Fault-risk coupling global assessment method and system for network-oriented offshore wind turbine, medium

Abstract

This invention discloses a fault-risk coupling comprehensive assessment method, system, and medium for grid-type offshore wind turbines, belonging to the field of offshore wind turbine fault diagnosis technology. The method includes the following steps: S1: Operating condition classification and adaptive preprocessing; S2: Multi-dimensional joint fault feature mining; S3: Multi-operating condition adapted intelligent fault identification; S4: Fault-risk coupling transmission modeling; S5: Comprehensive risk coupling quantitative assessment; S6: Closed-loop operation and maintenance decision-making. This invention combines the characteristics of grid-type turbine coupled operation with the grid, integrating multi-operating condition adaptive data processing, multi-dimensional fault identification, fault cascading transmission analysis, and comprehensive risk quantitative assessment into a novel method. It overcomes the shortcomings of existing technologies in operating condition adaptation, fault prediction, and coupled risk assessment, providing technical support for the offshore operation and intelligent operation and maintenance of grid-type offshore wind turbines.

Description

Technical Field

[0001] This invention relates to a fault-risk coupling global assessment method, system, and medium for grid-type offshore wind turbines, belonging to the field of intelligent fault diagnosis technology for offshore wind turbines. Background Technology

[0002] The offshore wind power industry is gradually developing towards deep-sea and high-power directions, with grid-type offshore wind turbines becoming a key area of ​​research and development. For example, a 25MW grid-type offshore wind turbine is equipped with 150m long flexible blades, a 160m or more high-flexibility tower, and a high-density electromechanical transmission system. During operation, these turbines exhibit deep coupling characteristics between the electromechanical system and the power grid, and are constantly exposed to complex marine environments such as high salt spray, typhoons, and strong turbulence, making various components prone to wear, deformation, flutter, and other failures.

[0003] In existing technologies, there are several safety risk assessment and fault diagnosis schemes for offshore wind turbines. Patent application publication number CN121352197A discloses a method, device, and storage medium for safety risk assessment of the transmission system of an offshore wind turbine. This scheme collects acoustic signals, vibration signals, temperature data, and meteorological environmental data from the transmission system, extracts spatiotemporal correlation features through a multimodal fusion model, fuses them, and inputs them into an artificial intelligence assessment model to output a safety and health score and risk level. However, this scheme has the following shortcomings: First, its assessment object is limited to a single component of the transmission system, not covering key components such as blades and towers, and cannot achieve a full-domain assessment of the entire turbine. Second, it does not consider the unique characteristics of deep electromechanical-grid coupling in grid-connected turbines, and does not introduce grid-specific features such as grid interaction parameters, making it unable to identify grid-connected faults. Third, it does not distinguish between different operating scenarios such as normal operating conditions, grid disturbance conditions, and extreme sea conditions, and adopts a fixed data processing strategy, which is prone to signal distortion under complex operating conditions. Fourth, it lacks fault propagation analysis and cannot assess the cascading risks of a single-point fault spreading to the entire turbine.

[0004] Patent application publication number CN122046023A discloses an intelligent fault diagnosis system and method for diesel generator sets based on a digital twin system. This scheme collects multi-dimensional physical signals through a multi-source heterogeneous sensing module, processes them through a time-series-physical coupling enhancement module, and then uses an intelligent diagnostic model to identify faults. It also infers potential fault chain paths based on a directed fault propagation graph and classifies risk levels based on subsystem states. However, this scheme has the following shortcomings: First, its application is to diesel generator sets, not offshore wind turbines, and it completely ignores the structural characteristics and fault modes of large flexible components such as blades and towers of offshore wind turbines. Second, it does not address the grid-connected characteristics of grid-type units and does not introduce grid interaction parameters as fault features. Third, it is not specifically designed for extreme marine environments such as typhoons and high salt spray faced by offshore wind turbines, and lacks a correction mechanism for risk assessment based on marine environmental factors. Fourth, its directed fault propagation graph is based on the subsystem topology of the diesel generator set and cannot be directly applied to fault propagation modeling between offshore wind turbine blades, towers, and transmission systems.

[0005] Furthermore, existing offshore wind power monitoring and assessment systems suffer from the following common technical shortcomings: Traditional technologies generally fail to differentiate between normal operating conditions, grid disturbance conditions, and extreme sea conditions, employing fixed data processing strategies that are prone to signal distortion and incomplete noise filtering under scenarios such as grid fluctuations and strong typhoons. Simultaneously, conventional fault identification methods rely solely on single parameter thresholds for judgment, failing to uncover deep time-frequency and entropy domain characteristics of the signal, and lacking sufficient ability to identify subtle faults such as blade micro-vibration and early wear of transmission components. More critically, the structural components of large offshore wind turbines are closely interconnected, and a single-point failure can easily trigger a chain reaction of risks. However, existing assessment methods mostly analyze the state of individual components in isolation, ignoring the transmission effects between faults, resulting in one-sided risk assessments that fail to reflect the true safety level of the entire turbine.

[0006] In the face of industry bottlenecks, there is an urgent need to break through the limitations of existing technologies and, in combination with the coupled operation characteristics of grid-type offshore wind turbines, develop a new method that integrates multi-condition adaptive data processing, multi-dimensional fault identification, fault cascade propagation analysis, and full-domain risk quantitative assessment. This method will make up for the shortcomings of existing technologies in terms of operating condition adaptation, fault prediction, and coupled risk assessment, and provide technical support for the offshore operation and intelligent maintenance of grid-type offshore wind turbines. Summary of the Invention

[0007] To address the aforementioned problems in the existing technology, this invention provides a fault-risk coupling global assessment method, system, and medium for grid-type offshore wind turbines.

[0008] To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, the fault-risk coupled global assessment method for grid-type offshore wind turbines described in this invention includes the following steps: S1: Operating condition classification and adaptive preprocessing: Based on the electromechanical-grid coupled operation of grid-connected units, three scenarios are classified: normal operating condition, grid disturbance operating condition, and extreme sea state. According to the current operating condition, the filtering strategy and data repair rules are dynamically matched to complete the data noise reduction and missing data compensation. S2: Multi-dimensional joint fault feature mining: Time domain, frequency domain, and entropy domain features are extracted simultaneously from the preprocessed monitoring data, and network electromechanical coupling feature parameters are introduced to construct a joint feature set; S3: Multi-condition adaptive intelligent fault identification: Based on a bidirectional long short-term memory network model trained for different operating conditions, combined with dynamic identification thresholds, it distinguishes between single component faults and multi-component coupled faults, and locates the fault type and physical location. S4: Fault-Risk Coupling Transmission Modeling: Based on the structural correlation of wind turbine components and the law of fault propagation, a fault-derived chain risk transmission path model is built to quantify the impact coefficient of a single fault spreading to the whole machine; S5: Comprehensive Risk Coupled Quantitative Assessment: Integrating FMECA failure mode analysis, improved DEMATEL correlation algorithm, and AHP-entropy weight combination weighting method, combined with failure impact coefficient and marine environmental factors, the computer group comprehensively couples the risk index and classifies the risk level; S6: Closed-loop operation and maintenance decision-making: Automatically generates differentiated operation and maintenance strategies based on risk level, and continuously executes the monitoring and evaluation process in a loop.

