Offshore wind turbine generator system-level health degree evaluation method based on dynamic weight distribution

Abstract

The invention belongs to the technical field of offshore wind power operation and maintenance and reliability engineering, and discloses an offshore wind turbine generator system-level health degree evaluation method based on dynamic weight distribution, and the method comprises the following steps: S1, collecting and preprocessing the multi-source data of an offshore wind turbine generator; s2, constructing a layered health degree evaluation index system; s3, fusing the mechanism model, the data driving model and the statistical reliability model, and constructing a subsystem degradation model to obtain a subsystem health degree; s4, performing weight fusion in combination with the static weight and the dynamic correction factor to obtain a dynamic weight; s5, fusing the subsystem health degree and the dynamic weight to obtain a system-level health degree; and S6, performing confidence interval sorting and risk grade division according to the system-level health degree. According to the evaluation method provided by the invention, through fusion of a mechanism model, a data driving model and a statistics model, the physical interpretability is reserved, the fitting capability of nonlinearity and coupling degradation is improved, and the misjudgment and missed judgment risks are reduced.

Description

Technical Field

[0001] This invention relates to the field of offshore wind power operation and maintenance and reliability engineering technology, and in particular to a method for evaluating the system-level health of offshore wind turbines based on dynamic weight allocation. Background Technology

[0002] As offshore wind farms continue to expand, wind turbines are exposed to harsh environments such as high humidity, strong winds and waves, and high salt spray for extended periods, making key components susceptible to corrosion, fatigue, and failure. Traditional health assessments often focus on single components or subsystems, lacking a system-level perspective, and their static weighting methods fail to reflect the dynamic changes in health under different operating conditions.

[0003] The existing technology has the following shortcomings:

[0004] The indicator system is incomplete: it lacks a comprehensive health assessment indicator that covers mechanical, electrical, control, structural, and environmental adaptability.

[0005] The weight allocation method is singular: it mostly uses fixed weights, which cannot reflect the dynamic changes in the importance of subsystems under different operating conditions;

[0006] Degradation modeling is crude: it is mostly based on empirical methods and does not make full use of multi-source monitoring data;

[0007] Insufficient system-level integration: The lack of a comprehensive integration method for health status makes it difficult to provide accurate support for predictive maintenance and operation and maintenance decisions.

[0008] Therefore, there is an urgent need for a new method that can realize the construction of system-level indicators, subsystem degradation modeling, dynamic weight allocation and health fusion calculation, so as to improve the scientificity and reliability of offshore wind turbine operation and maintenance. Summary of the Invention

[0009] To address the issues of incomplete overall evaluation indicators, simplistic weight allocation methods, coarse degradation modeling, and insufficient system-level integration in existing technologies, this invention proposes a system-level health evaluation method for offshore wind turbines based on dynamic weight allocation. This method enables a comprehensive, dynamic, and reliable quantitative assessment of the system-level health status of the turbines under complex marine environments and variable operating conditions.

[0010] This invention is achieved through the following technical solution, including the following steps:

[0011] S1. Collect and preprocess multi-source data from offshore wind turbines;

[0012] S2. Construct a hierarchical health evaluation index system, which consists of a first-level system-level health, a second-level subsystem health, and a third-level observable index.

[0013] S3, integrating the mechanism model, data-driven model and statistical reliability model, constructs a subsystem degradation model to obtain the subsystem health;

[0014] S4. Combine static weights and dynamic correction factors to perform weight fusion and obtain the dynamic weight of the health of each subsystem.

[0015] S5. Combine the health of subsystems and dynamic weights to obtain the system-level health.

[0016] S6. Sort confidence intervals and classify risk levels based on system-level health.

