A wind power pile structure monitoring method and system

Abstract

The present application relates to the technical field of wind turbine pile monitoring, and in particular to a method and system for monitoring a wind turbine pile structure. The present application proposes the following scheme: generating a radar decision scheme set based on wind speed, wave, ocean current and structural response perception information; controlling a rotating radar to collect echo signals in multiple observation directions based on the radar decision scheme set; inputting the echo signal into a monitoring model, outputting the stress state of the wind turbine pile structure; and generating structural risk prediction information based on the state; the method triggers a game strategy when the disturbance is significant, models the underwater propagation path through Monte Carlo simulation, and selects the optimal observation angle combination; the monitoring model adopts a graph neural network structure, integrates directional context and structural modeling information, and realizes decoding of stress states of multiple parts; the present application improves the structural recognition accuracy and monitoring efficiency in complex environments.

Description

Technical Field

[0001] The present application relates to the technical field of wind power pile monitoring, and in particular to a wind power pile structure monitoring method and system. Background Art

[0002] During the construction and operation of offshore wind farms, wind turbine piles, as the foundation support structures, are exposed to the coupling effects of multiple environmental loads such as wind, waves, and currents for a long time. They are prone to fatigue accumulation, structural damage, and local instability, which affect the operational safety and service life of wind turbines. In order to ensure the integrity and reliability of wind turbine pile structures, there is an urgent need to carry out high-frequency and high-coverage structural health monitoring. Existing structural monitoring methods mostly rely on local sensors such as strain gauges and accelerometers installed inside or on the surface of the pile body, but such methods are complex to deploy, difficult to maintain, and difficult to achieve continuous monitoring of the overall state of the structure. Some technologies attempt to introduce passive sensing methods such as radar to obtain structural responses, but their observation strategies generally adopt fixed directions or polling methods, which cannot adapt to changes in sea disturbances, resulting in large fluctuations in echo signal quality and unstable identification of key parts.

[0003] For example, a Chinese patent with authorization announcement number CN115017822B discloses an integrated monitoring method for offshore wind turbine pile foundations and submarine cables, which obtains wind turbine pile information and establishes a wind turbine pile model; determines a first installation position and supplements a first ultrasonic radar model; obtains second ultrasonic radar information, marks the corresponding second installation position in the wind turbine pile model, and supplements the second ultrasonic radar model in the wind turbine pile model; installs the first ultrasonic radar and the second ultrasonic radar according to the wind turbine pile model; installs a fiber optic interferometer on the structure platform, and connects the two sensing optical fibers in the submarine cable to the fiber optic interferometer; leads another sensing optical fiber from the submarine cable to the submarine cable lateral displacement monitoring system; sets up a data processing platform on the structure platform to receive the collected information from the collection equipment, and the data processing platform integrates the received collected information, evaluates the safety of the entire wind turbine pile foundation, and sends the evaluation results to the onshore monitoring center.

[0004] The above patents have the problems raised by this background technology: ignoring the coupling characteristics between structural directional responses, it is difficult to establish an accurate structural state identification model, and limiting the application effect and engineering value of radar in the field of wind turbine pile structure monitoring. In order to solve the above problems, this application designs a wind turbine pile structure monitoring method and system. Summary of the Invention

[0005] The technical problem to be solved by this application is to address the deficiencies of the existing technology and provide a wind turbine pile structure monitoring method and system, which generates a radar decision plan set based on wind speed, wave, ocean current and structural response perception information; controls the rotating radar to collect echo signals in multiple observation directions based on the radar decision plan set; inputs the echo signal into the monitoring model, outputs the stress state of the wind turbine pile structure; and generates structural risk prediction information based on the state. The method triggers the game strategy when the disturbance is significant, models the underwater propagation path through Monte Carlo simulation, and selects the optimal combination of observation angles. The monitoring model adopts a graph neural network structure, integrates directional context and structural modeling information, and realizes the decoding of stress states in multiple parts. This solution improves the accuracy of structural recognition and monitoring efficiency in complex environments.

[0006] To achieve the above objectives, this application provides the following technical solutions:

[0007] A wind power pile structure monitoring method, the method comprising:

[0008] Obtaining a radar decision solution set based on the perception information;

[0009] According to the decision plan set, controlling the radar to observe the wind power pile and collect radar echo signals;

[0010] Inputting the radar echo signal into a preset monitoring model, and outputting the structural stress state of the wind power pile through the monitoring model, wherein the monitoring model is trained with structural modeling data of the wind power pile;

[0011] According to the structural stress state, structural risk prediction information of the wind power pile is generated.

[0012] According to the decision plan set, the radar is controlled to observe the wind power pile, including:

[0013] Selecting a first observation plan from the decision plan set for performing an initial observation phase, the first observation plan including a plurality of preset observation directions to obtain echo response data in each direction, wherein the initial observation phase is not determined based on a game result of the decision plan set;

[0014] During the execution of the first observation scheme, determining whether a triggering condition for entering a strategic game observation phase is satisfied based on the echo response data;

[0015] If the trigger condition is met, selecting a second observation scheme from the decision scheme set for controlling the radar to perform a subsequent observation phase based on a game result among the observation value of the echo response data, the structural response sensitivity, and the angle switching cost, wherein the subsequent observation phase is determined based on the game result of the decision scheme set;

[0016] The radar is controlled to adjust the observation angle according to the second observation plan.

[0017] The triggering conditions include one or more of the following:

[0018] The observation time of the first observation plan exceeds a preset time threshold;

[0019] The amplitude variance between the radar echo signals in multiple directions collected by the first observation scheme is greater than a preset variance threshold;

[0020] The first observation scheme does not cover the key force directions calibrated in the wind turbine pile structure modeling;

[0021] The difference between the direction combination of the historical observation plan and the current first observation plan in the radar system is greater than a preset difference threshold;

[0022] The direction of wind, wave and current disturbances tends to change in a concentrated manner in the first observation scheme, and the rate of change exceeds the preset rate threshold.

[0023] Selecting a second observation scheme from the decision scheme set for controlling the radar to perform a subsequent observation phase includes:

[0024] Each observation direction in the decision solution set is used as multiple nodes in the Monte Carlo search tree;

[0025] Based on the acquired real-time underwater environmental parameters, a Monte Carlo random simulation process is performed, wherein the random simulation process includes multiple random simulations of the underwater environmental disturbance state and the signal propagation path;

[0026] In each Monte Carlo random simulation process, the observation value probability, structural response sensitivity probability and angle switching cost probability of each node are calculated respectively, and the probabilities are integrated into the node benefit value;

[0027] By statistically summarizing random simulation results, the node benefit value is recursively propagated upward from the node in a tree structure;

[0028] An observation direction corresponding to a node with a highest benefit value is determined from a root node of the Monte Carlo search tree, and a solution corresponding to the observation direction is used as a second observation solution.

[0029] The real-time underwater environmental parameters are acquired by a hydrological sensor array arranged on the wind power pile, and the hydrological sensor array includes a conductivity-temperature-depth probe and an acoustic Doppler current profiler.

[0030] The Monte Carlo random simulation process applies random perturbations to real-time underwater environmental parameters to generate multiple sets of signal propagation paths, and randomly determines reflection, refraction or attenuation events according to a preset scattering model when encountering the seabed or the outer wall of a wind turbine pile.

[0031] The step of obtaining a radar decision solution set based on the perception information includes:

[0032] Based on the perceived information of wind speed, wave parameters, current speed and structural response trend, samples are collected for the feasibility of multiple observation directions to generate a directional feature sampling set;

[0033] Constructing a plurality of observation angle combinations based on the directional feature sampling set, each observation angle combination corresponding to a subset of candidate solutions, wherein the observation angles of the observation angle combinations include at least a preset angle interval;

[0034] Each candidate solution subset is classified into a decision solution set, and the decision solution set is used to provide an angle control solution of the radar for the initial observation phase or the strategic game observation phase.