[0009] Further, in step S1, the adaptive preprocessing for operating conditions specifically includes: for normal operating conditions, fixed-window sliding mean filtering is used to reduce noise and linear interpolation is used to complete missing data; for grid disturbance operating conditions, variable-window Kalman filtering is used to suppress grid coupling noise, and missing data prediction compensation is performed based on the historical operating condition dataset of the grid-connected units; for extreme sea conditions, a multi-level joint noise reduction algorithm is used, and noise interference intervals are marked. The formula for fixed-window sliding mean filtering noise reduction is: ; in, For the original number Frame monitoring data, For the filtered first Frame data, To fix the length of the sliding window; The formula for linear interpolation to complete missing data is: ; in, , For the valid data moments before and after the missing interval, , For the corresponding valid monitoring values, For the time when data is missing, The data is completed after interpolation; The state equation and observation equation of the variable window Kalman filter are as follows: ; ; in, for The system state vector at any given time. for Observe data in real time. , These are the state transition matrix and the observation matrix, respectively, with the window dynamically adjusted according to the intensity of the power grid disturbance. , These are process noise and observation noise, respectively. The hard threshold formula for the multi-level joint noise reduction algorithm is: ; in, For the first Layer wavelet decomposition coefficients, These are the coefficients after noise reduction. The threshold is adaptive and is dynamically updated based on turbulence intensity and salt spray level.

[0010] Furthermore, in step S2, in addition to the conventional features in the time domain, frequency domain, and entropy domain, the joint feature set adds network electromechanical coupling features, including three types of parameters: unit dynamic support current hysteresis, short-circuit current duration, and grid oscillation amplitude; feature extraction is completed using db4 continuous wavelet transform, and the conventional features and electromechanical coupling features are concatenated into a multi-dimensional joint feature set; The formula for the db4 continuous wavelet transform is: ; in, These are the wavelet transform coefficients. As a scale factor, The translation factor is... For the db4 mother wavelet function, This is the preprocessed time series data; The concatenation expression for the joint feature set is: ; in, For time-domain feature vectors, For frequency domain eigenvectors, For entropy domain eigenvectors, This is the characteristic vector of electromechanical coupling in the network.

[0011] Furthermore, in step S3, the bidirectional long short-term memory network model adopts an independent training mode for different operating conditions, and constructs training datasets for normal, power grid disturbance, and extreme sea conditions respectively; at the same time, a dynamic recognition threshold is set, and the threshold tolerance range is automatically relaxed under power grid disturbance and extreme sea conditions. The formula for calculating the dynamic recognition threshold is: ; in, This is the baseline threshold for normal operating conditions. The disturbance factor is the operating condition factor under power grid disturbance conditions and extreme sea states. , This is the threshold relaxation factor; The memory cell update formula of the bidirectional long short-term memory network model is as follows: Input Gate: ; Forgotten Gate: ; Output gate: ; Memory cells: ; Hidden layer output: ; Bidirectional fusion output: ; in, It is the Sigmoid activation function. For element-wise multiplication, For vector concatenation, , These are the model weight matrix and bias vector, respectively, obtained through independent training for each work condition. As candidate memory cells, Input features at time t, For t Hidden layer output at time 1 For t 1. Memory of cell state at any moment Output of the feedforward long short-term memory network model; Output of the feedforward long short-term memory network model.

[0012] Furthermore, in step S4, the fault-risk coupling transmission modeling method is as follows: sort out the fault propagation paths of the three major components of blades, tower, and transmission system, and establish a directed transmission topology graph; The formula for calculating the fault impact coefficient is as follows: ; in, For components Faulty components Influence coefficient, for After the fault occurred Operating parameters for Normal operating parameters; The formula for calculating the probability of fault propagation is: ; in, For the fault from the component Conducted to components The probability, This is the inherent propagation probability. For components Faulty components Influence coefficient, These are marine environmental factors.

[0013] Furthermore, in step S5, the improved DEMATEL algorithm, based on traditional correlation analysis, introduces the failure impact coefficient and marine environmental weighting factors such as salt spray corrosion and turbulence intensity, and recalculates the influence and affected degree of each risk factor; the corrected direct influence matrix is ​​expressed as: ; in, For traditional DEMATEL, the direct influence matrix, This is the fault impact coefficient matrix. This is the marine environment weight matrix. The effect matrix after correction; The AHP-entropy weight combination weighting method calculates the comprehensive weight according to the following formula: ; in, For subjective weights in the analytic hierarchy process, For the objective weight of the entropy weight method, The weighted fusion coefficient; Comprehensive Coupling Risk Index The calculation formula is: ; in, For the first Risk indicator weights The severity of the fault, This represents the probability of fault propagation. This represents the total number of risk indicators.

[0014] Further, in step S6, the unit is divided into four levels according to the comprehensive coupling risk index R: R∈(0,0.3) is the safety level, R∈(0.3,0.6) is the general coupling risk level, R∈(0.6,0.85) is the high coupling risk level, and R∈(0.85,1.0) is the major coupling risk level. A closed-loop operation and maintenance strategy is automatically generated based on the risk level: routine inspections are performed for the safety level; targeted offline detection is performed for the general coupling risk level; the unit output is reduced and time-limited maintenance is carried out for the high coupling risk level; and the unit shutdown protection command is triggered for the major coupling risk level.

[0015] Furthermore, in step S1, the multi-source monitoring data includes blade load, torsion angle, and vibration data; tower strain, tilt angle, and vibration data; electromechanical transmission system temperature, meshing vibration, and acoustic vibration data; and grid interaction voltage, current, and inertia support response data of grid-type units. The data acquisition frame rate is uniformly 50Hz, and each frame of data is bound to a real-time operating condition tag.