[0017] As a further preferred option, the specific steps of step S3 are as follows:

[0018] S3.1 Mechanism Model: Based on fatigue accumulation, thermo-mechanical coupling, lubricating film thickness / abrasive particle formation mechanism, and load frequency domain response, the degradation amount is obtained. ;

[0019] S3.2 The data-driven model obtains degradation representations based on time-series learning, anomaly detection, or graph learning. ;

[0020] S3.3 The statistical reliability model uses a degradation process combined with Bayesian updates to obtain the remaining lifetime. ;

[0021] S3.4, The degradation amount obtained in step S3.1 The degradation characterization obtained in step S3.2 and the remaining lifetime obtained in step S3.3 The fusion degradation amount is obtained by fusion, and the formula is as follows:

[0022] ;

[0023] ;

[0024] In the formula, This represents the fusion degradation of the i-th subsystem; This represents the degradation weight of the i-th subsystem; This represents the degradation amount of the i-th subsystem; This represents the degradation representation weight of the i-th subsystem; This represents the degradation characteristic of the i-th subsystem; This represents the remaining lifetime weight of the i-th subsystem; This represents the remaining lifetime of the i-th subsystem;

[0025] S3.5. Based on the fusion degradation obtained in step S3.4, the subsystem health is obtained through monotonic health mapping, as shown in the following formula:

[0026] ;

[0027] In the formula, This represents the health status of the i-th subsystem; Indicates the mapping relationship; This represents the Sigmoid function or a piecewise linear mapping; Indicates the offset parameter; This represents the scaling parameter.

[0028] As a further preferred option, the specific steps of step S4 are as follows:

[0029] S4.1 Setting Initial Weights ;

[0030] S4.2. Based on the deviation between the current environmental operating conditions and the normal environmental operating conditions of the offshore wind turbine, construct the operating condition factor R. i (t);

[0031] S4.3. Based on the sensitivity of the subsystem to the system health, construct a sensitivity factor S. i (t);

[0032] S4.4 Construct a degradation rate factor based on the degradation rate of the subsystem. ;

[0033] S4.5, Based on the operating condition factor R i (t), Sensitivity Factor S i (t) and degradation rate factor D i (t), construct the dynamic correction factor, as shown in the following formula:

[0034] ;

[0035] In the formula, This represents the dynamic correction factor of the i-th subsystem; These represent the weight parameters for the operating condition factor, sensitivity factor, degradation rate factor, and abnormal warning factor, respectively. Represents the abnormal alarm / critical event triggering coefficient of the i-th subsystem;

[0036] S4.6. Based on the static weights and the dynamic correction factor, the dynamic weights are obtained, as shown in the following formula:

[0037] ;

[0038] ;

[0039] In the formula, represents the dynamic weight; j represents the total number of subsystems.

[0040] As a further preferred embodiment, the formula for obtaining the system-level health status in step S5 is as follows:

[0041] ;

[0042] ;

[0043] In the formula, Indicates system-level health; This represents the dynamic weight of the i-th subsystem; Represents the health of the i-th subsystem; Var represents the overall system-level health. The variance; It is a weighted column vector. ,in It is the number of subsystems; This represents the covariance of the subsystem health estimate.

[0044] As a further preferred embodiment, the smoothing and constraint formulas applied to the dynamic weights in step S3 are as follows:

[0045] ;

[0046] ;

[0047] In the formula, Indicates the smoothing parameter; These represent the lower and upper limits of the dynamic weights, respectively. Indicates the sampling time interval.

[0048] As a further preferred embodiment, the multi-source data in step S1 includes the operating data, environmental data, and maintenance data of the offshore wind turbine; the operating data of the offshore wind turbine includes the speed, power, torque, temperature, vibration, and load of the blades, gearbox, and generator; the environmental data of the offshore wind turbine includes the wind speed / direction of the sea breeze, the wave height, the salt spray level, the humidity, and the air temperature; and the maintenance data includes fault, replacement, and maintenance records.

[0049] As a further preferred option, the preprocessing method for multi-source data in step S1 includes denoising, outlier identification, missing value imputation, and dimensional normalization.

[0050] As a further preferred option, the subsystem health in step S2 includes the rotor system, transmission system, generator and power electronics system, control and communication system, floating platform and tower system, and environmental adaptability.

[0051] As a further preferred option, the observable indicators in step S2 include blade fatigue damage accumulation, gearbox lubrication deterioration, winding temperature rise margin, pitch response error, foundation stress utilization rate, and salt spray corrosion index.

