Comprehensive evaluation method for health state of fan complete machine based on fuzzy analytic hierarchy process

Abstract

The application discloses a fan whole machine health state comprehensive evaluation method based on a fuzzy analytic hierarchy process, comprising the following steps: through the construction of original health index collection, index edge distribution identification and correlation analysis, Copula function coupling modeling and fuzzy analytic hierarchy process fusion output, the complex relationship between multiple source health indexes is realized, and the limitations of traditional fuzzy analytic hierarchy process in index independence assumption, subjective weight setting and single evaluation output are broken through; through introducing a coupling strength matrix into the fuzzy analysis process, a mechanism innovation of dynamic adjustment of health weight is realized; through generating a grade interval by using historical health evaluation data, adaptive optimization of health label output is realized; and the accuracy and sensitivity of fan whole machine health evaluation are effectively improved.

Description

Technical Field

[0001] This invention relates to the technical field of health management and intelligent assessment of the operating status of wind power equipment, and in particular to a comprehensive evaluation method for the overall health status of wind turbines based on fuzzy hierarchical analysis. Background Technology

[0002] As wind power accounts for an increasing proportion of the global renewable energy system, the operational reliability and health management capabilities of wind turbine units have become core issues in the construction of operation and maintenance systems. This is especially true for wind turbines deployed in large-capacity, high-altitude, or complex wind-condition areas, where the operating environment is even harsher, leading to performance degradation or failure of critical components such as the main shaft, gearbox, and generator, severely impacting the overall power generation efficiency and safety. Therefore, in recent years, extensive research has been conducted on wind turbine health assessment based on multi-source operational data. This research utilizes signals such as temperature, vibration, and load to construct a health indicator system, and combines methods such as the analytic hierarchy process (AHP), fuzzy logic reasoning, and Bayesian networks to perform multi-dimensional quantitative modeling of wind turbine health status. However, traditional fuzzy AHP methods rely on manual experience to determine the membership weights of health indicators. The weighted hierarchical structure is ill-suited to the dynamic correlations and nonlinear couplings among health indicators in actual operating environments. During wind turbine operation, various indicators (such as vibration amplitude and load impact) often exhibit complex statistical dependencies. Ignoring these dependencies and relying solely on subjective judgment or linear weighting to assess the overall health status can easily lead to model assumption biases and fail to accurately reflect the actual health level. Furthermore, while some studies have attempted to introduce data-driven methods to enhance the objectivity of assessment models, they have not yet resolved the continuity issue of the key technology chain: "modeling nonlinear couplings among health indicators - integrating with fuzzy hierarchical structures - output level decision-making." Therefore, there is still an urgent need to propose a whole-machine-level health assessment method that integrates the ability to model multi-source indicator correlations with the advantages of fuzzy hierarchical analysis reasoning.

[0003] CN114417704A discloses a method for assessing the health of wind turbine generators based on an improved stacked autoencoder network. The method proposes to extract latent features from sensor data using an improved stacked autoencoder network and score the health status of the generator. While the scheme has some innovation in feature extraction, its health scoring mechanism is still based on the activation level of the output neural nodes. It lacks an interpretable index structure partitioning and weight modeling process, and does not involve modeling the coupling relationship between indicators, resulting in insufficient interpretability of the assessment results.

[0004] CN120450241A discloses a multi-source data fusion intelligent decision-making and early warning method for preventing false interlocking at wind farm booster stations. Based on expert knowledge and rule base, it performs fusion analysis on booster station monitoring data to achieve fault early warning and avoidance of false interlocking. Although the solution improves the operation and maintenance response capability, it still relies mainly on empirical rules in the processing of indicators and does not build a statistical coupling model between indicators. Especially when there is weak correlation or nonlinear dependence, it is difficult to fully reveal the health evolution law of the system and lacks the ability to structure and model the health status of the "whole station".

[0005] It is evident that existing wind turbine health assessment methods suffer from the following technical problems: difficulty in characterizing the nonlinear statistical dependencies between health indicators; lack of an integrated modeling mechanism that combines coupled analysis results with fuzzy hierarchical structures; and the fact that health status output structures are mostly single-value scores, lacking the linkage expression of level labels and state boundary information.

[0006] To address the aforementioned issues, this invention provides a comprehensive evaluation method for the overall health status of wind turbines based on fuzzy hierarchical analysis. By integrating edge distribution identification, Copula function coupling modeling, and segmented output of health levels, it achieves the modeling of nonlinear dependencies between multi-source health indicators, the combined calculation with multi-layer fuzzy structures, and the comprehensive evaluation of overall health level labels and boundary-aware output, effectively improving the accuracy and sensitivity of wind turbine health assessment. Summary of the Invention

[0007] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract and title of the invention. Such simplifications or omissions shall not be used to limit the scope of the present invention.

[0008] In view of the aforementioned existing problems, the present invention is proposed.

