foundation health status distribution map of the target offshore wind farm based on the underwater foundation fault comprehensive score of the wind turbine corresponding to each of the target detection points.

[0179] The above examples are only used to illustrate the technical solutions of the present application, but not to limit the same; although the present application has been described in detail with reference to the foregoing examples, it should be understood by those of ordinary skill in the art that: it can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for remote monitoring and management of an offshore wind farm, characterized in that, The method comprises the following steps: Based on the seabed topographic information of the target offshore wind farm, a plurality of target detection points are determined; Control the unmanned underwater detector to reach each target detection point and obtain the corresponding underwater foundation image information of each target detection point; Based on the underwater foundation image information corresponding to each target detection point, the underwater foundation fault comprehensive score of the wind turbine corresponding to each target detection point is determined; Based on the underwater foundation fault comprehensive score of the wind turbine corresponding to each target detection point, the underwater foundation health state distribution map of the target offshore wind farm is determined; Based on the underwater foundation image information corresponding to each target detection point, the underwater foundation fault comprehensive score of the wind turbine corresponding to each target detection point is determined, which comprises: According to the foundation type information, the fault weight coefficients corresponding to the underwater foundation images at different positions are determined, wherein the underwater foundation images at different positions include images corresponding to the outer wall area of the pile body, the connection area between the pile bottom and the seabed, and the intertidal zone area. Based on the underwater foundation images at different positions and the defect recognition model based on deep learning, the underwater foundation fault scores at different positions are determined. According to the foundation type, the fault weight coefficients corresponding to the underwater foundation images at different positions, and the underwater foundation fault scores at different positions, the underwater foundation fault comprehensive score is determined. In the case of single pile foundation, the underwater foundation fault comprehensive score E1 is determined based on the following formula: wherein, is a failure score of the outer wall of the pile body, is a failure score of the connection between the pile bottom and the seabed, is a failure score of the intertidal zone region, is a weight coefficient corresponding to the failure score of the outer wall of the pile body, is a weight coefficient corresponding to the failure score of the connection between the pile bottom and the seabed, is a weight coefficient corresponding to the failure score of the intertidal zone region, and ε, θ, ξ, and τ are all adjustment parameters. Based on the underwater foundation fault comprehensive score of the wind turbine corresponding to each target detection point, the underwater foundation health state distribution map of the target offshore wind farm is determined, which comprises: Map the underwater foundation fault comprehensive score of the wind turbine corresponding to each target detection point to the geographical space map of the target area to construct the foundation health state scatter point distribution map; Based on the foundation health state scatter point distribution map, interpolation and smoothing operations are performed to obtain the underwater foundation health state distribution map.

2. A method of offshore wind farm remote monitoring management according to claim 1, characterized in that, Based on the seabed topographic information of the target offshore wind farm, a plurality of target detection points are determined, which comprises: According to the seabed topographic information, the terrain convolutional neural network model is analyzed and classified to extract the terrain feature vector of different detection points, wherein the terrain feature vector includes slope information, geological density information and sediment loose degree information; the seabed topographic information is generated based on the detection data; According to the terrain feature vector, DBSCAN method is used for clustering analysis to obtain terrain clustering analysis result, wherein the terrain clustering analysis result includes detection point position information and clustering label information; Based on the terrain clustering analysis result and the terrain feature vector, the risk degree is calculated, and the preliminary detection points are determined according to the risk degree; According to the space coverage rate information and the resource limitation information, the preliminary detection points are distributed and optimized by the greedy algorithm to obtain the target detection points.

3. A method of offshore wind farm remote monitoring management according to claim 2, characterised in that, The risk degree F is calculated based on the following formula: wherein S is slope information, D is geologic density information, L is sediment loose degree information, and R is other environmental factor information, is a first weight coefficient corresponding to the slope information, is a second weight coefficient corresponding to the geologic density information, is a third weight coefficient corresponding to the sediment loose degree information, is a first adjustment parameter, is a second adjustment parameter, is a third adjustment parameter, is a fourth adjustment parameter, is a fifth adjustment parameter, is a clustering label function, is a position distance information function, k is a sixth adjustment parameter, is a seventh adjustment parameter, C represents clustering label information, and d represents position information of a point to be measured.