[0052] As a further preferred embodiment, the following steps are included after step S6:

[0053] S7, Based on system-level health and remaining service life, RUL i Prioritize inspections, optimize spare parts plans and time windows, and when When an alert is triggered, among which The failure threshold representing the system-level health status; It is the future range that the predictive horizon definition is concerned with; This serves as a warning threshold for the probability of health risks.

[0054] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0055] 1. The evaluation method provided by this invention, through a hierarchical index system and unified caliber, covers mechanical, electrical, control / communication, structural and environmental adaptability, ensuring the comparability and traceability of health across subsystems.

[0056] 2. The evaluation method provided by this invention integrates three models: mechanism, data-driven, and statistics. This not only preserves physical interpretability but also improves the fitting ability to nonlinearity and coupling degradation, thereby reducing the risk of misjudgment and missed judgment.

[0057] 3. The dynamic weight allocation in the evaluation method provided by this invention is adaptively adjusted according to the working conditions, sensitivity and degradation rate, and the weight jitter is suppressed by smoothing and boundary constraints, so as to balance sensitivity and stability.

[0058] 4. The evaluation method provided by this invention gives a confidence interval and risk level through system-level health, and the early warning threshold is upgraded from "point value" to "probability + interval", which is more in line with the actual operation and maintenance decision.

[0059] 5. The evaluation method provided by this invention supports online / offline deployment and is compatible with mainstream SCADA / CMS platforms; the subsystem model and weight strategy can be modularly replaced and expanded according to aircraft type and sea area, which facilitates large-scale promotion.

[0060] 6. The evaluation method provided by this invention directly uses system-level health H in the optimization of maintenance priority, spare parts, and window periods. xys Remaining Useful Life (RUL) i This, along with risk threshold logic, forms a closed-loop decision-making process, improving availability and reducing total lifecycle costs. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of the overall process of the health assessment method of the present invention.

[0062] Figure 2 This is a schematic diagram illustrating the hierarchical structure of the indicator system in the health evaluation method of the present invention.

[0063] Figure 3 This is a schematic diagram of the subsystem degradation modeling method of the present invention, which includes the fusion of mechanistic model, data-driven model and statistical model.

[0064] Figure 4 This is a schematic diagram of the dynamic weight allocation mechanism in the health assessment method of the present invention.

[0065] Figure 5 This is a schematic diagram of the system health calculation and risk classification results in the health assessment method of the present invention. Detailed Implementation

[0066] The advantages and features of the present invention will be illustrated and explained by the following non-limiting description of preferred embodiments, which are given by way of example only with reference to the accompanying drawings.

[0067] like Figure 1 As shown, this invention proposes a system-level health assessment method for offshore wind turbines based on dynamic weight allocation, which can achieve a comprehensive, dynamic, and reliable quantitative assessment of the system-level health status of the turbines under complex marine environments and variable operating conditions. Figure 2 As shown, this invention further unifies the mechanical, electrical, control / communication, structural, and environmental adaptability dimensions through a hierarchical index system. For example... Figure 3 As shown, this invention introduces a subsystem degradation modeling framework that integrates mechanistic models, data-driven models, and statistical models, forming a monotonic and interpretable health mapping. For example... Figure 4 As shown, this invention designs a dynamic weight allocation mechanism that enables the weights of each subsystem to adaptively evolve with operating conditions, sensitivity, and degradation rate. When the system is integrated, it outputs a health confidence interval and risk classification warning, which can directly support predictive maintenance and repair priority decision-making.

[0068] like Figure 1 As shown, this invention proposes a system-level health assessment method for offshore wind turbines based on dynamic weight allocation, which specifically includes the following steps:

[0069] Step 1: Collect and preprocess multi-source data from offshore wind turbines, including operational data, environmental data, and maintenance data of offshore wind turbines;

[0070] The operating data of the offshore wind turbine includes data such as the speed, power, torque, temperature, vibration, and load of the blades, gearbox, and generator.

[0071] The environmental data for the offshore wind turbines includes wind speed / direction, wave height, salt spray level, humidity, and temperature.

[0072] The maintenance and repair data includes fault, replacement, and maintenance records.

[0073] Preprocessing methods for multi-source data include denoising, outlier identification, missing value imputation, and dimensional normalization to form a usable dataset. Missing value imputation can be achieved using the K-Nearest Neighbors (KNN) algorithm or temporal interpolation.