[0035] Selecting a first observation plan for executing the initial observation phase from the decision plan set includes:

[0036] Based on the perception information and historical observation data, the effectiveness of the observation angle combinations in each candidate solution subset is evaluated, wherein the effectiveness evaluation includes a comprehensive analysis of the signal sampling stability index and the angle coverage breadth index;

[0037] Determine a preferred solution subset including effectiveness evaluation scores among multiple candidate solution subsets;

[0038] An observation angle combination is selected from at least one subset of preferred solutions as a first observation solution for performing the initial observation phase.

[0039] Inputting the radar echo signal into a preset monitoring model, and outputting the structural stress state of the wind power pile through the monitoring model, including:

[0040] The radar echo signals collected at multiple different observation angles and in different time periods are combined with their corresponding observation angle information into time series data with directional context labels;

[0041] The time series data is constructed as a sparse multi-channel input tensor, and directional time series features are extracted through a graph-based echo encoder;

[0042] The directional temporal features are input into the graph attention guidance module to capture the spatial response coupling relationship between each observation direction and generate a structural response potential map;

[0043] Modeling the structural response potential map by embedding the wind power pile structure modeling feature information obtained by pre-training;

[0044] The modeling results are decoupled to decode the stress trends of different structural parts respectively, and the output is the structural stress state including the node stress magnitude, direction and fatigue change rate.

[0045] A wind power pile structure monitoring system, the system comprising:

[0046] An acquisition module is configured to obtain sensory information on wind speed, wave parameters, current speed, and structural response trends, and control the radar to observe at different angles and collect corresponding echo signals;

[0047] a decision module configured to generate a radar decision solution set based on the perception information and select a target observation solution from the decision solution set based on a game strategy when a preset trigger condition is met;

[0048] a control module configured to adjust an observation angle of the radar according to a target observation plan;

[0049] A modeling module is configured to input the collected echo signal into the monitoring model and output the structural stress state of the wind power pile in combination with the structural modeling data;

[0050] An assessment module is configured to generate structural risk prediction information according to the structural stress state.

[0051] Compared with the prior art, the present invention has the following advantages:

[0052] This application introduces a dynamic selection mechanism of observation angles based on game strategy, combines real-time underwater disturbance perception with Monte Carlo simulation modeling, and realizes adaptive radar observation and control of key parts of wind turbine pile structures under variable sea conditions, solving the problems that traditional fixed strategy observation methods cannot effectively cope with changes in disturbance direction, fluctuations in signal quality, and local differences in structural response. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0054] Figure 1 This is a schematic diagram of an exemplary application scenario of an embodiment of the present application;

[0055] Figure 2 This is a flow chart of a wind power pile structure monitoring method according to an embodiment of the present application;

[0056] Figure 3 This is a flow chart of a radar echo signal acquisition method according to an embodiment of the present application;

[0057] Figure 4 This is a schematic diagram of the observation game principle of the embodiment of the present application;

[0058] Figure 5 This is a schematic diagram of the monitoring model structure of an embodiment of the present application. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments.

[0060] References to "embodiments" herein mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it refer to independent or alternative embodiments that are mutually exclusive of other embodiments. It will be understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0061] In the method provided in the embodiments of the present application, during the radar rotation decision-making process, a radar device with rotation control capabilities (hereinafter referred to as a rotating radar) is used to perform angle scheduling and echo signal acquisition of structural targets. A rotating radar specifically refers to a radar unit that has the ability to achieve continuous multi-angle rotation in the horizontal direction through servo control on a fixed installation basis. Its rotation angle range covers no less than 180° and has a minimum controllable angle accuracy of no less than 1°. It supports continuous scanning of multi-angle sequences and fixed-point jumps between non-equidistant angles. In actual deployment, the rotating radar is fixedly installed in the near-sea area above the wind turbine pile, with its central axis parallel to the sea level and its rotation direction located in the horizontal plane to cover wind-wave-current coupled disturbance signals from multiple directions.

[0062] Optionally, all observation strategies in this embodiment can be executed based on the rotating radar, including the multi-angle coverage strategy in the initial observation phase, the directional feature sampling and acquisition during the trigger condition judgment process, and the optimal angle combination execution in the strategy game phase. A linkage mechanism is formed between the radar's control logic and the decision-making scheme. The selection of the observation angle is driven by the decision-making scheme. At the same time, the execution status of the radar angle adjustment result is also fed back to optimize the subsequent angle switching cost evaluation. The angle control unit of the rotating radar is integrated with the radar transmitting and receiving unit, and is equipped with an anti-marine corrosion coating and a wave-proof spray structure to meet the conditions for long-term unmanned deployment at sea.

[0063] Furthermore, to ensure the adaptability of the strategic game in this embodiment to radar angle resource scheduling, the control strategy design limits the radar to a single-antenna structure with continuously adjustable rotation capabilities and without multi-beam separation capabilities, avoiding the use of fixed array or multi-beam scanning radars outside the scope of this embodiment. In addition, to ensure that the basis for the game strategy is based on the limited angle resource scheduling problem, the rotating radar in this application only supports serial angle switching mode and does not include parallel multi-view acquisition capabilities, ensuring the predictability of angle control and the rational convergence of the observation game model.

[0064] See also Figure 1 , which is a schematic diagram of an exemplary application scenario provided in an embodiment of the present application.

[0065] like Figure 1 As shown, the present application is applied to a single-pile wind turbine pile structure in an offshore wind farm, including a radar device arranged on the main body of the wind turbine pile and a sensing device arranged in the underwater area of ​​the wind turbine pile, which are respectively used to obtain wind, wave, and flow disturbance information and underwater environmental parameters.

[0066] Figure 1 In the project, the radar device is installed on the top of the wind turbine pile near the sea level and has a rotatable angle control mechanism to control the radar observation angle according to the decision plan. The radar collects echo signals in sequence at multiple observation angles to reflect the sea surface fluctuations and disturbance characteristics in the current direction. The radar echo signals provide basic data for subsequent structural stress state prediction, including: Figure 1 The dotted arrow in the figure indicates the direction of rotation.

[0067] Figure 1 In the experiment, the sensing device is set in the underwater area where the pile body of the wind turbine is located. It includes a conductivity-temperature-depth sensor (CTD) and an acoustic Doppler current profiler (ADCP), which are used to measure the temperature, salinity, flow rate and other environmental parameters of the seawater in real time, forming underwater environmental input to provide support for disturbance state modeling and signal propagation path prediction in Monte Carlo simulation.

[0068] In the scenario shown, the monitoring system first acquires environmental perception information, generates multiple feasible radar observation scenarios, and selects observation angles with good coverage for data collection during the initial observation phase. If the initial echo results meet the trigger conditions, the system selects the angle with the best observation value from the current decision-making set based on a game strategy to perform subsequent observations.

[0069] The collected echo signals are processed and input into the structural monitoring model. Combined with the wind turbine pile structure modeling data, the structural stress state information including stress magnitude, stress direction and fatigue trend is output. Figure 1The monitoring system is deployed on the pile body and has the overall functions of perception, observation, judgment and analysis. It is suitable for health status assessment and risk monitoring during the entire operation process of wind turbine piles.

[0070] Next, a wind power pile structure monitoring method provided in an embodiment of the present application is introduced with reference to the accompanying drawings.