4. A method of offshore wind farm remote monitoring management according to claim 2, characterised in that, According to the space coverage rate information and the resource limitation information, the preliminary detection points are distributed and optimized by the greedy algorithm to obtain the target detection points, which comprises: Based on the preliminary detection points, the space coverage rate information and the resource limitation information, the initial parameter setting operation of the greedy algorithm is performed to obtain the parameter list after initialization; According to the monitoring radius and the position information corresponding to the preliminary detection points, the detection point coverage range information corresponding to each preliminary detection point is calculated; Based on all the detection point coverage range information and the current uncovered area, the spatial coverage rate contribution value corresponding to each preliminary detection point is calculated; According to the spatial coverage rate contribution value corresponding to all the preliminary detection points and the resource limit information, a greedy selection operation is performed to obtain the updated uncovered area and the remaining preliminary detection points; Based on the updated uncovered area and the remaining preliminary detection points, a spatial coverage rate detection operation and a resource limit detection operation are performed to determine whether the termination condition is met; If the termination condition is not met, the uncovered area is updated, the selected detection points are removed, and the coverage rate contribution value is recalculated, and the greedy selection operation is repeated; If the termination condition is met, the current remaining preliminary detection points are determined as the target detection points.

5. A method of offshore wind farm remote monitoring management according to claim 1, characterized in that, The unmanned underwater explorer is controlled to reach each target detection point to obtain the underwater foundation image information corresponding to each target detection point, including: Based on the position information, environmental information and seabed obstacle information corresponding to each target detection point, a target path is obtained based on the PPO algorithm; The unmanned underwater explorer is controlled to reach each target detection point based on the target path; In the case that the unmanned underwater explorer reaches the target detection point, the base type information of the wind turbine corresponding to the current target detection point is determined; Based on the base type information, target shooting position information is determined; Based on the target shooting position information, underwater foundation image information is shot.

6. A method of offshore wind farm remote monitoring management, characterized in that, It includes: Based on the seabed topographic information of the target offshore wind farm, a plurality of target detection points are determined; The unmanned underwater explorer is controlled to reach each target detection point to obtain the underwater foundation image information corresponding to each target detection point; Based on the underwater foundation image information corresponding to each target detection point, the underwater foundation fault comprehensive score of the wind turbine corresponding to each target detection point is determined; Based on the underwater foundation fault comprehensive score of the wind turbine corresponding to each target detection point, the underwater foundation health status distribution map of the target offshore wind farm is determined; Based on the underwater foundation image information corresponding to each target detection point, the underwater foundation fault comprehensive score of the wind turbine corresponding to each target detection point is determined, including: According to the base type information, the fault weight coefficients corresponding to the underwater foundation images at different positions are determined, wherein the underwater foundation images at different positions include images corresponding to the edge area of the base, the connection area of the tower and the surrounding seabed area; Based on the underwater foundation images at different positions and the deep learning-based feature recognition model, the underwater foundation fault scores at different positions are determined; According to the base type, the fault weight coefficients corresponding to the underwater foundation images at different positions and the underwater foundation fault scores at different positions, the underwater foundation fault comprehensive score is determined; In the case that the base type information is a gravity foundation, the underwater foundation fault comprehensive score E2 is determined based on the following formula: wherein, is a failure score for the tower connection, is a failure score for the surrounding seabed, is a failure score for the tower connection, is a failure score for the tower connection, is a failure score for the surrounding seabed, is a failure score for the surrounding seabed, A, B, C, M and N are all adjustment parameters; Based on the underwater foundation fault comprehensive score of the wind turbine corresponding to each target detection point, the underwater foundation health status distribution map of the target offshore wind farm is determined, including: mapping the comprehensive score of the subsea foundation fault of the wind turbine corresponding to each target detection point to a geographical space graph of the target area to construct a foundation health state scatter plot; performing interpolation and smoothing operations based on the foundation health state scatter plot to obtain a subsea foundation health state distribution map.