[0016] Secondly, the present invention provides a fault-risk coupled global assessment system for grid-type offshore wind turbines, based on the aforementioned fault-risk coupled global assessment method for grid-type offshore wind turbines, comprising: The data acquisition and preprocessing module is used to collect monitoring data of blades, towers, electromechanical transmission systems and grid electrical parameters. Based on the current operating conditions, grid disturbance conditions or extreme sea conditions, it dynamically matches filtering strategies and data repair rules to complete data noise reduction and missing data compensation. The feature mining module is used to simultaneously extract time-domain, frequency-domain, and entropy-domain features from the preprocessed monitoring data, and introduce network electromechanical coupling feature parameters to construct a joint feature set; The intelligent fault identification module has a built-in bidirectional long short-term memory network model trained under different working conditions. Combined with dynamic identification thresholds, it can distinguish between single component faults and multi-component coupled faults, and locate the fault type and physical location. The fault propagation modeling module is used to build a fault-derived chain risk propagation path model based on the structural correlation of wind turbine components and the fault propagation law, and to calculate the impact coefficient of a single fault spreading to the whole machine. The comprehensive risk assessment module integrates FMECA failure mode analysis, improved DEMATEL correlation algorithm, and AHP-entropy weight combination weighting method, and combines failure impact coefficient with marine environmental factors to calculate a comprehensive coupled risk index and classify risk levels. The closed-loop operation and maintenance decision-making module is used to automatically generate and execute differentiated operation and maintenance strategies based on risk levels; The above modules are connected in sequence to form an automated cyclical evaluation link from data acquisition to closed-loop operation and maintenance.

[0017] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the aforementioned fault-risk coupled global assessment method for grid-type offshore wind turbines.

[0018] Compared with existing technologies, the fault-risk coupling global assessment method, system, and medium of the present invention for grid-type offshore wind turbines exhibit the following beneficial effects in terms of technical performance and practical application: First, the invention employs adaptive data preprocessing under three operating conditions to address the technical problem of signal distortion under complex operating conditions. This invention addresses three differentiated operating scenarios faced by grid-connected offshore wind turbines: conventional operating conditions, grid disturbance conditions, and extreme sea states. It designs three preprocessing strategies: fixed-window sliding mean filtering with linear interpolation, variable-window Kalman filtering, and multi-level joint noise reduction, dynamically switching between them according to real-time operating conditions. Compared to the fixed filtering strategies used in existing technologies, this invention effectively suppresses grid-machine coupling noise under grid disturbance conditions, achieving a noise suppression rate of 89.3%; and achieves a signal-to-noise ratio of 43.6 dB under extreme sea states, an improvement of 15.4 dB compared to the fixed strategy, thus avoiding misjudgment problems caused by signal distortion under complex operating conditions.

[0019] Second, this invention introduces network electromechanical coupling characteristics to fill the blind spots in mechanical monitoring and achieve early identification of minor faults. Based on traditional time-domain, frequency-domain, and entropy-domain characteristics, this invention adds three new types of network electromechanical coupling characteristics: dynamic support current hysteresis, short-circuit current duration, and network oscillation amplitude. For minor mechanical faults such as early blade flutter (vibration signal fluctuation <5%), the variation range of pure mechanical characteristics is limited, while the network electromechanical coupling characteristics are significantly amplified (network oscillation amplitude increases by up to 630%), becoming the core basis for fault identification.

[0020] Third, the BiLSTM model trained independently for each operating condition, combined with a dynamic identification threshold, significantly improves fault identification accuracy. This invention trains BiLSTM models independently for three types of operating conditions and sets dynamic identification thresholds, automatically widening the threshold tolerance range under power grid disturbance conditions and extreme sea conditions.

[0021] Fourth, a wind power-specific fault-risk coupling transmission model is constructed to upgrade the assessment from single-point detection to whole-machine cascading risk assessment. This invention analyzes the fault propagation paths of the three major components—blades, tower, and transmission system—establishes a directed transmission topology diagram, and quantifies the probability and severity of fault propagation. By introducing marine environmental factors to correct the fault propagation probability, this invention can accurately calculate the risk of a single fault spreading to the entire machine, thus proactively avoiding the risk of cascading overloads.

[0022] Fifth, this invention improves upon the DEMATEL algorithm by integrating it with AHP-entropy weighting to achieve scientific quantification of risk across the entire domain. Based on traditional DEMATEL correlation analysis, this invention introduces a fault impact coefficient matrix and a marine environment weight matrix to double-correct the direct impact matrix. Simultaneously, it employs the AHP-entropy weighting method to integrate subjective experience and objective data to calculate a comprehensive coupled risk index. The risk index error is controlled within 0.015.

[0023] Sixth, a four-level closed-loop operation and maintenance strategy is implemented to achieve fully automated closed-loop management of assessment, control, and operation and maintenance. Based on the comprehensive coupling risk index, this invention classifies the units into four levels: safe, general coupling risk, high coupling risk, and major coupling risk, corresponding to four differentiated strategies: routine inspection, offline detection, time-limited maintenance with reduced output, and shutdown protection. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the overall process of the method in this embodiment of the invention. Figure 2 This is a logic diagram of adaptive data preprocessing for three types of working conditions in an embodiment of the present invention; Figure 3 The following is a waveform diagram of multi-sensor time-domain detection under normal operating conditions in an embodiment of the present invention; in the figure, (a) is the vibration amplitude of the blade; (b) is the vibration amplitude of the tower; (c) is the temperature of the transmission system; and (d) is the dynamic support current hysteresis. Figure 4 This is a waveform diagram showing the comparison of signal noise reduction effects in an embodiment of the present invention; Figure 5 This is a frequency domain detection spectrum of blade faults under extreme sea conditions (typhoon) in an embodiment of the present invention; Figure 6 This is a line graph showing the detection of three major network features under five modes in the embodiments of the present invention; Figure 7 This is a scatter plot of fault classification detection data from a BiLSTM model in an embodiment of the present invention. Figure 8 This is a thermal detection diagram of fault conduction probability in an embodiment of the present invention; Figure 9 This is a bar chart comparing the fault indicators of the embodiments of the present invention and existing methods. Detailed Implementation

[0025] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0026] Example 1: This invention applies to a 25MW grid-connected offshore wind turbine, which features 150m long flexible blades, a 162m high flexible tower, and a high-density electromechanical transmission system. The turbine comes standard with a complete set of monitoring sensors covering the three core components, with a unified data acquisition frame rate of 50Hz. An edge computing unit is integrated to handle algorithm computation, data storage, and remote transmission, reusing the turbine's existing sensor network.

[0027] 150m long flexible blade: Load, torsion, and vibration sensors are installed at the blade root, middle section, and blade tip.

[0028] 162m high flexible tower: Strain, tilt, and vibration sensors are evenly distributed along the height.

[0029] High-density electromechanical transmission system: Temperature, vibration, and acoustic sensors are installed at the locations of the gearbox and generator.

[0030] Dedicated monitoring unit for network construction: Real-time acquisition of electromechanical coupling parameters such as dynamic support current, short-circuit current duration, and network oscillation.