[0074] This invention performs spatiotemporal alignment on multi-source data; employs robust metric (IQR / MAD) to suppress outliers; uses constraint-preserving temporal interpolation for missing data intervals to maintain reasonable energy and extrema; and ensures data quality and robustness.

[0075] Step 2: Construct a hierarchical health evaluation index system, which consists of a first-level system-level health index, a second-level subsystem health index, and a third-level observable index.

[0076] The evaluation of the health of the subsystems includes the rotor system, transmission system, generator and power electronics system, control and communication system, floating platform and tower system, and environmental adaptability.

[0077] The observable indicators include blade fatigue damage accumulation, gearbox lubrication deterioration, winding temperature rise margin, pitch response error, foundation stress utilization rate, and salt spray corrosion index.

[0078] like Figure 2 As shown, the indicator system established in this invention includes three levels:

[0079] Primary indicator: System-level health;

[0080] Secondary indicators: six subsystems: rotor, drive train, generator and power electronics, control and communication, floating platform and tower, and environmental adaptability;

[0081] Level 3 indicators: These are the key performance indicators of each subsystem, such as blade fatigue damage accumulation, gearbox lubrication deterioration, winding temperature rise margin, foundation stress utilization rate, and salt spray corrosion index.

[0082] The health status of each subsystem is derived from the data of the three-level indicators, and the system-level health status is evaluated by combining the health status of each subsystem with its corresponding weight.

[0083] Step 3: Integrate the mechanistic model, data-driven model, and statistical reliability model to construct a subsystem degradation model and obtain the subsystem health.

[0084] Step 3.1: Mechanism Model. Based on fatigue accumulation, thermo-mechanical coupling, lubricating film thickness / abrasive grain formation mechanism, load frequency domain response, etc., establish an interpretable degradation amount. .

[0085] The fatigue accumulation in this invention can be obtained from Miner linear damage.

[0086] Step 3.2: The data-driven model obtains the degradation representation based on time-series learning, anomaly detection, or graph learning. .

[0087] The temporal learning in this invention is a Long Short-Term Memory (LSTM) neural network model or a Temporal Convolutional Network (TCN) model. Anomaly detection is performed by an autoencoder, and graph learning is performed by the subsystem coupling topology.

[0088] Step 3.3: The statistical reliability model uses a degradation process combined with Bayesian updates to obtain the remaining lifetime. .

[0089] The statistical reliability model of this invention mainly provides a probabilistic framework for the degradation of a subsystem, describing its uncertainty over time. Different degradation models should be selected for different degradation characteristics. For example, the gamma distribution is mainly for monotonically cumulative damage processes, such as fatigue accumulation and wear, which are degradations that only increase and never decrease. The Weibull process is used to describe the life distribution and failure rate changes, such as mechanical damage and crack accumulation. The Gompertz curve is used to describe the physical creep and aging of components, such as the degradation processes of composite materials such as towers and blades. The selection is not limited to these three.

[0090] Step 3.4: Calculate the degradation amount obtained in Step 3.1. The degradation characterization obtained in step 3.2 and the remaining lifetime obtained in step 3.3 The fusion degradation amount is obtained by fusion, and the formula is as follows:

[0091] ;

[0092] ;

[0093] In the formula, This represents the fusion degradation of the i-th subsystem; This represents the degradation weight of the i-th subsystem; This represents the degradation amount of the i-th subsystem; This represents the degradation representation weight of the i-th subsystem; This represents the degradation characteristic of the i-th subsystem; This represents the remaining lifetime weight of the i-th subsystem; This represents the remaining lifetime of the i-th subsystem.

[0094] Step 3.5: Based on the fusion degradation obtained in Step 3.4, the subsystem health is obtained through monotonic health mapping, as shown in the following formula:

[0095] ;

[0096] In the formula, This represents the health status of the i-th subsystem; Indicates the mapping relationship; This represents the Sigmoid function or piecewise linear mapping, ensuring that health does not increase as degradation increases, i.e., monotonicity. Indicates the offset parameter; This represents the scaling parameter.