7. A method of offshore wind farm remote monitoring management according to claim 6, characterized in that, Based on the seabed topographic information of the target offshore wind farm, a plurality of target detection points are determined, including: According to the seabed topographic information, the terrain convolutional neural network model is used for analysis and classification operation to extract the terrain feature vector of different detection points, wherein the terrain feature vector includes slope information, geological density information and sediment loose degree information; the seabed topographic information is generated based on the detection data; According to the terrain feature vector, the DBSCAN method is used for cluster analysis to obtain the terrain cluster analysis result, wherein the terrain cluster analysis result includes the position information of the detection point and the cluster label information; Based on the terrain cluster analysis result and the terrain feature vector, the risk degree is calculated, and the preliminary detection point is determined according to the risk degree; According to the spatial coverage information and the resource limitation information, the greedy algorithm is used for distribution optimization operation on the preliminary detection point to obtain the target detection point.

8. A method of remote monitoring management of an offshore wind farm according to claim 7, characterized in that, The risk degree F is calculated based on the following formula: wherein S is slope information, D is geologic density information, L is sediment loose degree information, and R is other environmental factor information, is a first weight coefficient corresponding to the slope information, is a second weight coefficient corresponding to the geologic density information, is a third weight coefficient corresponding to the sediment loose degree information, is a first adjustment parameter, is a second adjustment parameter, is a third adjustment parameter, is a fourth adjustment parameter, is a fifth adjustment parameter, is a clustering label function, is a position distance information function, and k is a sixth adjustment parameter, is a seventh adjustment parameter, C represents clustering label information, and d represents position information of a point to be measured.

9. A method of offshore wind farm remote monitoring management according to claim 6, characterized in that, According to the spatial coverage information and the resource limitation information, the greedy algorithm is used for distribution optimization operation on the preliminary detection point to obtain the target detection point, including: Based on the preliminary detection point, the spatial coverage information and the resource limitation information, the initial parameter setting operation of the greedy algorithm is performed to obtain the initialized parameter list; According to the monitoring radius and the position information corresponding to the preliminary detection point, the detection point coverage range information corresponding to each preliminary detection point is calculated; Based on all the detection point coverage range information and the current uncovered area, the spatial coverage contribution value corresponding to each preliminary detection point is calculated; According to the spatial coverage contribution value corresponding to all the preliminary detection points and the resource limitation information, the greedy selection operation is performed to obtain the updated uncovered area and the remaining preliminary detection points; Based on the updated uncovered area and the remaining preliminary detection points, the spatial coverage detection operation and the resource limitation detection operation are performed to determine whether the termination condition is met; If the termination condition is not met, the uncovered area is updated, the selected detection points are removed, and the coverage contribution value is recalculated, and the greedy selection operation is repeated; If the termination condition is met, the current remaining preliminary detection point is determined as the target detection point.

10. A method of offshore wind farm remote monitoring management according to claim 6, characterized in that, The unmanned underwater explorer is controlled to reach each target detection point to obtain the subsea foundation image information corresponding to each target detection point, including: Based on the position information, the environment information and the seabed obstacle information corresponding to each target detection point, the target path is obtained based on the PPO algorithm; The unmanned underwater explorer is controlled to reach each target detection point based on the target path; In the case that the unmanned underwater explorer reaches the target detection point, the base type information of the wind turbine corresponding to the current target detection point is determined; Based on the base type information, the target shooting position information is determined; Based on the target shooting position information, the subsea foundation image information is shot.

Images