[0009] To address the aforementioned technical problems, this invention provides the following technical solution: Temperature, vibration, and load data of various components of the wind turbine are collected during the wind turbine's operating cycle to form a raw dataset of health indicators. A multi-level indicator structure for overall turbine health evaluation is established based on this raw dataset. Based on the raw dataset, the marginal distribution characteristics of each health indicator are determined, and the statistical correlation between indicators is calculated using time series correlation analysis to generate a health indicator association dataset. Based on the health indicator association dataset, a coupling model characterizing the nonlinear relationship of health indicators is constructed using a Copula function. The coupling strength value between health indicators is calculated based on the joint distribution function output by the coupling model, constructing a health indicator coupling strength matrix. The health indicator coupling strength matrix is ​​combined with the membership information corresponding to the multi-level indicator structure, and a fuzzy hierarchical analysis method is used to calculate the overall turbine health evaluation score. The overall turbine health status is output according to a preset health level interval, wherein the health level interval is divided based on the distribution characteristics of historical health evaluation scores.

[0010] As a preferred embodiment of the comprehensive health status evaluation method for wind turbines based on fuzzy hierarchical analysis as described in this invention, the step of outputting the overall health status of the turbine according to a preset health level range includes: The overall health status is classified into three levels: a first level, a second level, and a third level. The score range of the first level is greater than or equal to 0.8, the score range of the second level is greater than or equal to 0.5 and less than 0.8, and the score range of the third level is less than 0.5. The overall health evaluation score is compared sequentially with the grade intervals to determine the corresponding interval number, and the corresponding grade label is extracted to generate a structured health status data unit. Based on the difference between the overall health evaluation score and the corresponding interval boundary value, a difference record representing the degree of proximity to the current level boundary is generated, and the difference record is written into the additional fields of the health status output result, wherein the health status output result includes three fields: overall health evaluation score, corresponding health level label, and degree of proximity to the level boundary.

[0011] As a preferred embodiment of the comprehensive health status evaluation method for wind turbines based on fuzzy hierarchical analysis as described in this invention, the calculation of the overall health evaluation score includes: Set the initial membership weight for each health indicator in the multi-level indicator structure and establish a record table; Read each coupling value in the health indicator coupling strength matrix and perform weighted correction with the corresponding membership weight to form a corrected membership record table; Based on the multi-level indicator structure, hierarchical weighting is performed from bottom to top to generate a comprehensive membership record table. The coupling sensitivity is introduced as an additional field in the comprehensive membership record table and used as one of the hierarchical weighting parameters of the fuzzy hierarchical analysis to output the final overall health evaluation score.

[0012] As a preferred embodiment of the comprehensive health status evaluation method for wind turbines based on fuzzy hierarchical analysis as described in this invention, the generation of the health index coupling strength matrix includes: The health indicators are combined in pairs according to the numbering order of the multi-level indicator structure to generate health indicator pairs, and an index is established. Based on the health indicator pairs, joint value samples are extracted from the health indicator association dataset to form a relational sample subset; For each subset, the Copula function is applied for fitting, and the range of values ​​of the two health indicators in the subset is divided into segments according to the preset value range division rules. The coupling strength value is calculated based on the fitting parameters, and an intermediate coupling strength record table is constructed. Based on the indicator number, construct a two-dimensional matrix with health indicators as row and column indices, and fill each coupling strength value into the off-diagonal position to form the health indicator coupling strength matrix.

[0013] As a preferred embodiment of the comprehensive health status evaluation method for wind turbines based on fuzzy hierarchical analysis as described in this invention, the preset value interval division rules include: Read the numerical range of each health indicator in the subset of the relational samples, where the numerical range includes the minimum and maximum values ​​of the health indicator in all samples; The value range of each health indicator is segmented according to a preset interval division rule: If the value range of the health indicator is less than or equal to 1, then its value range is divided into 5 intervals, each interval having the same range length. If the value range of the health indicator is greater than 1 and less than or equal to 5, then its value range is divided into 4 intervals, each interval having the same range length. If the value range of the health indicator is greater than 5 and less than or equal to 10, then its value range is divided into 3 intervals, each interval having the same range length. If the value range of the health indicator is greater than 10, then its value range is divided into two intervals, each with the same range length.

[0014] As a preferred embodiment of the comprehensive health status evaluation method for wind turbines based on fuzzy hierarchical analysis as described in this invention, each segmented interval is labeled with a level according to the magnitude of the numerical value, wherein: If the value of the health indicator is in the first range, it is defined as low level; if the value of the health indicator is in the middle range, it is defined as medium level; if the value of the health indicator is in the last range, it is defined as high level. The grade labels after the interval division are matched with the corresponding health indicators one by one, generating a segmented statistical record of the non-linear relationship between each pair of health indicators, and generating a corresponding frequency statistical value for each segment.

[0015] As a preferred embodiment of the comprehensive health status evaluation method for wind turbines based on fuzzy hierarchical analysis as described in this invention, the health indicator association dataset is generated, including: Read the value sequence of each health indicator in the original dataset of the health indicators, extract the single indicator distribution features from the value sequence to describe the fluctuation trend, offset amplitude and concentration position, and write the single indicator distribution features into the marginal distribution feature record table; Based on the distribution characteristics of each health indicator in the marginal distribution feature record table, pairwise comparisons are made between each health indicator to calculate the statistical correlation results used to characterize the linear relationship between the indicators, and each statistical correlation result is written into the indicator correlation record table. In the indicator association record table, an indicator association strength identifier is generated based on the range of numerical changes in the statistical association results, and each indicator association strength identifier is written into the association strength field. In the correlation strength field, the correlation strength identifiers of each health indicator are summarized according to their numbering order to generate a health indicator correlation dataset.