[0097] like Figure 3 As shown, this invention performs subsystem degradation in the subsystem health assessment, and the subsystem degradation adopts a fusion of three types of models:

[0098] Mechanistic models, such as fatigue accumulation models, thermo-mechanical coupling models, and lubrication film thickness models, are used to quantitatively describe the physical degradation process and degradation trend of key components in a subsystem.

[0099] Data-driven models, such as deep learning-based time-series prediction models (LSTM models, TCN models) and anomaly detection models, are used to capture complex nonlinear degradation patterns, primarily based on the time-series characteristics of sensors.

[0100] Statistical reliability models, such as Weibull distribution process, Gamma distribution process, or Bayesian update method, are used to estimate remaining lifetime and failure probability.

[0101] The fusion degradation of the subsystem is obtained through evidence synthesis or weighted fusion methods, and then transformed into health level using a monotonic health mapping function. .

[0102] Figure 3 Taking the Sino-Israeli gearbox as an example:

[0103] The fatigue accumulation model was selected as the mechanistic model. Based on the SN curve of the gear material and the design manual, the fatigue life under different load levels was determined.

[0104] Real-time acquisition of torque sensor signals, and counting of the number of cycles n at each load level. k .

[0105] According to Miner's linear damage rule, the theoretical damage degree D output by the computer model is... m :

[0106] ;

[0107] In the formula, the theoretical damage degree D m It is a quantity that accumulates from 0 (no damage) to 1 (theoretical failure); n k N represents the number of cycles for the k-th load level; k This represents the theoretical fatigue life value obtained from the material SN curve and design manual.

[0108] The data-driven model selected is a time series forecasting model based on LSTM. LSTM-based time series forecasting models excel at learning complex degradation trends from historical data. The general process of the model is as follows:

[0109] Training phase: The time series prediction model based on the LSTM model is trained using data from the early stage of the gearbox's entire life cycle. The data collected in the early stage of the gearbox's entire life cycle is regarded as healthy data. The time series prediction model based on the LSTM model is trained using the healthy data of the gearbox, so that it can predict the root mean square value (RMS) of vibration at future times with high accuracy, and the trained time series prediction model based on the LSTM model is obtained.

[0110] Online application: Input real-time vibration features into a pre-trained LSTM-based time-series prediction model to obtain predicted values.

[0111] Calculate the degradation index: Calculate the relative error between the predicted and observed values. As the gearbox degrades, the prediction error systematically increases. After processing this error sequence through moving average filtering and normalization, the output is the degradation index D of the data-driven model. d The range is between [0, 1].

[0112] The statistical reliability model selected is a failure probability model based on Bayesian updates. The general process of the model is as follows:

[0113] First, a priori model is established: based on historical failure data of a group of gearboxes of the same model, an initial Weibull life distribution is fitted.

[0114] The outputs of the aforementioned mechanistic model and data-driven model are used as new evidence to update the parameters of the lifetime distribution using Bayes' theorem. When the theoretical damage degree D of the mechanistic model... m Or the degradation index D of the data-driven model d When the number of cases increases rapidly, the probability of failure should increase accordingly.

[0115] The statistical model outputs the failure probability D at the current moment. f Given all the current evidence, this is the probability that the device will fail in the next operating cycle.

[0116] Consistency test: Compare the outputs D of the three models m D d D f Based on the consistency threshold setting, if the condition is within the threshold, it is considered "consistent"; otherwise, it is considered "inconsistent".

[0117] When the models are consistent, a fixed-weighted average is used. Since the data-driven model better reflects actual working conditions, it is given a higher weight; the mechanistic and statistical models have lower weights. The fusion degradation after fusion is:

[0118] ;

[0119] in, Indicates the amount of fusion degradation in the gearbox; They are respectively The weights are all non-negative parameters.

[0120] When models are inconsistent: a dynamic weighting based on confidence level is used. The volatility (variance) of the recent outputs of each model is calculated; the lower the volatility, the higher the confidence level, and the greater the weight. The weights are then dynamically calculated before fusion.

[0121] The fusion degradation is expressed by a monotonically decreasing function. Mapped to intuitive subsystem health The formula is as follows:

[0122]

[0123] The subsystem health H1 has a range of [0, 1], and a higher value indicates a healthier subsystem.