[0071] See also Figure 2 , which is a flow chart of a wind power pile structure monitoring method provided in an embodiment of the present application. Figure 2 The wind power pile structure monitoring method shown includes the following steps S1-S4, and the specific steps are as follows:

[0072] S1: Obtain the radar’s decision plan set based on the perception information;

[0073] In this embodiment, the sensing information includes wind speed, wave parameters, ocean current velocity, and wind turbine pile structural response trends. This information is acquired jointly by radar and hydrological sensing devices. By comprehensively analyzing this information, a directional feature sampling set consisting of multiple preset observation angles is generated. Multiple observation angle combination schemes are then constructed and organized into a decision-making set for subsequent observation phases. This set of schemes not only includes the observation feasibility of each direction but also determines the coverage and expected signal quality of each scheme based on historical echo results and structural modeling data.

[0074] S2: According to the decision plan set, control the radar to observe the wind power pile and collect radar echo signals;

[0075] In this embodiment, during the initial observation phase, a combination of angles with high coverage is selected from the decision-making set to perform basic data collection. If pre-set trigger conditions are met during the acquisition process, such as a significant fluctuation in the observation direction or insufficient observation of key structural features, a new observation direction is selected as the subsequent observation plan through a multi-factor game mechanism based on observation value, structural response sensitivity, and angle switching cost. After the radar is controlled to adjust the angle, echo data collection continues, forming a time-sharing, multi-directional echo response sequence.

[0076] S3: Inputting the radar echo signal into a preset monitoring model, and outputting the structural stress state of the wind power pile through the monitoring model, wherein the monitoring model is trained by the structural modeling data of the wind power pile;

[0077] In this embodiment, echo signals collected at multiple angles over different time periods and their corresponding angle information are constructed as input data with a temporal structure and directional context, which is then fed into a wind turbine pile structure monitoring model. This monitoring model incorporates physical prior information, such as structural morphology, material properties, and boundary conditions, derived from wind turbine pile structure modeling. This model establishes a nonlinear relationship between perturbation excitation and structural response, ultimately outputting stress state indicators such as stress level, stress direction, and fatigue trend at key structural locations.

[0078] S4: generating structural risk prediction information of the wind power pile according to the structural stress state;

[0079] In this embodiment, a structural risk assessment index system is constructed based on the structural stress state output by the monitoring model. This index includes fatigue accumulation indicators, ultimate stress warning factors, and abnormal deformation trends. Based on this information, structural modeling tolerance thresholds are compared to generate wind turbine pile structural risk prediction information. This information can be used as a criterion for wind farm operational safety, supporting subsequent maintenance scheduling, operating condition adjustments, and lifespan assessment.

[0080] In actual offshore wind turbine operation scenarios, wind turbines are constantly exposed to a complex, turbulent environment characterized by the coupling of ocean currents, waves, and wind. Signal disturbances in this environment are highly random and directional. Furthermore, due to the nonlinear refraction, reflection, and multipath attenuation involved in underwater signal propagation, the fixed observation angles or preset polling schemes used by traditional radar observation strategies make it difficult to stably obtain effective echo signals from key structural locations in complex sea conditions. In particular, in certain high-current or highly reflective environments, blind-scan radar strategies often lead to a significant increase in invalid observations, significantly increasing energy consumption and causing large fluctuations in the accuracy and reliability of structural force identification, seriously impacting the stability and timeliness of structural risk assessment results.

[0081] In this embodiment, the selection of the observation angle is not regarded as an isolated single-objective optimization process. Instead, each optional angle is regarded as a strategic participant between the observation effect, structural response sensitivity and system angle adjustment cost under different environmental disturbance conditions. Multiple rounds of games are carried out under the constraints of limited energy consumption and structural identification accuracy to achieve a dynamic optimal angle selection that takes into account both risk coverage and energy efficiency. The game model uses underwater real-time perception information as input, and randomly models the signal propagation path under different disturbance fields through Monte Carlo simulation, thereby generating a benefit function corresponding to each angle and constructing a tree-like strategy diagram in the observation strategy set. Finally, through benefit tracing and strategy pruning, the dynamic selection of the current optimal observation angle is achieved.

[0082] In one practical application example, when the direction of submarine current changes from east-southeast to northwest, the traditional observation angle configuration still maintains the original direction sequence, resulting in an increase in spurious paths in the echo signal, a decrease in the signal-to-noise ratio in critical structural areas, and increased model prediction errors. However, in this embodiment, after real-time monitoring of underwater velocity vector changes through hydrological sensors, the disturbance field model is automatically updated and the game strategy reconstruction is triggered, quickly converging to a set of observation directions that are consistent with the current current coupling direction and have low propagation loss. This not only effectively improves the confidence of structural stress identification, but also enhances the fatigue risk warning accuracy in key areas without significantly increasing observation time.

[0083] The specific steps of S1 are as follows:

[0084] S1.1: Based on the perceived information of wind speed, wave parameters, current speed, and structural response trend, sample feasibility of multiple observation directions is collected to generate a directional feature sampling set;

[0085] Specifically, in the marine environment where wind turbines are located, disturbance information has obvious temporal non-stationarity and spatial directionality, and there is a coupling relationship between the disturbance intensity in different observation directions and the response capability of the radar echo signal. If this correlation cannot be reflected in the angle selection stage, the subsequent radar echo acquisition is likely to have problems such as redundant observations in invalid directions and insufficient observations in structural risk directions. Therefore, in order to improve the directional adaptability of radar observations, multi-source perception information such as wind speed, wave parameters, current speed, and structural response trends is introduced in the angle decision stage as input for directional feasibility assessment to support more accurate sample collection strategies.

[0086] In this embodiment, the environmental disturbance data within the current observation period is first collected by radar and supporting hydrological sensors. The wind speed is obtained by the anemometer installed on the top of the pile, and the wave parameters include wave height, wave period, and wave direction, which are obtained by analyzing the sea surface scattered waveform; the current velocity information is measured vertically by the underwater ADCP, and the structural response trend is analyzed by the strain gauge or accelerometer arranged inside the wind power pile. After aligning these information with a unified timestamp, each preset observation direction is used as the reference direction, and a group of disturbance vectors with a relatively small angle with the direction are extracted to form a directional disturbance subset. By statistically analyzing the three dimensions of the disturbance intensity mean, directional stability, and echo predictability of the subset, the observation feasibility of the direction is comprehensively evaluated and recorded as a direction sample.

[0087] S1.2: Constructing a plurality of observation angle combinations based on the directional feature sampling set, each observation angle combination corresponding to a subset of candidate solutions, wherein the observation angles of the observation angle combinations include at least one preset angle interval;

[0088] Specifically, after the directional feature sampling set is generated, it cannot be used directly as an observation strategy because actual radar rotation is subject to engineering constraints such as angle continuity restrictions, minimum rotation resolution, time constraints, and power consumption constraints. In addition, it must meet the requirement of complete structural coverage within the observation area. Therefore, it is necessary to construct an observation angle combination that meets physical executable requirements based on the feature sampling set and use it as the core content of the candidate solution subset. This ensures that subsequent game strategies or initial observation strategy selection are operated within a combination solution with real execution capabilities, improving system stability and efficiency.

[0089] In this embodiment, the 360° horizontal angle range is pre-divided into multiple adjacent non-overlapping angle intervals, and the width of each interval is determined by the minimum resolvable rotation angle of the radar mechanism and the system response frequency. When constructing a combination, directions with higher feature scores are selected from the directional feature sampling set, and weighted arrangement is performed according to structural symmetry, the main direction of environmental disturbance, and the location of structural weak areas to form a directional priority sequence. During the combination construction process, a sliding window method is used to generate multiple directional combinations. Each combination contains 3 to 6 directions, and each direction falls within a preset angle interval to ensure uniform distribution of the coverage range. Each combination is regarded as a subset of candidate solutions, and the physical location of the direction in the combination, the corresponding angle interval, the structural coverage target, and the disturbance coupling strength are recorded.