[0016] As a preferred embodiment of the comprehensive health status evaluation method for wind turbines based on fuzzy hierarchical analysis as described in this invention, the calculation of statistical correlation results used to characterize the linear relationship between health indicators includes: Read the value sequence of each health indicator, perform discretization processing on the value sequence, and then write the processed data into the discretized indicator sequence record table; For each pair of health indicators in the discretized indicator sequence record table, the aligned data segment is extracted according to the time window sliding method to form a comparative analysis sample, and the sample is written into the indicator comparison sample field. Based on the aligned data segments in the comparison sample field of the aforementioned indicators, the average difference coefficient, the consistency rate of change direction, and the correlation coefficient of the rank sequence are calculated respectively, and the three types of statistical results are written into the indicator association analysis field; The three types of statistical results in the correlation analysis field of the indicator are weighted and synthesized to generate the final statistical correlation result.

[0017] As a preferred embodiment of the comprehensive health status evaluation method for wind turbines based on fuzzy hierarchical analysis as described in this invention, the original dataset of the health indicators is formed, including: During the wind turbine's operating cycle, temperature, vibration, and load data of each component are collected according to a preset data sampling interval, and the original measured values ​​corresponding to each data source are written into the initial monitoring data field. The data in the initial monitoring data field is processed for time synchronization, and a uniformly structured sampling dataset is generated based on the mapping relationship between the measurement point number and the indicator type, and written into the time-aligned data record table; For the sampled data in the time-aligned data record table, outlier removal and missing value completion are performed, and the processing results are written into the data cleaning record table; All fields in the data cleaning record table are reordered according to the health indicator dimension, and a raw dataset of health indicators containing indicator number, indicator type, and measurement sequence is generated.

[0018] As a preferred embodiment of the comprehensive health status evaluation method for wind turbines based on fuzzy hierarchical analysis as described in this invention, a multi-level index structure for overall turbine health evaluation is established based on the original dataset of health indicators, including: Based on the functional division relationship of the wind turbine system components, and according to the component to which the health indicators belong in the original data set of the health indicators, all health indicators are divided into primary structural units of generator system, main shaft system, pitch system, and yaw system, and the division results are written into the primary indicator structure mapping table. In the primary index structure mapping table, based on the monitoring parameter type of the index contained in each primary structural unit, they are further classified into temperature, vibration and load indexes, and written into the secondary index structure table; Based on the number, affiliated component, and type information of each indicator in the secondary indicator structure table, a multi-level structure path identifier is generated and written into the indicator structure path field. The indicator structure path field is bound to the unique number of each health indicator in the original dataset to form a multi-level indicator structure.

[0019] The beneficial effects of this invention are as follows: By constructing a system for collecting original health indicators, identifying and analyzing the marginal distribution of indicators, coupling modeling using Copula functions, and fusing output using fuzzy hierarchical analysis, this invention achieves explicit modeling and hierarchical fusion of complex relationships among multi-source health indicators, overcoming the limitations of traditional fuzzy hierarchical analysis methods in terms of indicator independence assumptions, subjective weight setting, and single evaluation output; by introducing the coupling strength matrix into the fuzzy analysis process, it achieves an innovative mechanism for dynamic adjustment of health weights; by using historical health evaluation data to generate grade intervals, it achieves adaptive optimization of health label output; and it effectively improves the accuracy and sensitivity of the overall health assessment of wind turbines. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a flowchart illustrating the comprehensive evaluation method for the overall health status of a wind turbine based on fuzzy hierarchical analysis, as shown in this invention. Detailed Implementation

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0022] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort should fall within the scope of protection of this invention.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] According to an embodiment of the present invention, in combination Figure 1 The flowchart shown illustrates a comprehensive evaluation method for the overall health status of a wind turbine based on fuzzy hierarchical analysis, which specifically includes the following steps: S1. Collect temperature, vibration, and load data for various components of the wind turbine during its operating cycle (e.g., one year) to form a raw dataset of health indicators. Based on this raw dataset, establish a multi-level indicator structure for overall turbine health evaluation. Note that the following points should be noted in this step: S1.1 During the wind turbine's operating cycle, collect temperature, vibration, and load data of each component according to a preset data sampling interval (e.g., 10 minutes), and write the original measured values ​​corresponding to each data source into the initial monitoring data field; As an example, at each preset sampling time point, the following component status data are synchronously collected through the wind turbine online monitoring system: generator end cover temperature, main bearing vibration amplitude, and yaw drive load, covering the main subsystems of the wind turbine. The temperature data is collected using a PT100 thermal resistance sensor, the vibration data is collected using a triaxial accelerometer, and the load data is recorded in real time using a strain gauge load sensor. The collected data is exported in CSV format and mapped to key-value fields using a JSON structure standard. This data is then written to a database table named "Initial Monitoring Data Field." Each record includes a timestamp, component number, indicator type, and measurement value. Example fields are shown below: { "timestamp": "2024-05-01T10:00:00Z", "component_id": "GEN_END", "metric_type": "temperature", "value": 68.3 } S1.2 Perform time-series synchronization processing on the data in the initial monitoring data field, and generate a uniformly structured sampling dataset based on the mapping relationship between the measurement point number and the indicator type, and write it into the time-aligned data record table; In a preferred embodiment, data from different acquisition time points in the initial monitoring data field are interpolated to ensure a consistent time sequence. A linear interpolation algorithm is used to align the time points to the nearest sampling interval. A mapping table is configured to clarify the one-to-one correspondence between the measurement point number and the indicator type, such as mapping GEN_END_T01 to the generator end cover temperature. The aligned data is then written into a time-aligned data record table, with fields including: timestamp, measurement point number, indicator type, and aligned measurement value. S1.3. For the sampled data in the time-aligned data record table, use the three sigma criterion to remove outliers exceeding 3σ, use the nearest neighbor interpolation algorithm to fill in missing values, and write the processing results into the data cleaning record table. S1.4 Reorder all fields in the data cleaning record table according to the health indicator dimension (i.e., indicator number) (e.g., ascending order) and generate the original health indicator dataset containing indicator number, indicator type, and measurement sequence. Examples are shown in Table 1: Table 1. Original Data Set of Health Indicators