[0031] The sensor's acquisition dimensions and range are shown in Table 1: Table 1 Sensor Acquisition Dimensions and Range

[0032] Model training: Based on historical operating data of the unit, laboratory simulated fault data, and marine measured fault samples, three sets of BiLSTM models were independently trained for normal operating conditions, power grid disturbance conditions, and extreme sea conditions, respectively.

[0033] Parameter configuration: In the filtering parameters, the sliding window for normal operating conditions is N=20, the extreme sea state is db4 wavelet 5-level decomposition, and the adaptive noise reduction threshold is T=0.85; in the dynamic threshold coefficients, the threshold relaxation coefficient for power grid disturbance and extreme sea state is α=0.2, the baseline threshold Thd0=0.5; the combined weighting fusion coefficient is λ=0.5; and the fault propagation baseline probability is P0=0.45.

[0034] Database: Built-in FMECA fault mode library, fault propagation topology rules, and four-level risk assessment criteria. Risk levels are divided into four categories: Safe (comprehensive coupling risk index range 0~0.3), General coupling risk (0.3~0.6), High coupling risk (0.6~0.85), and Major coupling risk (0.85~1.0). The corresponding handling strategies for each level are as follows: Safe level: routine periodic inspections; General coupling risk: offline special testing; High coupling risk: remote output reduction and time-limited maintenance; Major coupling risk: immediate shutdown and protection.

[0035] like Figure 1 As shown, this embodiment provides a fault-risk coupled global assessment method for grid-type offshore wind turbines, which is performed according to the following steps: S1: Adaptive Preprocessing of Operating Conditions and Multi-Source Monitoring Data like Figure 2 As shown, the operating conditions of grid-connected generating units are divided into three categories: normal operating conditions, grid disturbance operating conditions, and extreme sea conditions. Based on the current operating conditions, dynamic matching filtering strategies and data repair rules are used to complete data noise reduction and missing data compensation.

[0036] After the system is powered on and started, the edge computing unit collects structural monitoring data and network electromechanical coupling data in real time and automatically determines the operating conditions based on real-time marine meteorology and power grid operation status, executes differentiated preprocessing strategies, and binds all data to operating condition tags in real time.

[0037] Under normal operating conditions (stable sea surface, no power grid fluctuations): a fixed-window moving average filter is used to remove minor environmental noise, and a linear interpolation algorithm is used to complete the missing short-time signal data. The formula for the fixed-window moving average filter is: ; in, For the original number Frame monitoring data, For the filtered first Frame data, To maintain a fixed sliding window length, this scheme uses N=20. The linear interpolation formula is: ; in, , For the valid data moments before and after the missing interval, , For the corresponding valid monitoring values, For the time when data is missing, This is the data after interpolation and completion.

[0038] like Figure 3 The figure shown is a waveform diagram of time-domain detection by multiple sensors under normal operating conditions. Figure 3 (a) shows the time-domain waveform of the blade vibration, with the vertical axis in g and a range of 0.100–0.140 g. The blade vibration value fluctuates between 0.103 and 0.134 g, with peak values ​​appearing near the 150th and 650th sampling points. The steady-state mean is 0.117 g. The waveform is smooth without spikes or abrupt changes, indicating that the 150 m long flexible blade operates stably under normal conditions.

[0039] Figure 3(b) shows the time-domain waveform of the tower vibration, with the vertical axis in g and a range of 0.075~0.105g. The tower vibration value fluctuates in the range of 0.078~0.101g, with the overall fluctuation amplitude slightly smaller than that of the blade vibration, and the amplitude stabilizes at around 0.09g, proving that the 162m high flexible tower has a normal structural response under normal operating conditions.

[0040] Figure 3 (c) shows the time-domain waveform of the transmission system temperature, with the vertical axis in °C and a range of 38.12~38.30℃. The temperature fluctuates slightly within the range of 38.14~38.28℃, with the overall temperature change within 0.14℃, remaining constant at around 38.2℃. There are no abnormal temperature rise fluctuations, indicating that the high-density electromechanical transmission system has normal heat dissipation and is operating well.

[0041] Figure 3 The middle (d) is the time-domain waveform of the dynamic support current hysteresis, with the vertical axis in ms and a range of 5.95~6.25ms. The hysteresis fluctuates in the range of 5.97~6.23ms and stabilizes at around 6.1ms, which is within the normal range (0~10ms), indicating that the inertial support response of the grid-type unit is normal and there are no electrical abnormalities.

[0042] The four sub-figures comprehensively illustrate the time-domain characteristics of multi-source monitoring data under normal operating conditions: all waveforms are continuous, stable, distortion-free, and free of abnormal pulses, proving that under normal operating conditions with a stable sea surface and normal power grid, the multi-source data acquisition and basic preprocessing module of this invention is working normally, the sensor data is authentic and reliable, and the unit's mechanical and electrical systems are in a healthy operating state. This figure serves as a benchmark reference, providing a normal sample basis for waveform comparison under various subsequent fault conditions.

[0043] Taking a normal test wind field off the coast of Shandong as an example, the wind speed was 8-10 m / s, the turbulence intensity was 0.08, the power grid was stable, and the rated output of the generating unit was 25 MW. The test lasted 10 minutes, and 30,000 frames of data were collected. The original 20 consecutive frames of vibration data are as follows: [0.12,0.11,0.13,0.12,0.10,0.11,0.12,0.13,0.11,0.12,0.12,0.11,0.13,0.12,0.10,0.11,0.12,0.13,0.11,0.12]; The moving average was calculated to be 0.117g. After filtering, the signal-to-noise ratio (SNR) of the entire component signal was 48.2dB, with no missing data, and linear interpolation was not enabled.

[0044] Grid disturbance condition (small fluctuations in grid voltage / frequency, unit startup inertia support function): Switch to variable window Kalman filtering to suppress interference signals generated by grid-generator coupling, and use historical operating condition datasets to predictively compensate for missing data. The state equation and observation equation for the Kalman filter are as follows: ; ; in, for The system state vector at any given time. for Observe data in real time. , These are the state transition matrix and the observation matrix, respectively, with the window dynamically adjusted according to the intensity of the power grid disturbance. , These are process noise and observation noise, respectively.

[0045] Taking the offshore wind field in Shandong as an example, the wind speed was 9 m / s, the turbulence intensity was 0.10, and the short-term fluctuation of the power grid frequency was 49.2~50.8 Hz. The test lasted for 8 minutes, and 24,000 frames of data were collected. The state transition matrix A=[0.98,0.01;0.02,0.97], and the observation matrix H=[1,0;0,1]. After filtering, the noise suppression rate of the electromechanical signal was 89.3%, the signal SNR was 39.7 dB, and there was no distortion of the mechanical component signal.