[0090] S1.3: Assign each candidate solution subset to a decision solution set, wherein the decision solution set is used to provide a radar angle control solution for the initial observation phase or the strategic game observation phase;

[0091] Specifically, the candidate solution subset itself is merely a set of potentially feasible angle combinations. However, the angle control during the actual execution phase must be organized and scheduled according to the policy timing logic. This involves categorizing the solutions as either executed during the initial observation phase or invoked during the strategic game observation phase based on task attributes. Control attributes such as policy-level labels, expected observation targets, and reserved energy budgets are then added to form a complete set of schedulable solutions. Therefore, the candidate solution subset needs to be encapsulated as a control unit with multi-dimensional scheduling attributes and integrated into the decision solution set for storage and management.

[0092] In this embodiment, the system reconstructs the decision solution set at the beginning of each monitoring cycle. Before being added to the solution set, each candidate solution subset is assigned a corresponding strategy label based on the structural parts covered by its combination, the degree of directional disturbance coupling, and the angle execution cost: for example, global coverage category, high-risk priority category, low-cost and fast category, and directional fluctuation response category. During the initial observation phase, priority is given to selecting angle combinations from the global coverage category to perform basic observations; during the strategic game phase, priority is given to selecting corresponding solutions from the high-risk priority category for game selection. All solutions are uniformly stored and a number index, priority queue, and state cache are established for the control unit to call according to conditions.

[0093] It is understandable that the initial observation phase does not involve any game-playing, but merely determines an initial observation plan based on the current monitoring situation.

[0094] See also Figure 3 , which is a flow chart of a radar echo signal acquisition method provided in an embodiment of the present application. Figure 3 The method shown can be applied to step S2 of the aforementioned method. The specific steps of S2 are as follows:

[0095] S2.1: Selecting a first observation plan from the decision plan set for performing an initial observation phase, the first observation plan including a plurality of preset observation directions to obtain echo response data in each direction, wherein the initial observation phase is not determined based on a game result of the decision plan set;

[0096] Specifically, the initial monitoring phase lacks a priori knowledge of stable disturbance environment trends and structural responses, necessitating the need to ensure that the data collected in this initial phase possesses structural coverage and directional distribution integrity. Directly filtering the initial plan through a payoff game can lead to critical areas being incorrectly eliminated before the payoffs are fully realized, compromising data quality. To ensure the system maintains basic monitoring capabilities even before identifying structural risk signatures, a pre-set angular coverage scheme is employed during the initial observation phase, independent of echo feedback and game calculation outputs for strategic decision-making.

[0097] As an example, the specific steps of S2.1 are as follows:

[0098] S2.1.1: Divide the decision solution set into a plurality of candidate observation solution subsets, each candidate observation solution subset including at least one observation angle combination;

[0099] Specifically, in practical applications, radar cannot achieve full coverage observations in all directions at the same time, and observation resources must be properly organized and scheduled. Therefore, to effectively manage observation angles, it is necessary to structure the existing angle decision plans into multiple independently executable and prioritized observation plan units, allowing subsequent control processes to be called on demand and deployed in stages. Dividing the plan set into subsets can also improve execution efficiency, reduce duplicate observation rates, and avoid the waste of resources caused by full polling.

[0100] In this embodiment, the generated decision solution set consists of multiple observation angle combinations, each of which represents a set of directional focus areas and their associated angle intervals. The system first divides the angle combinations into subsets targeting structural targets such as the pile foot, pile body, connection section, and pile cap, based on the coverage requirements for different parts of the wind turbine pile in the structural modeling information. Secondly, based on the distribution of main disturbance directions such as wind speed, wave direction, and flow direction, the observation combinations are further clustered according to environmental relevance. Ultimately, multiple candidate observation solution subsets are formed, and the observation directions within each subset are correlated in terms of structural targets or environmental coupling characteristics to enhance the consistency of local effects during subsequent execution.

[0101] S2.1.2: Based on the perception information and historical observation data, perform an effectiveness evaluation on the observation angle combinations in each candidate solution subset, wherein the effectiveness evaluation includes a comprehensive analysis of the signal sampling stability index and the angle coverage breadth index;

[0102] Specifically, once an observation angle combination is generated, it cannot be put into use directly, as some combinations may suffer from directional failure, insufficient structural coverage, or strong signal interference in the current cycle. Therefore, a comprehensive evaluation of the observation angle combinations in the candidate solution subset is necessary to screen for the most promising observation solutions in the current cycle. This evaluation not only considers the quality of individual points in each direction, but also incorporates the structural range, observation stability, and historical execution results to quantify effectiveness across three dimensions: spatial distribution, signal characteristics, and target correspondence.

[0103] In this embodiment, the system acquires sensory data such as wind speed, flow velocity, wave height, and directional change rate within the current cycle. Combined with historical data such as the mean signal-to-noise ratio, fluctuation range, and echo energy decay rate of echo signals from each direction within the previous cycle, the system assigns a current observability label to each angle combination. The specific evaluation method is: on the one hand, it evaluates signal sampling stability, that is, whether the signal quality of each angle in the combination changes smoothly within a continuous time window; on the other hand, it evaluates the angle coverage breadth, that is, the proportion of the structural target area covered by the observation angles in the combination in the structural modeling coordinate system. Finally, a normalized weighted evaluation model is used to generate an effectiveness score for each angle combination.

[0104] S2.1.3: From multiple candidate solution subsets, determine a preferred solution subset that includes effectiveness assessment scores;

[0105] Specifically, to select target combinations for initial observation from multiple candidate subsets, ranking and selection are performed based on the evaluation results from the previous step. This process comprehensively considers the structural coverage capability of the angular distribution within the combination, the ability to adapt to environmental disturbances, and signal stability. It also ensures that different structural areas are included in at least one scenario to avoid missing key areas.

[0106] In this embodiment, the system sorts all candidate solution subsets from high to low according to their corresponding effectiveness scores, initially selecting a subset as the preferred candidate set. These candidate subsets are then analyzed for conflicts, such as whether there are duplicate coverage targets, excessive overlap in observation directions, or physical conflicts in angle execution. If conflicts exist, they are pruned by reducing their priority, ultimately selecting a subset of preferred solutions with balanced structural coverage, high directional diversity, and reasonable execution costs.

[0107] S2.1.4: Select an observation angle combination from at least one preferred solution subset as a first observation solution for performing the initial observation phase;

[0108] Specifically, the final first observation plan must be an angle combination that is directly executable in the current cycle, structurally representative, directionally reasonable, and controllable. This combination must not only be derived from the aforementioned subset of preferred plans but also possess a reasonable directional sequence and acquisition window distribution to meet the requirements for coordinated operation between the radar angle control unit and the data acquisition unit. The selected angle combination serves as the execution unit for the initial observation and will serve as the basic data input for subsequent decisions on whether to enter the game strategy. Its data quality directly affects the reliability of the game strategy modeling.

[0109] In this embodiment, the system prioritizes the angle combination with the most balanced structure target distribution and the highest adaptability to the disturbance direction from the preferred solution subset as the first observation solution. The selected angles are then sorted by path based on the radar's current initial angle to construct an execution queue. The execution time of each observation angle in the first observation solution is determined by the historical signal volatility of its combination.