[0025] S1.5. Based on the functional division relationship of the components of the wind turbine system, and according to the component to which the health indicators belong in the original data of health indicators, all health indicators are divided into first-level structural units of generator system (GEN, including generator stator temperature, end cover temperature, generator bearing vibration), main shaft system (SHAFT, including main bearing temperature, axial and radial vibration), pitch system (PITCH, including pitch motor load, motor temperature rise), and yaw system (YAW, including yaw motor current, yaw drive temperature). The division results are written into the first-level indicator structure mapping table, where each record contains the first-level structure number, the corresponding health indicator number and name. S1.6 In the primary index structure mapping table, based on the monitoring parameter type of the index contained in each primary structural unit, it is further classified into temperature, vibration and load indexes, and written into the secondary index structure table. Example as follows: Table 2. Structure of Secondary Indicators

[0026] S1.7 Based on the number, affiliated component and type information of each indicator in the secondary indicator structure table, a multi-level structure path identifier is generated using the path concatenation rule of "first-level number-second-level type-indicator number", such as GEN-TEMP-GEN_END_T01, and the multi-level structure path identifier is written into the indicator structure path field. S1.8 Bind the indicator structure path field to the unique number of each health indicator in the original dataset to form a multi-level indicator structure.

[0027] In a preferred embodiment, to establish a stable, logically mapped, multi-level structure of wind turbine health indicators, the structural path identifier of each health indicator needs to be bound to its unique number in the original dataset. This binding operation is completed using a one-to-one mapping of multiple fields, and the specific steps are as follows: Extract the multi-level structure path information of each health indicator from the indicator structure path field. The path is composed of a first-level structure system identifier, a second-level indicator type identifier, and an indicator unique number concatenated in a fixed format, using the first-level number-second-level type-indicator number rule, such as GEN-TEMP-GEN_END_T01. Extract the unique ID of all health indicators from the original health indicator dataset. This ID is a unique identifier for each type of indicator in the entire structure system. Establish an indicator number-structure path binding table to ensure that each health indicator number can uniquely correspond to a structure path identifier; the structure of this mapping table is shown in the table below: Table 3. Comparison Table of Health Indicator Numbers and Multi-level Structure Path Identifiers

[0028] The binding results are written into the multi-level indicator structure record table as multi-level structure definition fields, while the index relationship between the original indicator number and its measurement sequence is retained for structure tracing in the subsequent analysis stage. Regular expression validation is performed on the structure path identifier field after binding to ensure that the path structure meets the system abbreviation-type abbreviation-number format, avoid ambiguous paths or number conflicts, and unify the field naming convention.

[0029] For example, in this embodiment, taking the temperature monitoring index of a wind turbine as an example, the raw data collected by the generator end cover temperature sensor is numbered GEN_END_T01, and its corresponding structural information is as follows: Primary system structure: Generator system (GEN); Secondary indicator type: Temperature (TEMP); Indicator number: GEN_END_T01; The constructed multi-level path is identified as GEN-TEMP-GEN_END_T01, and a unique mapping relationship is established with the number GEN_END_T01. The complete record is as follows: { "metric_id": "GEN_END_T01", "hierarchical_path": "GEN-TEMP-GEN_END_T01", "data_source": "Temperature sensor_PT100", "metric_series": [68.3, 69.1, 67.8, ...] } It should be noted that in this binding structure, all subsequent analysis steps (such as correlation analysis, Copula coupling, and fuzzy hierarchical scoring) can accurately locate their respective component systems and monitoring types based on the GEN-TEMP-GEN_END_T01 structural path.

[0030] S2. Based on the original health indicator dataset, determine the marginal distribution characteristics of each health indicator, and use time series correlation analysis to calculate the statistical correlation between the indicators, generating a health indicator correlation dataset. Note that the following should be noted in this step: S2.1 Read the value sequence of each health indicator in the original health indicator dataset, and extract the single indicator distribution features from the value sequence to describe the fluctuation trend (such as the mean and direction of the first difference of the sequence), the offset magnitude (such as the difference between the maximum and minimum values, the skewness coefficient), and the concentration position (such as the mean, median, quantile), and write the single indicator distribution features into the marginal distribution feature record table. For example, as shown in the table below: Table 4. Record of Edge Distribution Characteristics

[0031] Each feature vector maintains a field identifier consistent with the indicator number to ensure consistency in field references in subsequent steps; As an example, this embodiment takes the generator end cover temperature sequence GEN_END_T01 as an example, and its measurement sequence is as follows: Then it can be extracted: Fluctuation trend characteristics, such as the first-order difference mean of 0.12 and the difference sign consistency rate of 0.63; Offset magnitude characteristics, such as range 3.4 and skewness 0.18; Central tendency features, such as mean 68.86, median 68.9, 75th percentile 69.5; The above information is written into the record table in vector form.