[0046] Extreme sea states (typhoons, strong turbulence): A multi-level joint denoising algorithm is employed to suppress high-intensity environmental noise, while noise interference intervals are marked to prevent invalid data from entering subsequent analysis stages. The multi-level joint denoising uses db4 wavelet decomposition, and the hard threshold formula is as follows: ; in, For the first Layer wavelet decomposition coefficients, These are the coefficients after noise reduction. The threshold is adaptive and is dynamically updated based on turbulence intensity and salt spray level.

[0047] Taking the offshore and deep-sea area near Shandong as an example, the test was conducted under conditions of a Category 8 typhoon, turbulence intensity of 0.38, and high salt spray. The test lasted 12 minutes and collected 36,000 frames of data. A 5-level decomposition using db4 wavelet was employed, with an adaptive threshold T=0.85.

[0048] like Figure 4The image shows a waveform comparison of the signal denoising effect. The upper waveform is the original typhoon environmental signal, which is noisy and cluttered, with a signal-to-noise ratio (SNR) of 28.2 dB. The lower waveform is the signal after multi-level joint denoising using the db4 wavelet of this invention, which is smooth and retains fault features completely, with an SNR of 43.6 dB. The comparison of the two waveforms intuitively demonstrates the excellent performance of the multi-level joint denoising algorithm designed for extreme sea conditions in this invention, which filters out high-intensity environmental noise while retaining effective fault features. The SNR is significantly improved by 15.4 dB, verifying the technical advantages of the three-condition differentiated preprocessing strategy and solving the problem of signal distortion under extreme sea conditions in existing technologies.

[0049] S3: Multi-dimensional joint fault feature mining: The db4 wavelet mother function is used to perform continuous wavelet transform on the preprocessed data to extract three basic features in the time domain, frequency domain, and entropy domain. The specific electromechanical coupling features of the grid-type unit (dynamic support current hysteresis, short-circuit current duration, and grid oscillation amplitude) are superimposed to form a multi-dimensional joint feature set, which fully covers two types of feature information: mechanical faults and abnormal grid-machine interaction.

[0050] The formula for db4 continuous wavelet transform is: ; in, These are the wavelet transform coefficients. As a scale factor, The translation factor is... For the db4 mother wavelet function, This is the preprocessed time series data. The joint feature set concatenation expression is: ; in, For time-domain feature vectors, For frequency domain eigenvectors, For entropy domain eigenvectors, This is the characteristic vector of electromechanical coupling in the network.

[0051] like Figure 5 The image shows a frequency domain spectrum for blade fault detection under extreme sea conditions (typhoon). The horizontal axis represents frequency (0-20Hz), and the vertical axis represents vibration amplitude. The red curve represents a normal blade with a low overall amplitude; the blue curve represents early blade flutter under extreme sea conditions (Category 8 typhoon), with a significantly increased amplitude compared to a normal blade. Traditional single-threshold detection relies solely on time-domain amplitude and cannot identify subtle early faults such as blade micro-flutter. The frequency domain spectrum, however, can capture newly added characteristic frequencies, intuitively demonstrating the technical advantages of this invention in extracting multi-dimensional features across the time, frequency, and entropy domains, thus solving the problem of missed detection of subtle faults in existing technologies. Frequency domain features are the core basis for identifying early flutter in flexible blades.

[0052] like Figure 6 The graph shown is a line graph of the three main network characteristics detected under five modes. The horizontal axis represents five operating states: normal, blade flutter, transmission wear, converter fault, and multi-component coupling fault. The three curves correspond to the three main network-specific characteristics: the dynamic support current hysteresis changes from the normal value of 6.2ms to 24.8ms, 19.5ms, 37.1ms, and 45.3ms (threshold 10ms); the short-circuit current duration changes from 15.1ms to 18.3ms, 26.7ms, 42.5ms, and 51.2ms (threshold 20ms); and the grid oscillation amplitude changes from 0.021pu to 0.146pu, 0.068pu, 0.192pu, and 0.275pu (threshold 0.05pu).

[0053] Significant differences exist in the degradation patterns of different fault types: mechanical faults are mainly characterized by increased oscillation amplitude and current hysteresis, electrical faults are mainly characterized by excessive short-circuit duration, and all three indicators of coupled faults are severely exceeded. This data demonstrates that network coupling characteristics can accurately distinguish fault types and severity.

[0054] The average characteristics under power grid disturbance conditions are as follows: all mechanical characteristics are normal, and the network coupling characteristics fluctuate slightly (current lag 11.3ms, short-circuit current 16.2ms, and machine-grid oscillation 0.062pu), which are within the normal response range.

[0055] Mean characteristics under extreme sea conditions and early blade flutter: blade vibration acceleration 0.31g (+165% from baseline), spectral entropy 0.48 (+128%), current hysteresis 24.8ms (+307%), and machine-network oscillation amplitude 0.146pu (+630%). The network coupling characteristics are significantly amplified, becoming the core basis for identifying weak faults.

[0056] S3: Multi-condition adaptive intelligent fault identification: The corresponding trained BiLSTM model is invoked based on the current operating conditions, and fault determination is performed using a dynamic recognition threshold. The BiLSTM model consists of a forward LSTM and a backward LSTM, and the output fused features are used for fault classification. The core memory cell update formula is: Input Gate: ; Forgotten Gate: ; Output gate: ; Memory cells: ; Hidden layer output: ; Bidirectional fusion output: ; in, It is the Sigmoid activation function. For element-wise multiplication, For vector concatenation, , These are the model weight matrix and bias vector, respectively. The weight matrix and bias vector are obtained through independent training for each work condition. As candidate memory cells, Forward LSTM output; : Output to LSTM Input features at time t, For t Hidden layer output at time 1 For t 1. Memory cell state at any given moment.

[0057] The formula for the dynamic recognition threshold is: ; in, This is the baseline threshold for normal operating conditions. Operating disturbance factor (grid disturbance / extreme sea state) ), This is the threshold relaxation factor. The threshold tolerance range is automatically relaxed under extreme sea conditions and power grid disturbances to reduce the probability of misjudgment; the model can accurately identify single-component faults and multi-component coupled faults, and label the fault type and specific location.

[0058] like Figure 7 The figure shows a scatter plot of fault classification detection using the BiLSTM model. The horizontal and vertical axes represent the two-dimensional feature values ​​output by the BiLSTM model. Blue dots (normal samples) are concentrated in the low-feature region; yellow triangles (single-component faults) have medium feature values; and red squares (multi-component coupled faults) have significantly high feature values. The cluster boundaries of the three types of samples are clear, proving that the BiLSTM model trained under different operating conditions can accurately distinguish between fault-free, single-component, and multi-component coupled faults. The dynamic threshold line distinguishes the judgment criteria for different operating conditions, verifying the rationality of the design of automatically relaxing the threshold under power grid disturbances and extreme sea conditions, effectively reducing misjudgments caused by environmental interference. The overall comprehensive recognition accuracy is 97.9%, and the detection rate of weak faults is 96.0%.