[0124] This invention establishes a consistent normalized standard for each of the three-level indicators, standardizing it to the highest possible value. Specifically, in the subsystem health degradation model, the deviation from the final system health residual is represented by a monotonically decreasing function; the smaller the difference, the higher the final output, and thus the higher the health. Physical threshold anchoring (such as temperature rise margin and fatigue damage limit) ensures comparability across subsystems, achieving a unified health mapping and standard.

[0125] In step 3 of this invention, the DS evidence theory can also be incorporated, triggering model retraining or sensor verification when the conflict coefficient is higher than a threshold.

[0126] Step 4: Combine static weights and dynamic correction factors to perform weight fusion and obtain the final weight of the health of each subsystem, i.e., the dynamic weight;

[0127] Step 4.1: Set initial weights Initial weights These are static weights, determined by the analytic hierarchy process or expert experience.

[0128] Step 4.2: Based on the deviation between the current environmental operating conditions and the normal environmental operating conditions of the offshore wind turbine, construct the operating condition factor R. i (t);

[0129] Step 4.3: Construct a sensitivity factor S based on the subsystem's sensitivity to system health. i (t);

[0130] Step 4.4: Construct a degradation rate factor based on the degradation rate of the subsystem. ;

[0131] Step 4.5: Based on the operating condition factor R i (t), Sensitivity Factor S i (t) and degradation rate factor D i (t), construct the dynamic correction factor, as shown in the following formula:

[0132] ;

[0133] In the formula, This represents the dynamic correction factor of the i-th subsystem; These represent the weight parameters of the operating condition factor, sensitivity factor, degradation rate factor, and abnormal warning factor, respectively, and are all non-negative parameters; This represents the abnormal alarm / critical event trigger coefficient of the i-th subsystem, used to indicate whether the i-th subsystem has an abnormal alarm, such as a positive crack detection or a sudden increase in the ferrography of the oil.

[0134] The graph model in the data-driven model of this invention can abstract the coupling of subsystems into a graph, and use a graph convolutional network (GCN) to learn linkage degradation. It can be derived from graph attention scores.

[0135] Step 4.6: Based on the static weights and the dynamic correction factor, obtain the dynamic weights using the following formula:

[0136] ;

[0137] ;

[0138] In the formula, This represents the dynamic weight, i.e., the weight of the i-th subsystem after dynamic adjustment; Let represent the static weight of the i-th subsystem, i.e., the initial weight; j represents the total number of subsystems, i.e., i = 1, 2, 3, ..., j.

[0139] By normalizing and using boundary constraints, we ensure that the weight changes are smooth and that the sum is 1.

[0140] To avoid jitter and extreme bias in dynamic weights, smoothing and constraints can be applied to the dynamic weights, as shown in the following formula:

[0141] ;

[0142] ;

[0143] In the formula, Indicates the smoothing parameter; These represent the lower and upper limits of the dynamic weights, respectively. This represents the time interval, i.e., the period for system state sampling and weight updates.

[0144] like Figure 4 As shown, the weight calculation of this invention introduces operating condition factors, sensitivity factors, degradation rate factors and abnormal alarm factors on the basis of static weights.

[0145] Static weights are determined by the analytic hierarchy process (AHP) or expert experience. Operating condition factors are related to the environmental conditions of the offshore wind turbine, which include wind speed, sea state, temperature, and humidity. Operating condition factors are obtained by comparing the deviation of the current operating condition from the normal operating condition. i (t) represents the sensitivity of the i-th subsystem to the system health status, determined by the subsystem vibration, temperature, and other signals detected in real time by sensors, reflecting the operating status of the offshore wind turbine. The degradation rate factor is based on historical statistical data; for equipment with a higher degradation rate corresponding to a certain indicator, its weight should be increased, and it is derived according to the magnitude of the degradation rate.

[0146] The weights are dynamically adjusted over time, reflecting the importance of different subsystems under different operating conditions and enabling a rapid response to critical fault signals.