[0032] It should be noted that by constructing marginal distribution features, this step can provide the necessary basic statistics for the correlation analysis between indicators in the subsequent S2.2, so that the correlation analysis does not rely on the direct comparison of the original sequences, but is based on normalized statistical attributes, thereby improving the effectiveness of cross-indicator comparisons.

[0033] S2.2. Based on the distribution characteristics of each health indicator in the marginal distribution characteristic record table, perform pairwise comparisons between each health indicator, calculate the statistical correlation results used to characterize the linear relationship between indicators, and write each statistical correlation result into the indicator correlation record table. Further explanation is needed regarding the calculation of statistical correlation results used to characterize the linear relationship between health indicators. This includes: reading the value sequence of each health indicator, performing discretization on the value sequence, and then writing the processed data into a discretized indicator sequence record table; for each pair of health indicators in the discretized indicator sequence record table, extracting aligned data segments using a time window sliding method to form a comparative analysis sample, and writing this sample into the indicator comparison sample field; calculating the average difference coefficient, the consistency rate of change direction, and the rank sequence correlation coefficient based on the aligned data segments in the indicator comparison sample field, and writing the three types of statistical results into the indicator correlation analysis field; and weighting and synthesizing the three types of statistical results in the indicator correlation analysis field to generate the final statistical correlation result.

[0034] As an example, the discretized indicator sequence record table is read, and for each pair of indicators, a sliding window data segment of length L is extracted from the sequence according to the same time index and written into the indicator comparison sample field, where: Calculate the average difference coefficient :

[0035] in, For indicator A at the window position Discretized numerical values, For indicator B in the window position The discretized values, where L is the window length; Calculate the consistency rate of change direction :

[0036] in, This is an indicator function used to determine whether two signs are the same; it returns 1 if the signs are the same, and 0 otherwise. Calculate the correlation coefficient of the rank series (Constructed in Spearman format):

[0037] in, This represents the ranking value of indicator A within the window. The ranking value of indicator B within the window; The final statistical correlation results are calculated using a fixed weighting rule. :

[0038] in, , , These are the weighting coefficients for the three types of statistics, and their sum is 1. This weighted composite result is written into the index association analysis field.

[0039] S2.3 In the indicator association record table, generate indicator association strength identifiers based on the range of numerical changes in the statistical association results, and write each indicator association strength identifier into the association strength field; For example, read each statistical correlation result from the indicator correlation analysis field. Based on their numerical values, they are divided into three association strength levels and written into the association strength field; Table 5. Association Strength Indicator Table

[0040] It should be noted that this step uses a hierarchical identification structure to establish the correlation strength between indicators, providing a basis for the existence of significant linear relationships in subsequent Copula coupling analysis, so as to avoid performing unnecessary nonlinear coupling fitting on irrelevant indicator pairs.

[0041] S2.4 In the association strength field, according to the numbering order of health indicators, the association strength identifiers of each item are summarized to generate a health indicator association dataset.

[0042] Furthermore, following the numerical order of the health indicators, all association strength identifiers are re-aggregated in a row-column format to form a health indicator association dataset indexed by the indicator number; an example structure is as follows: Table 6. Health Indicator Correlation Data Table

[0043] If the calculation results of GEN_END_T01 and SHAFT_ACC01 are If the correlation strength is H1, then the corresponding cell in the above structure is recorded as H1.

[0044] S3. Based on the health indicator association dataset, construct a coupling model to characterize the nonlinear relationship between health indicators using a Copula function. Calculate the coupling strength value between health indicators based on the joint distribution function output by the coupling model, and construct a health indicator coupling strength matrix. Note that the following should be noted in this step: S3.1. Combine health indicators in pairs according to the numbering order of the multi-level indicator structure to generate health indicator pairs, and establish an index; It should be noted that the health indicator association dataset generated in step S2 is read, and the indicator pairs whose association strength identifier is not H3 (i.e., not low correlation) are combined in pairs to form a set of health indicator pairs, so as to avoid introducing low correlation meaningless indicator pairs into the calculation. In this embodiment, to avoid redundancy and symmetrical repetition, a unique key value is constructed by combining index numbers in ascending order and written into the index pair index table. The structure of each record is as follows: Table 7. Index Table of Indicators

[0045] If there is a high correlation between the number GEN_END_T01 and SHAFT_ACC01, then record it as D01, and the combination form is in ascending order (GEN first).

[0046] S3.2. Based on health indicator pairs, extract the complete measurement sequences of the two indicators in each indicator pair from the health indicator association dataset to form a relational sample subset; the structure is as follows: Table 8. Relational Sample Subset

[0047] Preferably, the above subset is used as input data for subsequent fitting of the Copula joint distribution model, reflecting the numerical relationship between the index pairs.