[0059] Specific applications are as follows: Under normal operating conditions with no faults: A BiLSTM model under normal operating conditions is called, with a baseline threshold Thd0 = 0.5, and the model fusion output value H... t =0.22<0.5, indicating no fault.

[0060] Fault-free operation under grid disturbance conditions: Using the grid disturbance condition BiLSTM model, the dynamic threshold Thd = 0.5 × (1 + 0.2 × 0.35) = 0.535, and the model output H.t =0.41<0.535, which is considered normal machine-network interaction.

[0061] Extreme sea state with early blade flutter: Using the extreme sea state BiLSTM model, the dynamic threshold Thd = 0.5 × (1 + 0.2 × 0.42) = 0.542, and the model output H... t =0.71>0.542, indicating early flutter in a 150m long flexible leaf (mid-section).

[0062] Multi-component coupled fault: Call the conventional operating condition model and output H t =0.92>0.5, indicating a multi-stage coupling fault from blade flutter to transmission wear to tower resonance.

[0063] To fully verify the performance of the fault identification model of this invention, 50 repeated tests were conducted for each of the four typical scenarios, for a total of 200 test samples. The comprehensive performance statistics are shown in the table below: Table 2 Test data for four typical scenarios of the present invention

[0064] As shown in Table 2, the overall fault identification accuracy of this invention reaches 97.9%, with an average response delay of only 22.7ms, maintaining stable high performance under different operating conditions.

[0065] S4: Fault-Risk Coupling Transmission Modeling The identification results are retrieved and matched against the built-in fault propagation topology map to trace the fault propagation paths of the three main components: blades, tower, and transmission system. The formula for calculating the fault influence coefficient is as follows: ; in, For components Faulty components Influence coefficient, for After the fault occurred Operating parameters for Normal operating parameters.

[0066] The formula for calculating the probability of fault propagation is: ; in, For the fault from the component Conducted to components The probability, This is the inherent propagation probability. For components Faulty components Influence coefficient, These are marine environmental factors.

[0067] like Figure 8 The diagram shown is a thermal mapping of fault propagation probability. The rows and columns correspond to the three core components: a 150m long flexible blade, a high-density transmission system, and a 162m high flexible tower. The color coding is: blue (0-30%, low risk), yellow (30-70%, medium risk), and red (70-100%, high risk). The measured probabilities are: blade to transmission 61%, transmission to tower 80%, tower to blade 78%, and environmental factor η. env =0.72 (high salt spray / typhoon). The three main components of an offshore wind turbine are structurally interconnected; a single point of failure can easily trigger a chain reaction, which is a major shortcoming of existing technology (isolated detection, ignoring fault propagation). The heatmap uses color and numerical values ​​to visually represent the probability of fault propagation: blade failure has a 61% probability of causing transmission overload, and transmission failure has an 80% probability of propagating to the tower, representing a high-risk link. The probability is calculated in conjunction with marine environmental factors, closely reflecting the complex operating environment of deep-sea areas.

[0068] S5: Comprehensive Risk Coupled Quantitative Assessment By integrating FMECA failure mode analysis, the improved DEMATEL algorithm, and the AHP-entropy weight combination method, a comprehensive coupling risk index is calculated.

[0069] The improved DEMATEL algorithm, based on traditional correlation analysis, introduces fault impact coefficients and marine environmental weighting factors such as salt spray corrosion and turbulence intensity. The corrected direct impact matrix is ​​expressed as follows: ; in, For traditional DEMATEL, the direct influence matrix, This is the fault impact coefficient matrix. This is the marine environment weight matrix. This is the revised impact matrix, used to recalculate the impact and affected degree of each risk factor.

[0070] The AHP-entropy weight combination method calculates the comprehensive weight according to the following formula: ; in, For subjective weights in the analytic hierarchy process, For the objective weight of the entropy weight method, This is the weighted fusion coefficient, with a value between 0 and 1, used to balance the contribution ratio of subjective experience and objective data.

[0071] Comprehensive Coupling Risk Index The calculation formula is: ; in, For the first Risk indicator weights The severity of the fault, This represents the probability of fault propagation. This represents the total number of risk indicators.

[0072] Specifically, the first step is to determine the weights of the risk indicators. The AHP-entropy weighting method is used to first calculate the subjective weights of the analytic hierarchy process. and the objective weight of the entropy weight method Then according to Calculate the overall weight.

[0073] The second step is to determine the severity of the fault. Based on FMECA failure mode analysis, the severity level of the failure (such as blade flutter, transmission wear, tower resonance), frequency of occurrence, and difficulty of detection are comprehensively evaluated, with a value range of (0, 1). The larger the value, the higher the degree of failure severity.

[0074] The third step is to determine the probability of fault propagation. Based on the fault-risk coupling transmission modeling results from step S4, the fault impact coefficient and marine environmental factors are substituted to calculate the value, which ranges from (0, 1). The larger the value, the higher the risk of fault propagation.

[0075] The fourth step is to calculate the comprehensive coupling risk index R. (The above...) , , Substituting into the formula, we obtain the overall coupling risk index R, which takes values ​​in the range of (0, 1). Based on the R value, the unit is divided into four levels: safe, general coupling risk, high coupling risk, and major coupling risk.

[0076] like Figure 9 The chart shown is a bar graph comparing the fault indicators of this invention and the D1 (CN121352197A) method. The test scenario was extreme sea conditions plus early minor blade faults. The comparison of the four core indicators is as follows: fault identification accuracy: this invention 98.2%, D1 61.4%; minor fault detection rate: this invention 96.0%, D1 47.2%; false alarm rate: this invention 1.8%, D1 12.6%; overall identification rate: this invention 97.9%, D1 64.1%. D1 is a traditional offshore wind turbine evaluation scheme, which only targets the transmission system, without distinguishing operating conditions or network characteristics. The data clearly shows that this invention is superior to the existing technology in terms of accuracy, minor fault detection capability, and anti-interference (low false alarm), especially for difficult faults in the industry such as blade flutter and early wear, where the performance improvement is significant.

[0077] S6: Data Archiving and Closed-Loop Operation and Maintenance Decisions The entire process, including raw data, feature sets, fault results, and risk reports, is synchronously stored locally and uploaded to the remote cloud. The system automatically executes corresponding maintenance strategies based on the risk level: routine inspections are performed for low-risk situations; targeted offline testing is performed for general coupled risks; unit output is reduced and time-limited maintenance is implemented for higher coupled risks; and unit shutdown protection commands are triggered for major coupled risks. After completing a single cycle, the system automatically jumps to the data acquisition phase, achieving uninterrupted monitoring.