[0110] It is understandable that if historical fluctuations are large, the observation window will be extended, while if fluctuations are small, the window will be compressed to improve data utilization efficiency. Once the solution is configured, it will be immediately written to the radar control task cache and automatically executed after the radar enters a stable state.

[0111] Preferably, a subset of solutions marked as full coverage or basic polling can be selected from the decision solutions generated in the previous cycle, and the locations of the key stress areas marked in the structural model can be used in combination with the statistical frequency of the multi-directional disturbance directions of wind, waves and currents to select an initial observation solution consisting of multiple angle directions. This solution includes multiple observation angle combinations, and the angle coverage range prioritizes coverage of the main direction of the pile body, the pile foot and the pile cap, with a minimum angle distribution uniformity constraint between directions. The observation time length for each direction is configured based on the signal echo time delay and intensity stability in the historical cycle, and is usually set to an average observation window of three consecutive waveform cycles to ensure the integrity and comparability of the signal echo.

[0112] S2.2: During the execution of the first observation scheme, determining whether a trigger condition for entering the strategic game observation phase is met based on the echo response data;

[0113] Specifically, to improve the efficiency of observation strategy execution and the rationality of resource allocation, this embodiment introduces a trigger judgment mechanism based on radar echo response data during the initial observation phase. This mechanism does not rely on structural state assessment results. Instead, it directly determines whether a disturbance structure has formed in the current monitoring period that can enter the strategic game phase based on the fluctuation characteristics, directional coverage characteristics, and environmental trend information of the initial observation data. By setting multiple independent or combined trigger conditions, the game strategy is ensured to be initiated under the premise of sufficient data support and clear execution value, avoiding the ineffective operation of the game model when the disturbance trend is not obvious or the directional change is insufficient.

[0114] In this embodiment, the trigger conditions include the following technical judgment logic, and if any one of them is met, the strategic game phase begins:

[0115] The observation time of the first observation plan exceeds a preset time threshold;

[0116] In actual operation, the system sets a maximum observation duration threshold to prevent prolonged initial observations, which can lead to data accumulation and delayed strategic responses. Once the radar begins executing the initial observation plan, the system monitors the elapsed time in real time. If the time exceeds the preset upper limit, the initial observation phase is deemed insufficient to determine the state of the structural disturbance, automatically triggering the strategic game phase to shorten the overall response time.

[0117] The amplitude variance between the radar echo signals in multiple directions collected by the first observation scheme is greater than a preset variance threshold;

[0118] After the radar completes initial observations in multiple directions, the system analyzes the echo signals from each direction and extracts the range of signal amplitude variations. If the signal strength fluctuations vary significantly across multiple directions, this indicates significant spatial directional differences in the ocean disturbance, suggesting the presence of a primary disturbance direction or a blind spot. At this point, to promptly adjust the observation strategy to capture more representative directional information, the system enters a strategic game phase, selecting a combination of directions with higher coverage value for further observation.

[0119] The first observation scheme does not cover the key force directions calibrated in the wind turbine pile structure modeling;

[0120] The system maps and compares the structural coverage of the first observation plan based on the predefined high-risk area annotations in the wind turbine pile structural modeling information, such as the pile foot area, connection flange, and pile top transition section. If important structural parts are not covered by the current observation direction, or the coverage area is insufficient to support the structural identification model to make an effective judgment, the system automatically triggers the strategic game phase and calls a subsequent observation plan that includes the relevant structural area direction to compensate for structural coverage.

[0121] The difference between the direction combination of the historical observation plan and the current first observation plan in the radar system is greater than a preset difference threshold;

[0122] The radar system compares the observation direction combination used in the current first observation plan with the observation angle combinations of one or more historical periods, analyzing the angle change and coverage overlap. If the system detects a significant deviation in the angle combination of the current period compared to the historical combination, this may lead to a trend interruption in the structural monitoring data or the accumulation of identification errors. To maintain the continuity of the observation strategy and data comparability, the system enters a strategic game phase, dynamically adjusting the observation direction to achieve the continuation or optimization of the historical coverage path.

[0123] The direction of wind, wave and current disturbances tends to change in a concentrated manner in the first observation scheme, and the rate of change exceeds the preset rate threshold;

[0124] The system uses hydrological sensors to obtain information on current current direction, wave direction, and wind speed changes. Combined with the echo distribution trend in the radar observation direction, it assesses whether the current disturbance direction is showing a trend of concentration. If the system determines that the disturbance intensity in a certain direction is continuously increasing, the directional stability is increasing, and there is directional overlap with an area of ​​high structural response sensitivity, it is determined that the current state has entered a state with a clear main disturbance direction. The system will respond promptly and enter the strategic game phase to maximize the value of structural identification in the current main direction.

[0125] S2.3: If the trigger condition is met, selecting a second observation plan from the decision plan set for controlling the radar to execute a subsequent observation phase based on a game result among the observation value of the echo response data, the structural response sensitivity, and the angle switching cost, wherein the subsequent observation phase is determined based on the game result of the decision plan set;

[0126] Specifically, after the trigger conditions are met, traditional echo quality maximization strategies struggle to cope with the combined risks of sudden changes in disturbance direction and localized structural response. Therefore, the observation direction selection problem needs to be modeled as a dynamic game between disturbance response, structural sensitivity, and control overhead. This game design aims to achieve an optimal balance between the number of angle rotations and delay consumption while maintaining structural coverage and recognition confidence by constructing a multi-objective payoff function.

[0127] In this embodiment, the system uses the current disturbance environment state, structural prediction feedback, and radar angle rotation history as inputs, and utilizes a Monte Carlo simulation-based policy tree model to construct a policy evolution path diagram between multiple directional nodes. The benefit function value on the path is calculated by integrating three factors. The connection relationship between nodes is dynamically updated based on the direction conversion angle, control response delay, and historical switching cost. Finally, the recursive benefit propagation and policy pruning of the tree structure are executed upward, and the current optimal observation path direction combination is output as the second observation plan. Each direction in this plan contains executable angle instructions, expected observation window, target structure coverage information, and expected confidence assessment target value.

[0128] After selecting a second observation plan, the system will evaluate its feasibility for the remainder of the current cycle and optimize it in conjunction with the signal quality information collected during the execution of the initial plan. If it is not expected to be completed within the current cycle, the system will automatically perform structural compression on the plan, retaining only the high-weighted directions and prioritizing information coverage stability in key structural areas.

[0129] S2.4: Controlling the radar to adjust the observation angle according to the second observation scheme;

[0130] Specifically, once the second observation plan is generated, the radar angle must be adjusted quickly and precisely at the physical device level. Due to the inherent inertia of the radar device, overly rapid or drastic angle changes can cause angle drift and echo mismatch. Therefore, the control method must provide smooth transitions and high-precision locking capabilities. Furthermore, the angle priority in the game output plan must be considered to rationally arrange the angle execution order to avoid ineffective switching.

[0131] In this embodiment, after receiving the second observation plan, the radar device's control module first constructs an angle execution list based on the direction sequence and calculates the optimal rotation path based on the difference between the current radar angle and the target angle. The radar device is equipped with a high-precision servo motor and angle encoder. The control logic uses a speed-position dual-loop control strategy for angle adjustment, ensuring that each rotation is within limits and does not jitter. After locking, the specified angle time window is stably maintained before signal acquisition begins. To improve efficiency, the system allows for a small angle slip control strategy between multiple consecutive directions, allowing multiple target angles to be covered in a single continuous rotation, avoiding repeated starts and stops.