[0048] S3.3. Perform Copula function fitting on each subset, and divide the value range of the two health indicators in the subset into segments according to the preset value range division rules. Calculate the coupling strength value based on the fitting parameters and construct an intermediate coupling strength record table. It should be noted that the preset value range division rules include: Read the numerical range of each health indicator in the subset of relational samples. The numerical range includes the minimum and maximum values ​​of the health indicator in all samples. The value range of each health indicator is segmented according to a preset interval division rule: If the value range of a health indicator is less than or equal to 1, then its value range is divided into 5 intervals, each with the same range length. If the value range of a health indicator is greater than 1 and less than or equal to 5, then its value range is divided into 4 intervals, each with the same range length. If the value range of the health indicator is greater than 5 and less than or equal to 10, then its value range is divided into 3 intervals, each with the same range length. If the value range of a health indicator is greater than 10, its value range is divided into two intervals, each with the same range length. Furthermore, for each segmented interval, a level is assigned based on the magnitude of the numerical value, where: If the health indicator value is in the first range, it is defined as low level; if the health indicator value is in the middle range, it is defined as medium level; if the health indicator value is in the last range, it is defined as high level. The grade labels after the interval division are matched with the corresponding health indicators one by one, generating a segmented statistical record of the non-linear relationship between each pair of health indicators, and generating a corresponding frequency statistical value for each segment.

[0049] As an example, for each index pair and sample subset, a Gaussian Copula function constructed based on the empirical marginal distribution is selected for joint fitting, using the following formula:

[0050] in, , is the marginal cumulative distribution value of indicator A; , is the marginal cumulative distribution value of indicator B; This is the inverse standard normal transform; The correlation coefficient is The two-dimensional normal distribution function; For the joint Copula function; the result after fitting This is the initial value of the nonlinear coupling strength of the index pair.

[0051] To perform piecewise analysis on the coupled distribution structure of the fitted index, the following numerical range segmentation rule is adopted (applicable to cases where the original data is not normalized): Table 9. Numerical Range Division Table

[0052] Each interval after division is defined as low level (H3), medium level (H2), and high level (H1) according to the value range, and is then marked to form a combination of level pairs; It should be noted that the reason for segmenting the numerical range of each health indicator in the above manner is based on the following three points: The raw data of wind turbine health indicators have significant non-dimensional distribution characteristics: temperature indicators (such as generator end cover temperature) usually fluctuate between 40 and 90, vibration RMS values ​​are between 0.1 and 2, and load stress values ​​are usually between 10 and 30; if they are not standardized within intervals, it is impossible to use a unified level space for nonlinear statistics in coupled analysis. The interval division rule in this embodiment is used to solve the technical problem of establishing comparability of cross-dimensional indicators. When the indicator value range is small (such as vibration sequences ≤1), subdividing it into 5 segments can capture more small change intervals and avoid the loss of grade information caused by excessively wide intervals. However, when the value range is large (such as load indicators >10), if it is still divided into 5 segments, some interval data will be sparse, leading to statistical instability. Therefore, by gradually reducing the number of intervals, the stability of frequency statistics can be improved, the coupling value shift caused by sparse intervals can be avoided, and the unified grade structure between different indicators can be strengthened. After the interval is divided, each segment is marked as: high grade (H1), medium grade (H2), low grade (H3) to map continuous values ​​into discrete grade relationships, which are used to form grade pair combinations (such as H1-H1, H1-H2, H2-H3, etc.). By transforming the complex continuous distribution relationship into a structured grade combination, the subsequent frequency statistics and coupling strength calculation have a unified dimension, avoiding the instability caused by direct coupling of continuous values.

[0053] For each indicator pair, a corresponding rank sequence is generated based on the interval division results, for example: Indicator A: H1, H1, H2, H3, H2, ... Indicator B: H2, H1, H2, H2, H3, ... The sequence of level pairs is then: (H1,H2), (H1,H1), (H2,H2), (H3,H2), (H2,H3), ... Each combination is considered a level pair event.

[0054] Preferably, this partitioning rule is used to solve the technical problem that it is difficult to uniformly measure indicators with different dimensions. By using a unified segmentation and level mapping mechanism, the original data is mapped to a relative level space to ensure the standardization and comparability of subsequent frequency statistics and to provide clear interval distribution characteristics for coupling strength. Furthermore, the sample frequency is statistically analyzed for each level of pair combination. After normalizing the frequency, the coupling strength is calculated based on the following coupling strength function:

[0055] in, Indicators and The coupling strength, Indicates the first The first level corresponds to the first The normalized frequency of each level for the occurrence of combinations The weighting coefficient for the rank combination (e.g., the high-high combination has the largest weight). , The result is written to the intermediate coupling strength record table, which represents the number of levels (e.g., 3 for a 3-level system).

[0056] It should be noted that recording the statistical frequency of each level of pair combination in the intermediate coupling strength record table is for the final coupling strength. Provides quantifiable underlying event data: coupling strength It does not come directly from the Copula function, but requires the fusion of rank with frequency, frequency weight, and weighted summation to obtain the final result. Furthermore, noise in the original data can cause outliers, while frequency statistics of the sequence based on the rank can significantly smooth out noise, improve robustness, and provide a normalized index structure corresponding to rows and columns for subsequent matrix construction.

[0057] S3.4. Based on the indicator number, construct a two-dimensional matrix with health indicators as row and column indices, and fill each coupling strength value into the off-diagonal position to form the health indicator coupling strength matrix.