[0147] Step 5: Combine the subsystem health score and dynamic weights to obtain the system-level health score, as shown in the following formula:

[0148] ;

[0149] ;

[0150] In the formula, Indicates system-level health; The dynamic weight of the i-th subsystem is obtained from step 4; The health of the i-th subsystem is represented by Var, obtained from step 3; Var represents the health of the entire system. The variance is used to indicate the range of fluctuations or uncertainty of the system health assessment value; It is a weighted column vector. ,in It is the number of subsystems; The covariance of the subsystem health estimate is represented by the model residuals, sensor noise, and posterior variance.

[0151] This invention transfers the posterior variance of subsystem health to system-level health through linear approximation or Monte Carlo sampling, which is used for confidence interval and risk communication.

[0152] Step 6: Sort confidence intervals and classify risk levels based on system-level health;

[0153] The system-level health confidence intervals and risk levels are shown in Table 1:

[0154] Table 1

[0155] grade Health indicators Running status Running status description <![CDATA[G1]]> <![CDATA[H sys ≥0.85]]> healthy The system is in high health and the equipment is operating safely. <![CDATA[G2]]> <![CDATA[0.70≤H sys <0.85]]> good The system is in good health and the equipment is relatively safe. <![CDATA[G3]]> <![CDATA[0.55≤H sys <0.70]]> qualified The system health status is acceptable, but the equipment is not entirely safe. <![CDATA[G4]]> <![CDATA[H sys <0.55]]> Danger The system health is low, and the relevant equipment needs to be inspected and repaired.

[0156] Step 7: Based on system-level health and remaining useful life (RUL) i (where i represents a subsystem), perform inspection priority ranking, spare parts planning, and window period optimization. When an alert is triggered, among which It is a probability value that represents the future. At any given time, the overall health of the system is below the threshold. The probability of. The failure threshold representing the system-level health status. It is the future range that the definition of the predictive horizon concerns. This serves as a warning threshold for the probability of health risks.

[0157] like Figure 5 As shown, this invention obtains a system-level health curve by weighted fusion of subsystem health, and provides the health confidence interval and risk level.

[0158] When the health status falls below the preset threshold or there is a significant risk of a decline in the future predicted value, the system will automatically trigger an early warning, prompting maintenance personnel to arrange for maintenance and spare parts preparation in advance.

[0159] The results can be directly applied to the operation and maintenance management system to achieve predictive maintenance and repair decision optimization, thereby improving the availability of wind turbine units and reducing operation and maintenance costs.

[0160] This invention employs recursive Bayesian or Kalman / particle filtering to... and Real-time updates, weighting Offline calibration and online fine-tuning are performed to constrain the learning rate and prevent overfitting.

[0161] In addition to the above embodiments, the present invention may have other implementation methods. All technical solutions formed by equivalent substitution or equivalent transformation fall within the protection scope claimed by the present invention.

Claims

1. A method for offshore wind turbine system-level health assessment based on dynamic weight allocation, characterized in that: The method comprises the following steps: S1, collecting and preprocessing multi-source data of offshore wind turbine; S2, constructing a hierarchical health degree evaluation index system, including system-level health degree, sub-system health degree and observable index; S3, fusing mechanism model, data-driven model and statistical reliability model to construct a subsystem degradation model and obtain the subsystem health degree; S4, combining static weight and dynamic correction factor to fuse the weight and obtain the dynamic weight of each subsystem health degree; S5, fusing the subsystem health degree and the dynamic weight to obtain the system-level health degree; S6, sorting the confidence interval and dividing the risk level according to the system-level health degree.

2. The offshore wind turbine system-level health assessment method based on dynamic weight distribution according to claim 1, characterized in that: The specific steps of step S3 are as follows: S3.1, Mechanism model according to fatigue accumulation, thermal-mechanical coupling, lubrication film thickness / abrasive particle generation mechanism, load frequency domain response, get degradation quantity ; S3.2, the data-driven model obtains the degradation characterization according to time series learning, anomaly detection or graph learning ; S3.3, The statistical reliability model adopts degradation process, combined with Bayesian update, to obtain the remaining life ; S3.4, the degradation quantity obtained in step S3.1 , the degradation representation obtained in step S3.2 , and the residual life obtained in step S3.3 , the fusion degradation quantity is obtained by fusion, and the formula is as follows: ； ； wherein, represents a fusion degradation amount of the i-th subsystem; represents a degradation amount weight of the i-th subsystem; represents a degradation amount of the i-th subsystem; represents a degradation representation weight of the i-th subsystem; represents a degradation representation of the i-th subsystem; represents a residual life weight of the i-th subsystem; represents a residual life of the i-th subsystem; S3.5, obtaining the subsystem health degree by monotonic health mapping according to the fusion degradation amount obtained in step S3.4, and the formula is as follows: ； wherein represents the health of the i-th subsystem; represents a mapping relationship; represents a Sigmoid function or a piecewise linear mapping; represents an offset parameter; represents a scaling parameter.