[0132] Taking the Monte Carlo search tree as an example, you can refer to Figure 4 To understand, Figure 4 A schematic diagram of the observation game principle provided in an embodiment of the present application shows that the Monte Carlo search tree uses the starting state of the current observation period as the main node, and expands multiple candidate observation directions under the main node, each direction corresponding to a potential starting point of the observation path. The first observation direction and the second observation direction in the figure represent the observation direction selection that can be executed immediately in the current stage, and belong to the first-level decision nodes of the game strategy. These directions are usually derived from the decision-making plan set of the previous stage and are preliminarily screened in combination with the current disturbance field information.

[0133] Figure 4 It shows that when performing simulation deduction, the system further simulates the observation paths of multiple subsequent direction combinations for each initial observation direction node, and forms a branch expansion of the search tree structure, such as Figure 4 The third, fourth, and fifth observation directions shown are candidate observation directions, derived from simulations of echo signal propagation paths and direction switching costs under the current environmental disturbance. These simulated directions are not actually executed but are used to calculate benefits within the tree to support path optimization.

[0134] The specific steps of S2.3 are as follows:

[0135] S2.3.1: Treat each observation direction in the decision solution set as multiple nodes in the Monte Carlo search tree;

[0136] Specifically, traditional observation direction selection strategies often select a direction or a set of directions based on single-point evaluations (e.g., maximum signal strength, highest estimated risk). These strategies lack a systematic perspective and the ability to model path evolution relationships. Especially when environmental perturbations are unstable, these methods often fail to accurately capture the evolutionary trends of high-value directions, leading to insufficient information utilization, misselected directions, or wasted resources. Therefore, we introduce a structured decision-making approach with strong strategy evolution capabilities, proposing to construct an observation direction strategy space based on a Monte Carlo search tree.

[0137] In this embodiment, the system uses all observation directions in the decision solution set generated in the current cycle as the bottom nodes of the search tree, and each node corresponds to a specific angle observation configuration, including attributes such as angle value, direction label, and structural action area. The tree is not constructed based on a fixed hierarchy, but is gradually expanded through the dynamic evolution of the strategy path. It supports the construction of a combined path from any angle starting node upward, allowing different directions to participate in the combined evolution in different ways in multiple rounds of simulation. Each node in the tree also records the state index information of its combination with the upper path, including the cumulative control cost in the path, the covered structural range, and the remaining resource budget. Through this organizational structure, the system establishes an extensible search framework with the ability to express the observation strategy space, supporting the subsequent game solution process based on the disturbance environment and structural response.

[0138] Furthermore, the search tree is not constrained by fixed depth or angular intervals, but allows dynamic insertion of new directions and pruning of inefficient paths according to the game payoff function during the strategy evolution process. It has evolutionary flexibility and strategic plasticity, and establishes a structural foundation for subsequent game payoff propagation and direction optimization.

[0139] S2.3.2: Based on the acquired real-time underwater environmental parameters, execute a Monte Carlo random simulation process, wherein the random simulation process includes multiple random simulations of the underwater environmental disturbance state and signal propagation path, wherein the real-time underwater environmental parameters are acquired by a hydrological sensor array deployed on the wind turbine pile, wherein the hydrological sensor array includes a conductivity-temperature-depth probe and an acoustic Doppler current profiler;

[0140] Specifically, in actual offshore wind farm radar structural monitoring scenarios, due to the complexity of the marine environment surrounding wind turbines, the disturbance state in the surrounding waters exhibits not only high spatial heterogeneity and temporal instability, but also significant nonlinear propagation, refraction, and multipath interference. This disturbance manifests itself in practical engineering scenarios as strong random variations in radar signal paths, abnormal waveform attenuation, and difficulty in accurately predicting refraction paths. In particular, the salinity distribution, water temperature gradient, sound velocity profile, and flow field state within different layers of the seawater vary significantly across seasons, tidal cycles, and wind and wave conditions, resulting in a high degree of uncertainty in radar echo path prediction. This high degree of environmental uncertainty makes it difficult for traditional single-deterministic propagation path modeling methods to fully reproduce the actual observation scenario. Therefore, a technical solution must be developed that can integrate multiple random factors in the underwater environment and model the diversity of radar signal propagation paths to meet the requirements for refined processing of environmental disturbance information for wind turbine structural health monitoring.

[0141] In this embodiment, to achieve the above-mentioned technical objectives, the radar system deploys a hydrological sensor array in the underwater area of ​​the wind turbine. This array includes multiple sensor units, such as conductivity-temperature-depth probes and acoustic Doppler current profilers. The conductivity-temperature-depth probes accurately measure the water's sound velocity gradient and density fields by measuring the temperature, salinity, and depth of the vertical profile of the seawater layer by layer. The acoustic Doppler current profiler monitors the flow velocity, direction, and shear velocity changes in different water layers from the sea surface to the seabed in real time. After these high-temporal and spatial resolution hydrological sensor data are transmitted in real time, the system performs data fusion processing on the data and reconstructs it into a three-dimensional disturbance field data model that can be used to describe the current disturbance state of the water area around the wind turbine in real time. This three-dimensional disturbance field model contains specific water characteristic parameters for each water layer location and can fully reflect the actual physical conditions of radar signal propagation at the current moment.

[0142] Based on the three-dimensional disturbance field model established above, the system further utilizes Monte Carlo random simulation methods to perform multiple random samplings of disturbance states to obtain a statistically representative set of environmental disturbance states. In its implementation, the system employs a random sampling method based on the statistical characteristics of currently measured underwater environmental parameters to generate a large number of disturbance sample scenarios with varying disturbance intensities and structures. These disturbance sample scenarios reflect the various possible states of the environment surrounding the wind turbine, including but not limited to varying salinity distributions, sudden changes in water temperature, amplitudes of changes in the sound velocity gradient, and random fluctuations in flow velocity and direction. In each random sample scenario, the system performs detailed signal propagation path simulations for each possible radar observation direction based on a pre-established physical model of underwater signal propagation. During the specific path simulation, the system tracks the spatial trajectory of the radar signal from the transmitting source point by point. At each trajectory point, the system determines the deflection angle and propagation loss of the signal refraction path based on the local disturbance state of the water body. The system also dynamically calculates the real-time attenuation of signal strength along each potential path.

[0143] In addition, each simulated path will encounter different physical boundary conditions during the propagation process, including the seabed interface, the outer wall surface of the wind turbine pile, etc. In order to accurately characterize the interaction characteristics of these boundaries and signals, when the signal path encounters the seabed or the outer wall of the pile, the system combines the material scattering parameter model calibrated in the actual project to simulate and judge the boundary interaction behavior of the signal one by one. According to the type of boundary material, surface roughness, signal incident angle and the current state of the underwater disturbance environment, the system determines in real time the next step of each path propagation behavior, such as continuing reflection propagation, penetrating refraction propagation, or directly terminating the propagation. Through this process, the system finally obtains a large number of propagation path sample sets under multiple observation directions. Each set contains multiple propagation path results that may appear under different disturbance conditions, providing a propagation path data basis for the game decision-making stage.

[0144] S2.3.3: During each Monte Carlo random simulation, calculate the observation value probability, structural response sensitivity probability, and angle switching cost probability for each node and combine the probabilities into a node benefit value, wherein the Monte Carlo random simulation process applies random perturbations to the real-time underwater environmental parameters to generate multiple sets of signal propagation paths and randomly determines reflection, refraction, or attenuation events when encountering the seabed or the outer wall of the wind turbine pile according to a preset scattering model;

[0145] Specifically, the observation value probability is calculated by counting the proportion of high-quality, valid echo paths obtained in each direction, reflecting the observation effectiveness of that direction under current environmental conditions. The structural response sensitivity probability is analyzed by comparing it with calibration data of key wind turbine pile structures to analyze the structural response capability corresponding to that direction. The angle switching cost probability is estimated by analyzing historical radar control data, reflecting the control cost of switching from the current radar observation angle to the target direction. The probability values ​​of these three indicators are statistically recorded for each random sample. Finally, the system statistically averages or integrates the data from multiple samples of each indicator to obtain a complete and comprehensive benefit evaluation for each direction.