[0058] Specifically, all indicators are numbered and ordered by indicator sequence. , Establish a two-dimensional matrix The values ​​at the off-diagonal positions represent the coupling strength, while the diagonal positions are empty or denoted as... Its mathematical expression formula is as follows:

[0059] in, Indicators of health and The Copula coupling strength value, The total number of health indicators, This indicates that no coupling strength is defined between the same indicators; For example, if there are three metrics: GEN_END_T01, SHAFT_ACC01, and PITCH_LOAD02, the coupling strength matrix constructed through the above steps is as follows: Table 11. Coupling Strength Matrix Index Record Table

[0060] Preferably, this embodiment solves the problem of difficulty in modeling the nonlinear interaction relationship between different types of health indicators by using the above method. It adopts the Copula function to model the joint distribution and combines the level interval statistics and frequency weighting mechanism to generate a quantifiable and matrix-based coupling strength expression structure, which provides solid data support for multi-dimensional weight fusion in fuzzy hierarchical analysis and enhances the coupling sensitivity of subsequent whole-machine health scoring.

[0061] S4. Combine the health indicator coupling strength matrix with the membership information corresponding to the multi-level indicator structure, calculate the overall machine health evaluation score using fuzzy hierarchical analysis, and output the overall machine health status according to the preset health level interval, where the health level interval is divided based on the distribution characteristics of historical health evaluation scores. Note that the following should be noted in this step: S4.1 Set the initial membership weight of each health indicator in the multi-level indicator structure and establish a record table; For example, each health indicator is assigned an initial membership weight within the secondary structural unit, denoted as . The setting method is as follows: For each primary system (such as GEN, SHAFT, etc.), similar indicators are scored according to their importance (such as proximity of heat sources, frequency of historical failures, etc.), and the normalized weight allocation method is used for calculation, as shown in the following formula:

[0062] in, The number of health indicators within the k-th system; For example, in the GEN system, which includes GEN_END_T01 (end cap temperature) and GEN_ACC01 (vibration acceleration), if expert experience determines that the former is more critical, then the following settings are made: .

[0063] S4.2 Read each coupling value in the health indicator coupling strength matrix and perform weighted correction with the corresponding membership weight to form a corrected membership record table; Furthermore, the average coupling strength value corresponding to each health indicator is read from the coupling strength matrix and denoted as . This is used to measure the strength of its interaction with other indicators within the system, and the initial membership degree is adjusted according to the following formula:

[0064] Revised This is the first-level input used for hierarchical weighting; in, As an indicator Membership degree after coupling strength correction; As an indicator Initial membership degree within a secondary structure; As an indicator The average coupling strength with other indicators within the primary structure; For the first The number of indicators contained within a primary structure; This is an index for traversing all indicators within the first-level structure; It should be noted that this step addresses the problem that membership degrees, which are based solely on subjective importance assessments, cannot reflect the dynamic correlations between indicators. By adjusting the membership degree distribution through coupling strength, it integrates interactive information between structures, thereby improving the practical reliability of health status modeling.

[0065] S4.3. Based on the multi-level indicator structure, perform hierarchical weighting from bottom to top to generate a comprehensive membership record table. In a preferred embodiment, within the secondary structure (e.g., temperature-related, vibration-related, load-related), all parameters are applied as follows:

[0066] in, For the first The overall membership degree of a combination of secondary indicators. As an indicator For secondary structure The degree of belonging (such as the membership function weight). This represents the number of indicators within this secondary structure. Subsequently, the three secondary structures within each primary structure are weighted and normalized to obtain the overall membership degree of the primary structure. .

[0067] S4.4. Introduce coupling sensitivity as an additional field into the comprehensive membership record table, and use it as one of the hierarchical weighting parameters of fuzzy hierarchical analysis to output the final overall health evaluation score. Define the coupling sensitivity factor for each primary structure. Its source is the average value of the coupling strength of various indicators within the primary structure:

[0068] in, For the first The coupling sensitivity factor of a primary structure; This is the row index for indicators within the first-level structure; It is a column index of the indicator within the first-level structure, and is not equal to ; As an indicator With indicators The coupling strength; For example, the final overall health evaluation score H is calculated as follows:

[0069] Where H is the final output overall system health score, and K is the total number of Level 1 systems. For the first The weighted comprehensive membership degree of each primary structure For the first The coupling sensitivity of the structure For the first The standard deviation normalization result of the volatility of the individual structural indicators is the integral variable (representing the standard scoring range [0,1]). It is a sensitive decay function with penalty weights. The score weights are normalized functions, and the integral results reflect the overall response fusion trend under multi-structure coupling. It should be noted that, in this embodiment, the above-mentioned whole-machine health scoring formula is constructed by integrating hierarchical structure membership degree, health indicator coupling sensitivity and nonlinear integral function, aiming to solve the problem that the traditional fuzzy hierarchical analysis method cannot dynamically reflect the structural interaction relationship in health evaluation.

[0070] Specifically, the formula is based on the comprehensive membership degree of each primary system, representing the hierarchical importance of each structural unit in the overall score. It also introduces a coupling sensitivity factor, calculating the average coupling strength between health indicators within the primary structure to measure the correlation between indicators within the system, further constructing a response weighting system. In the integral structure, by combining a parameter-adjustable logarithmic function and a normalization function, dynamic response adjustment to structural volatility is achieved, enabling the overall turbine scoring process to adaptively reflect the attenuating weight effect of high-volatility structures. Finally, the integral is summed over the standard health scoring interval [0,1] to output a comprehensive health score reflecting the current overall turbine operating status. This scoring method not only retains the hierarchical modeling capability of the traditional FAHP method for system structure hierarchy and indicator membership, but also introduces nonlinear interaction information between indicators through coupling sensitivity, and introduces a response accumulation mechanism for dynamic trends through integral form. This constructs a health evaluation model with structured, coupled-driven, and dynamically adjustable characteristics, significantly improving the sensitivity, accuracy, and adaptability of the overall turbine operating status assessment.