[0078] Example 2: This embodiment provides a fault-risk coupled global assessment system for grid-type offshore wind turbines. The system is applied to grid-type offshore wind turbines equipped with long flexible blades, highly flexible towers, and high-density electromechanical drive systems, and includes the following six functional modules: The data acquisition and preprocessing module is used to collect monitoring data on blades, towers, electromechanical transmission systems, and grid electrical parameters. Based on the current operating condition (normal, grid disturbance, or extreme sea state), it dynamically matches filtering strategies and data repair rules to perform data denoising and missing data compensation. Specifically, under normal operating conditions, fixed-window moving average filtering and linear interpolation are used for data denoising and missing data compensation; under grid disturbance conditions, variable-window Kalman filtering is used to suppress grid coupling noise, and missing data prediction compensation is performed based on historical operating condition datasets of the grid-connected units; under extreme sea states, a multi-level joint denoising algorithm is employed, while simultaneously marking noise interference intervals.

[0079] The feature mining module is used to simultaneously extract time-domain, frequency-domain, and entropy-domain features from the preprocessed monitoring data, and introduces grid-connected electromechanical coupling feature parameters to construct a joint feature set. The grid-connected electromechanical coupling features include three types of parameters: unit dynamic support current hysteresis, short-circuit current duration, and grid-connected oscillation amplitude; feature extraction is performed using db4 continuous wavelet transform.

[0080] The intelligent fault identification module incorporates a bidirectional long short-term memory network model trained under different operating conditions. Combined with dynamic identification thresholds, it distinguishes between single-component faults and multi-component coupled faults, and locates the fault type and physical location. The BiLSTM model is trained independently for normal conditions, power grid disturbances, and extreme sea conditions, with the dynamic identification thresholds automatically relaxed under power grid disturbance conditions and extreme sea conditions.

[0081] The fault propagation modeling module is used to build a fault-derived cascading risk propagation path model based on the structural correlation of wind turbine components and the fault propagation law, and to calculate the impact coefficient of a single fault spreading to the whole machine. This module sorts out the fault propagation paths of the three major components: blades, tower, and transmission system, calculates the impact coefficients and propagation probabilities of cascading faults from blade flutter to transmission overload, transmission wear to tower resonance, and tower deformation to blade yaw, and establishes a directed propagation topology graph.

[0082] The comprehensive risk assessment module integrates FMECA failure mode analysis, an improved DEMATEL correlation algorithm, and the AHP-entropy weighting method. It combines failure impact coefficients and marine environmental factors to calculate a comprehensive coupled risk index and classify risk levels. Specifically, the improved DEMATEL algorithm incorporates failure impact coefficients and marine environmental weight factors to correct the direct impact matrix. The AHP-entropy weighting method integrates subjective and objective weights. The comprehensive coupled risk index is calculated according to... The calculations categorize the units into four levels based on the R-value: safe, moderate coupling risk, high coupling risk, and significant coupling risk.

[0083] The closed-loop operation and maintenance decision module is used to automatically generate and execute differentiated operation and maintenance strategies based on risk levels: routine inspections are performed for safe levels; targeted offline detection is performed for general coupled risks; unit output is reduced and time-limited maintenance is carried out for higher coupled risks; and unit shutdown protection commands are triggered for major coupled risks.

[0084] The above modules are connected in sequence: the output of the data acquisition and preprocessing module is connected to the input of the feature mining module; the output of the feature mining module is connected to the input of the intelligent fault identification module; the output of the intelligent fault identification module is connected to the input of the fault propagation modeling module; the output of the fault propagation modeling module is connected to the input of the global risk assessment module; and the output of the global risk assessment module is connected to the input of the closed-loop operation and maintenance decision module, thus forming an automated cyclical assessment link from data acquisition to closed-loop operation and maintenance.

[0085] Example 3: This embodiment provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the fault-risk coupled global assessment method for grid-type offshore wind turbines as described in any of the above method embodiments.

[0086] The computer-readable storage medium may be any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical fiber, a portable compact disk read-only memory, an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0087] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A fault-risk coupled global assessment method for grid-type offshore wind turbines, characterized in that, Includes the following steps: S1: Operating condition classification and adaptive preprocessing: Based on the electromechanical-grid coupled operation of grid-connected units, three scenarios are classified: normal operating condition, grid disturbance operating condition, and extreme sea state. According to the current operating condition, the filtering strategy and data repair rules are dynamically matched to complete the data noise reduction and missing data compensation. S2: Multi-dimensional joint fault feature mining: Time domain, frequency domain, and entropy domain features are extracted simultaneously from the preprocessed monitoring data, and network electromechanical coupling feature parameters are introduced to construct a joint feature set; S3: Multi-condition adaptive intelligent fault identification: Based on a bidirectional long short-term memory network model trained for different operating conditions, combined with dynamic identification thresholds, it distinguishes between single component faults and multi-component coupled faults, and locates the fault type and physical location. S4: Fault-Risk Coupling Transmission Modeling: Based on the structural correlation of wind turbine components and the law of fault propagation, a fault-derived chain risk transmission path model is built to quantify the impact coefficient of a single fault spreading to the whole machine; S5: Comprehensive Risk Coupled Quantitative Assessment: Integrating FMECA failure mode analysis, improved DEMATEL correlation algorithm, and AHP-entropy weight combination weighting method, combined with failure impact coefficient and marine environmental factors, the computer group comprehensively couples the risk index and classifies the risk level; S6: Closed-loop operation and maintenance decision-making: Automatically generates differentiated operation and maintenance strategies based on risk level, and continuously executes the monitoring and evaluation process in a loop.

2. The fault-risk coupled global assessment method for grid-type offshore wind turbines according to claim 1, characterized in that, In step S1, the adaptive preprocessing for operating conditions specifically includes: for normal operating conditions, fixed window sliding mean filtering is used to reduce noise and linear interpolation is used to complete missing data; for grid disturbance operating conditions, variable window Kalman filtering is used to suppress grid coupling noise, and missing data prediction compensation is performed based on the historical operating condition dataset of the grid-connected units; for extreme sea conditions, a multi-level joint noise reduction algorithm is used, and noise interference intervals are marked. The formula for fixed-window sliding mean filtering noise reduction is: ； in, For the original number Frame monitoring data, For the filtered first Frame data, To fix the length of the sliding window; The formula for linear interpolation to complete missing data is: ； in, , For the valid data moments before and after the missing interval, , For the corresponding valid monitoring values, For the time when data is missing, The data is completed after interpolation; The state equation and observation equation of the variable window Kalman filter are as follows: ； ； in, for The system state vector at any given time. for Observe data in real time. , These are the state transition matrix and the observation matrix, respectively, with the window dynamically adjusted according to the intensity of the power grid disturbance. , These are process noise and observation noise, respectively. The hard threshold formula for the multi-level joint noise reduction algorithm is: ； in, For the first Layer wavelet decomposition coefficients, These are the coefficients after noise reduction. The threshold is adaptive and is dynamically updated based on turbulence intensity and salt spray level.