[0146] S2.3.4: Through statistical aggregation of random simulation results, recursively propagate node benefit values ​​from the node upward in a tree structure;

[0147] Specifically, to obtain an executable angle-path combination strategy, it is necessary not only to evaluate the expected return of a single observation direction but also to analyze the overall strategy sequence return formed during the execution of multiple angle combinations. To this end, this embodiment constructs a search tree structure based on direction nodes. After the Monte Carlo simulation is completed, the return value of each node is recursively propagated from the lower layer to the upper layer to construct a comprehensive return evaluation for the multi-path strategy combination.

[0148] In its implementation, each path in the search tree represents a possible radar angle scheduling sequence. The system sets the payoff of each intermediate node in the tree structure as a function of the payoffs of its child nodes. The accumulated payoffs take into account control resource consumption, structural coverage overlap, and directional switching continuity along the path to form the payoff for the complete path strategy. During the propagation process, if a path's payoff is significantly lower than the average, the system automatically prunes the path, removing downstream nodes to reduce interference from ineffective strategies.

[0149] S2.3.5: Determine, from the root node of the Monte Carlo search tree, the observation direction corresponding to the node with the highest benefit value, and use the solution corresponding to the observation direction as the second observation solution;

[0150] Specifically, after calculating the strategy tree path benefits, the system determines the cumulative benefits of each strategy path at the tree root node, selects the optimal path as the observation execution sequence for the current cycle, and chooses the starting direction within that path as the key angle instruction for the second observation plan. This direction not only meets favorable signal propagation conditions, but also exhibits strong structural response expression capabilities under the current disturbance context, and possesses control feasibility and resource conservation characteristics, forming a strategy selection result based on a three-party game equilibrium.

[0151] In this embodiment, the system generates a second observation plan, including execution parameters such as the selected observation direction, predicted control time window, structural target area identifier, and observation sequence number. This plan is written to the radar control task queue and synchronized to the game strategy cache, providing a reference for the next round of strategy adjustments and model feedback. This second observation plan not only dynamically adapts to changing sea conditions but also features execution path optimization, improving the efficiency of critical data acquisition and the continuity of structural monitoring responses.

[0152] Taking the monitoring model as an example, you can refer to Figure 5 To understand, Figure 5 This is a schematic diagram of the monitoring model structure provided in an embodiment of the present application. The monitoring model includes:

[0153] The input layer is used to combine radar echo signals collected at multiple different observation angles and their corresponding observation angle information into time series data with directional context labels;

[0154] Specifically, traditional structural monitoring models often only process time series based on single-point sensor data or single-viewpoint observations. They struggle to integrate distributed radar echo signals from multiple directions and different time windows, making it impossible to fully assess the stress state of wind turbine structures with spatially heterogeneous responses. In offshore wind turbine environments, in particular, there is a complex coupling relationship between the disturbance direction and the observation direction. If the model input cannot fully express the context of the observation angle, the subsequent feature extraction and recognition process will lead to misjudgments due to missing directions or ambiguous positions.

[0155] In this embodiment, to solve this problem, the input layer is designed as a composite input method that integrates time series and directional attributes. The model jointly encodes radar echo signals collected from multiple different observation angles and at different time points with their corresponding observation angle information to construct time series input data with directional context tags. The specific implementation method is as follows: the observation sample at each moment contains a radar echo signal segment, as well as the observation angle identifier, angle switching timestamp and structural target area index corresponding to the segment. All samples are constructed into a time series tensor in the input stage, in which each frame not only contains radar signal strength and spectrum information, but also carries directional semantic information such as the relative orientation of the angle of the frame, radar scanning angular velocity, signal response delay, etc.

[0156] Furthermore, to enhance the model's understanding of angular information, the input layer includes an angle embedding mechanism. By vectorizing the observed angles, the model can identify the relative differences and structural correlations between different directions, enhancing its ability to express the physical connections between directions in the subsequent spatial modeling phase. This input processing method provides a high-dimensional feature foundation for the subsequent graph neural network layer to establish spatial associations, ensuring that the echo data is not only temporally continuous but also structurally spatially identified.

[0157] A graph neural network layer, including an echo encoder and a graph attention guidance module, constructs the time series data into a sparse multi-channel input tensor, extracts directional temporal features through a graph-based echo encoder, and inputs the directional temporal features into the graph attention guidance module to capture the spatial response coupling relationship between each observation direction and generate a structural response potential map;

[0158] Specifically, after receiving time series data with directional context labels, traditional sequence models such as LSTM or CNN cannot display the structural connections between modeling different observation angles during the processing process, and lack the ability to express the response coupling relationship between directions. Especially in the process of structural force propagation, the stress changes in key parts often affect the signal distribution of adjacent parts. Therefore, it is necessary to introduce a feature extraction network with graph structure understanding capabilities.

[0159] In this embodiment, this part includes two closely coordinated modules: the echo encoder and the graph attention guidance module. First, the echo encoder is responsible for extracting local features from the radar echo signal in the input sequence. The encoder is built based on a graph structure, treating each observation direction as a graph node. The nodes are connected by "structural connectivity" and "directional adjacency" extracted from the structural modeling data to build graph edge relationships. Under the graph structure, the system encodes and extracts the time series signal of each direction node to generate a set of directional time series feature representations. These features not only include local features such as the signal's frequency domain energy distribution and time delay morphology, but also retain directional attributes and angle historical change patterns for subsequent spatial coupling processing.

[0160] Next, all extracted directional temporal features are input into the graph attention guidance module, which models the response correlation between different observation directions through a graph attention mechanism. Under this mechanism, the system calculates the attention coefficient between each pair of nodes based on the edge weights between the nodes (derived from the structural coupling degree, the historical synchronous response amplitude, and the directional angle similarity), thereby determining whether the current structural response has strong coupling paths between certain directions. This mechanism allows the model to automatically learn the nonlinear propagation relationships between multiple directions under the current perturbation state and generate a potential map of the structural response. This map describes the strength of each observation direction in identifying the structural state in the current period.

[0161] Furthermore, the graph not only updates the original nodes but also integrates the temporal response trends of adjacent directions through an attention mechanism, simulating information propagation at the structural level. This effectively supports the identification of the common manifestations of non-local damage such as structural fatigue and cracks across multiple directions, enhancing the model's understanding of the dynamics of structural evolution. This is unmatched by traditional methods that directly stack or statically fuse observational data.

[0162] The modeling decoupling layer is used to embed the wind turbine pile structure modeling feature information obtained in advance, model the structural response potential map, and decouple the modeling results to decode the stress trends of different structural parts respectively, and output the structural stress state including the node stress magnitude, direction and fatigue change rate.

[0163] Specifically, there is a nonlinear mapping relationship between radar observation data and structural response, and this relationship is affected by multiple physical factors such as the geometric shape of the wind turbine pile, material properties, and stress boundary conditions. If structural modeling features are not introduced as physical priors, the model will not be able to correctly identify the correspondence between the observation signal and the structural stress, resulting in a lack of physical interpretability and generalizability of the prediction results.

[0164] In this embodiment, the monitoring model decodes the response map into the predicted structural stress state through a modeling decoupling layer. This layer first accesses the wind turbine pile structural modeling features obtained through finite element analysis or historical monitoring data training. These features include high-dimensional vector information such as stress transfer paths, principal stress directions, and stress concentration factor distributions at various locations under different excitation conditions. This information is embedded in the map as structural nodes, providing the model with a physically consistent foundation for feature learning and response assignment.