3. The offshore wind turbine system-level health assessment method based on dynamic weight distribution according to claim 2, characterized in that: The specific steps of step S4 are as follows: S4.1, setting initial weights ; S4.2, constructing a working condition factor R according to the deviation degree of the current environmental working condition of the offshore wind turbine from the normal environmental working condition range i (t); S4.

3. Constructing a sensitivity factor S based on the subsystem's sensitivity to system health i (t); S4.

4. Constructing a degradation rate factor based on the degradation rate of the subsystem ; S4.5, according to the working condition factor R i (t), a sensitivity factor S i (t), and a degradation factor D i (t), a dynamic correction factor is constructed, and the formula is as follows: ； wherein, represents a dynamic correction factor of the i-th subsystem; respectively represent weight parameters of the operating condition factor, the sensitivity factor, the degradation rate factor, and the abnormality warning factor; represents an abnormality warning / critical event trigger coefficient of the i-th subsystem. S4.6, obtaining the dynamic weight according to the static weight and the dynamic correction factor, and the formula is as follows: ； ； wherein denotes the dynamic weight; j denotes the total number of subsystems.

4. The offshore wind turbine system-level health assessment method based on dynamic weight distribution according to claim 3, characterized in that: The formula for obtaining the system-level health degree in step S5 is as follows: ； ； wherein represents the system level health; represents the dynamic weight of the i-th subsystem; represents the i-th subsystem health; Var the variance of the system level health; is a weight column vector, wherein is the number of subsystems; represents the subsystem health estimate covariance.

5. The offshore wind turbine system-level health assessment method based on dynamic weight distribution according to claim 4, characterized in that: The smoothing and constraint formula for the dynamic weight in step S3 is as follows: ； ； In the formula, represents a smoothing parameter; respectively represent the lower and upper limits of the dynamic weight; represents a sampling time interval.

6. The offshore wind turbine system-level health assessment method based on dynamic weight distribution according to claim 4, characterized in that: The multi-source data in step S1 includes operation data, environmental data and maintenance data of offshore wind turbine; the operation data of offshore wind turbine includes the speed, power, torque, temperature, vibration and load of blades, gearboxes and generators; the environmental data of offshore wind turbine includes the wind speed / direction of sea wind, wave height, salt fog degree, humidity and air temperature; the maintenance data includes fault, replacement and maintenance records.

7. The offshore wind turbine system-level health assessment method based on dynamic weight distribution according to claim 6, characterized in that: The preprocessing method of multi-source data in step S1 includes denoising, outlier identification, missing data filling and dimensionless normalization processing.

8. The offshore wind turbine system-level health assessment method based on dynamic weight distribution of claim 4, wherein: The subsystem health degree in step S2 includes rotor system, transmission system, generator and power electronic system, control and communication system, floating platform and tower system, and environmental adaptability.

9. The offshore wind turbine system-level health assessment method based on dynamic weight distribution of claim 4, wherein: The observable index in step S2 includes blade fatigue damage accumulation, gearbox lubrication degradation degree, winding temperature rise margin, variable pitch response error, foundation stress utilization rate and salt fog corrosion index.

10. The offshore wind turbine system-level health assessment method based on dynamic weight distribution according to any one of claims 1 to 9, characterized in that: The step S6 further comprises the following steps: S7, prioritizing the inspection, spare parts planning and window optimization based on the system level health and the remaining useful life RUL i , triggering an early warning when , wherein represents a failure threshold of the system level health; is a future horizon defined by the prediction horizon; is an early warning threshold of the health risk probability.

Images