3. The fault-risk coupled global assessment method for grid-type offshore wind turbines according to claim 1, characterized in that: In step S2, in addition to the conventional features in the time domain, frequency domain, and entropy domain, the joint feature set adds network electromechanical coupling features, including three types of parameters: unit dynamic support current hysteresis, short-circuit current duration, and grid oscillation amplitude. The feature extraction is completed by using db4 continuous wavelet transform, and the conventional features and electromechanical coupling features are concatenated into a multi-dimensional joint feature set. The formula for the db4 continuous wavelet transform is: ； in, These are the wavelet transform coefficients. As a scale factor, The translation factor is... For the db4 mother wavelet function, This is the preprocessed time series data; The concatenation expression for the joint feature set is: ； in, For time-domain feature vectors, For frequency domain eigenvectors, For entropy domain eigenvectors, This is the characteristic vector of electromechanical coupling in the network.

4. The fault-risk coupled global assessment method for grid-type offshore wind turbines according to claim 1, characterized in that: In step S3, the bidirectional long short-term memory network model adopts an independent training mode for different operating conditions, and builds training datasets for normal, power grid disturbance, and extreme sea conditions respectively; at the same time, a dynamic recognition threshold is set, and the threshold tolerance range is automatically relaxed under power grid disturbance and extreme sea conditions. The formula for calculating the dynamic recognition threshold is: ； in, This is the baseline threshold for normal operating conditions. The disturbance factor is the operating condition factor under power grid disturbance conditions and extreme sea states. , This is the threshold relaxation factor; The memory cell update formula of the bidirectional long short-term memory network model is as follows: Input Gate: ; Forgotten Gate: ; Output gate: ; Memory cells: ; Hidden layer output: ; Bidirectional fusion output: ; in, It is the Sigmoid activation function. For element-wise multiplication, For vector concatenation, , These are the model weight matrix and bias vector, respectively, obtained through independent training for each work condition. As candidate memory cells, Input features at time t, For t Hidden layer output at time 1 For t 1. Memory of cell state at any moment Output of the feedforward long short-term memory network model; Output of the feedforward long short-term memory network model.

5. The fault-risk coupled global assessment method for grid-type offshore wind turbines according to claim 1, characterized in that: In step S4, the fault-risk coupling transmission modeling method is as follows: sort out the fault propagation paths of the three major components of blades, tower and transmission system, and establish a directed transmission topology graph; The formula for calculating the fault impact coefficient is as follows: ； in, For components Faulty components Influence coefficient, for After the fault occurred Operating parameters for Normal operating parameters; The formula for calculating the probability of fault propagation is: ； in, For the fault from the component Conducted to components The probability, This is the inherent propagation probability. For components Faulty components Influence coefficient, These are marine environmental factors.

6. The fault-risk coupled global assessment method for grid-type offshore wind turbines according to claim 5, characterized in that: In step S5, the improved DEMATEL algorithm, based on traditional correlation analysis, introduces fault impact coefficients and marine environmental weighting factors such as salt spray corrosion and turbulence intensity, and recalculates the influence and affected degree of each risk factor; the corrected direct influence matrix is ​​expressed as follows: ； in, For traditional DEMATEL, the direct influence matrix, This is the fault impact coefficient matrix. This is the marine environment weight matrix. This is the corrected direct influence matrix; The AHP-entropy weight combination weighting method calculates the comprehensive weight according to the following formula: ； in, For subjective weights in the analytic hierarchy process, For the objective weight of the entropy weight method, The weighted fusion coefficient; Comprehensive Coupling Risk Index The calculation formula is: ； in, For the first Risk indicator weights The severity of the fault, This represents the probability of fault propagation. This represents the total number of risk indicators.

7. The fault-risk coupled global assessment method for grid-type offshore wind turbines according to claim 6, characterized in that: In step S6, the unit is divided into four levels according to the comprehensive coupling risk index R: R∈(0,0.3) is the safety level, R∈(0.3,0.6) is the general coupling risk level, R∈(0.6,0.85) is the high coupling risk level, and R∈(0.85,1.0) is the major coupling risk level; a closed-loop operation and maintenance strategy is automatically generated according to the risk level: the safety level performs routine inspections; For general coupling risks, targeted offline detection is performed; for higher coupling risks, unit output is reduced and time-limited maintenance is required. Significant coupling risks trigger unit shutdown protection commands.

8. The fault-risk coupled global assessment method for grid-type offshore wind turbines according to claim 1, characterized in that: In step S1, the multi-source monitoring data includes blade load, torsion angle, and vibration data; tower strain, tilt angle, and vibration data; electromechanical transmission system temperature, meshing vibration, and acoustic vibration data; and grid interaction voltage, current, and inertia support response data of grid-type units. The data acquisition frame rate is uniformly set to 50Hz, and each frame of data is bound to a real-time operating condition label.

9. A fault-risk coupled global assessment system for grid-type offshore wind turbines, based on the fault-risk coupled global assessment method for grid-type offshore wind turbines as described in any one of claims 1-8, characterized in that, include: The data acquisition and preprocessing module is used to collect monitoring data of blades, towers, electromechanical transmission systems and grid electrical parameters. Based on the current operating conditions, grid disturbance conditions or extreme sea conditions, it dynamically matches filtering strategies and data repair rules to complete data noise reduction and missing data compensation. The feature mining module is used to simultaneously extract time-domain, frequency-domain, and entropy-domain features from the preprocessed monitoring data, and introduce network electromechanical coupling feature parameters to construct a joint feature set; The intelligent fault identification module has a built-in bidirectional long short-term memory network model trained under different working conditions. Combined with dynamic identification thresholds, it can distinguish between single component faults and multi-component coupled faults, and locate the fault type and physical location. The fault propagation modeling module is used to build a fault-derived chain risk propagation path model based on the structural correlation of wind turbine components and the fault propagation law, and to calculate the impact coefficient of a single fault spreading to the whole machine. The comprehensive risk assessment module integrates FMECA failure mode analysis, improved DEMATEL correlation algorithm, and AHP-entropy weight combination weighting method, and combines failure impact coefficient with marine environmental factors to calculate a comprehensive coupled risk index and classify risk levels. The closed-loop operation and maintenance decision-making module is used to automatically generate and execute differentiated operation and maintenance strategies based on risk levels; The above modules are connected in sequence to form an automated cyclical evaluation link from data acquisition to closed-loop operation and maintenance.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the fault-risk coupling global assessment method for grid-type offshore wind turbines as described in any one of claims 1 to 8.

Images