[0165] This layer then spatially decouples the potential map, dividing the overall structure into multiple subregions corresponding to actual structural segments such as the pile foot, pile body, pile cap, and connection section. Local predictive modeling is performed within each structural subregion to avoid global prediction errors caused by differences in local structural characteristics. Within each subregion, the model deconstructs the node information using a local decoder, outputting the stress state information for that section of the structure, including indicators such as equivalent stress magnitude, principal stress direction, stress change rate, and fatigue damage rate.

[0166] Furthermore, to enhance the robustness of the decoupling layer, the model incorporates an auxiliary task of structural state classification during training, such as determining whether a section is currently in compression, tension, or alternating states. This auxiliary task improves the model's adaptability to structural boundary conditions, reduces unreasonable fluctuations in force predictions, and improves the model's recognition stability and error convergence speed under actual deployment conditions.

[0167] The present invention provides a wind power pile structure monitoring system, which includes:

[0168] An acquisition module is configured to obtain sensory information on wind speed, wave parameters, current speed, and structural response trends, and control the radar to observe at different angles and collect corresponding echo signals;

[0169] a decision module configured to generate a radar decision solution set based on the perception information and select a target observation solution from the decision solution set based on a game strategy when a preset trigger condition is met;

[0170] a control module configured to adjust an observation angle of the radar according to a target observation plan;

[0171] A modeling module is configured to input the collected echo signal into the monitoring model and output the structural stress state of the wind power pile in combination with the structural modeling data;

[0172] An assessment module is configured to generate structural risk prediction information according to the structural stress state.

[0173] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A wind power pile structure monitoring method, characterized in that: The method comprises: Obtaining a radar decision solution set based on the perception information; According to the decision plan set, controlling the radar to observe the wind power pile and collect radar echo signals; Inputting the radar echo signal into a preset monitoring model, and outputting the structural stress state of the wind power pile through the monitoring model, wherein the monitoring model is trained with structural modeling data of the wind power pile; generating structural risk prediction information of the wind power pile according to the structural stress state; The step of obtaining a radar decision solution set based on the perception information includes: Based on the perceived information of wind speed, wave parameters, current speed and structural response trend, samples are collected for the feasibility of multiple observation directions to generate a directional feature sampling set; Constructing a plurality of observation angle combinations based on the directional feature sampling set, each observation angle combination corresponding to a subset of candidate solutions, wherein the observation angles of the observation angle combinations include at least a preset angle interval; Classifying each candidate solution subset into a decision solution set, wherein the decision solution set is used to provide an angle control solution for the radar during an initial observation phase or a strategic game observation phase; Inputting the radar echo signal into a preset monitoring model, and outputting the structural stress state of the wind power pile through the monitoring model, including: The radar echo signals collected at multiple different observation angles and in different time periods are combined with their corresponding observation angle information into time series data with directional context labels; The time series data is constructed as a sparse multi-channel input tensor, and directional time series features are extracted through a graph-based echo encoder; The directional temporal features are input into the graph attention guidance module to capture the spatial response coupling relationship between each observation direction and generate a structural response potential map; Modeling the structural response potential map by embedding the wind power pile structure modeling feature information obtained by pre-training; The modeling results are decoupled to decode the stress trends of different structural parts respectively, and the output is the structural stress state including the node stress magnitude, direction and fatigue change rate.

2. A wind power pile structure monitoring method according to claim 1, characterized in that: According to the decision plan set, the radar is controlled to observe the wind power pile, including: Selecting a first observation plan from the decision plan set for performing an initial observation phase, the first observation plan including a plurality of preset observation directions to obtain echo response data in each direction, wherein the initial observation phase is not determined based on a game result of the decision plan set; During the execution of the first observation scheme, determining whether a triggering condition for entering a strategic game observation phase is satisfied based on the echo response data; If the trigger condition is met, selecting a second observation scheme from the decision scheme set for controlling the radar to perform a subsequent observation phase based on a game result among the observation value of the echo response data, the structural response sensitivity, and the angle switching cost, wherein the subsequent observation phase is determined based on the game result of the decision scheme set; The radar is controlled to adjust the observation angle according to the second observation plan.

3. A wind power pile structure monitoring method according to claim 2, characterized in that: The triggering conditions include one or more of the following: The observation time of the first observation plan exceeds a preset time threshold; The amplitude variance between the radar echo signals in multiple directions collected by the first observation scheme is greater than a preset variance threshold; The first observation scheme does not cover the key force directions calibrated in the wind turbine pile structure modeling; The difference between the direction combination of the historical observation plan and the current first observation plan in the radar system is greater than a preset difference threshold; The direction of wind, wave and current disturbances tends to change in a concentrated manner in the first observation scheme, and the rate of change exceeds the preset rate threshold.

4. A wind power pile structure monitoring method according to claim 2, characterized in that: Selecting a second observation scheme from the decision scheme set for controlling the radar to perform a subsequent observation phase includes: Each observation direction in the decision solution set is used as multiple nodes in the Monte Carlo search tree; Based on the acquired real-time underwater environmental parameters, a Monte Carlo random simulation process is performed, wherein the random simulation process includes multiple random simulations of the underwater environmental disturbance state and the signal propagation path; In each Monte Carlo random simulation process, the observation value probability, structural response sensitivity probability and angle switching cost probability of each node are calculated respectively, and the probabilities are integrated into the node benefit value; By statistically summarizing random simulation results, the node benefit value is recursively propagated upward from the node in a tree structure; An observation direction corresponding to a node with a highest benefit value is determined from a root node of the Monte Carlo search tree, and a solution corresponding to the observation direction is used as a second observation solution.

5. A wind power pile structure monitoring method according to claim 4, characterized in that: The real-time underwater environmental parameters are acquired by a hydrological sensor array arranged on the wind power pile, and the hydrological sensor array includes a conductivity-temperature-depth probe and an acoustic Doppler current profiler.

6. A wind power pile structure monitoring method according to claim 4, characterized in that: The Monte Carlo random simulation process applies random perturbations to real-time underwater environmental parameters to generate multiple sets of signal propagation paths, and randomly determines reflection, refraction or attenuation events according to a preset scattering model when encountering the seabed or the outer wall of a wind turbine pile.

7. A wind power pile structure monitoring method according to claim 2, characterized in that: Selecting a first observation plan for executing the initial observation phase from the decision plan set includes: Based on the perception information and historical observation data, the effectiveness of the observation angle combinations in each candidate solution subset is evaluated, wherein the effectiveness evaluation includes a comprehensive analysis of the signal sampling stability index and the angle coverage breadth index; Determine a preferred solution subset including effectiveness evaluation scores among multiple candidate solution subsets; An observation angle combination is selected from at least one subset of preferred solutions as a first observation solution for performing the initial observation phase.

8. A wind power pile structure monitoring system, used to implement a wind power pile structure monitoring method according to any one of claims 1 to 7, characterized in that: The system comprises: An acquisition module is configured to obtain sensory information on wind speed, wave parameters, current speed, and structural response trends, and control the radar to observe at different angles and collect corresponding echo signals; a decision module configured to generate a radar decision solution set based on the perception information and select a target observation solution from the decision solution set based on a game strategy when a preset trigger condition is met; a control module configured to adjust an observation angle of the radar according to a target observation plan; A modeling module is configured to input the collected echo signal into the monitoring model and output the structural stress state of the wind power pile in combination with the structural modeling data; An assessment module is configured to generate structural risk prediction information according to the structural stress state.

Images