[0071] S4.5. Set the overall health status grading intervals. The grading intervals include the first level interval, the second level interval, and the third level interval. The score range of the first level interval (high) is greater than or equal to 0.8, the score range of the second level interval (medium) is greater than or equal to 0.5 and less than 0.8, and the score range of the third level interval (low) is less than 0.5. The overall health evaluation score is compared with the grade range in sequence to determine the corresponding range number, and the corresponding grade label is extracted to generate a structured health status data unit. Based on the difference between the overall health evaluation score and the corresponding interval boundary value, a difference record representing the degree of proximity to the current level boundary is generated, and the difference record is written into the additional fields of the health status output result. The health status output result includes three fields: overall health evaluation score (H), corresponding health level label (such as excellent, medium, poor), and degree of proximity to the level boundary (absolute value of the difference from the boundary value).

[0072] Example result fields are as follows: { "health_score": 0.76, "health_grade": "medium", "proximity_to_boundary": 0.04 } It should be further noted that the overall machine health evaluation score ,in: The equipment is in good working order; The equipment is in moderate operating condition; monitoring is recommended. The equipment is at risk of malfunction and requires maintenance or shutdown.

[0073] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A comprehensive evaluation method for the overall health status of a wind turbine based on fuzzy hierarchical analysis, characterized in that, include: Temperature, vibration and load data of various components of the wind turbine are collected during the wind turbine's operating cycle to form a raw dataset of health indicators. A multi-level indicator structure for overall machine health evaluation is then established based on the raw dataset of health indicators. Based on the original dataset of health indicators, the marginal distribution characteristics of each health indicator are determined, and the statistical correlation between indicators is calculated using time series correlation analysis to generate a health indicator association dataset. Based on the health indicator association dataset, a coupling model for characterizing the nonlinear relationship between health indicators is constructed using the Copula function. The coupling strength value between health indicators is calculated based on the joint distribution function output by the coupling model, and a health indicator coupling strength matrix is ​​constructed. The coupling strength matrix of the health indicators is combined with the membership information corresponding to the multi-level indicator structure. The fuzzy hierarchical analysis method is used to calculate the overall health evaluation score and output the overall health status according to the preset health level interval, wherein the health level interval is divided according to the distribution characteristics of historical health evaluation scores.

2. The comprehensive evaluation method for the overall health status of a wind turbine based on fuzzy hierarchical analysis as described in claim 1, characterized in that, The step of outputting the overall health status of the machine according to a preset health level range includes: The overall health status is classified into three levels: a first level, a second level, and a third level. The score range of the first level is greater than or equal to 0.8, the score range of the second level is greater than or equal to 0.5 and less than 0.8, and the score range of the third level is less than 0.

5. The overall health evaluation score is compared sequentially with the grade intervals to determine the corresponding interval number, and the corresponding grade label is extracted to generate a structured health status data unit. Based on the difference between the overall health evaluation score and the corresponding interval boundary value, a difference record representing the degree of proximity to the current level boundary is generated, and the difference record is written into the additional fields of the health status output result, wherein the health status output result includes three fields: overall health evaluation score, corresponding health level label, and degree of proximity to the level boundary.

3. The comprehensive evaluation method for the overall health status of a wind turbine based on fuzzy hierarchical analysis as described in claim 2, characterized in that, The calculation of the overall machine health evaluation score includes: Set the initial membership weight for each health indicator in the multi-level indicator structure and establish a record table; Read each coupling value in the health indicator coupling strength matrix and perform weighted correction with the corresponding membership weight to form a corrected membership record table; Based on the multi-level indicator structure, hierarchical weighting is performed from bottom to top to generate a comprehensive membership record table. The coupling sensitivity is introduced as an additional field in the comprehensive membership record table and used as one of the hierarchical weighting parameters of the fuzzy hierarchical analysis to output the final overall health evaluation score.

4. The comprehensive evaluation method for the overall health status of a wind turbine based on fuzzy hierarchical analysis as described in claim 3, characterized in that, Generating the coupling strength matrix of the health indicators includes: The health indicators are combined in pairs according to the numbering order of the multi-level indicator structure to generate health indicator pairs, and an index is established. Based on the health indicator pairs, joint value samples are extracted from the health indicator association dataset to form a relational sample subset; For each subset, the Copula function is applied for fitting, and the range of values ​​of the two health indicators in the subset is divided into segments according to the preset value range division rules. The coupling strength value is calculated based on the fitting parameters, and an intermediate coupling strength record table is constructed. Based on the indicator number, construct a two-dimensional matrix with health indicators as row and column indices, and fill each coupling strength value into the off-diagonal position to form the health indicator coupling strength matrix.

5. The comprehensive evaluation method for the overall health status of a wind turbine based on fuzzy hierarchical analysis as described in claim 4, characterized in that, The preset value range division rules include: Read the numerical range of each health indicator in the subset of the relational samples, where the numerical range includes the minimum and maximum values ​​of the health indicator in all samples; The value range of each health indicator is segmented according to a preset interval division rule: If the value range of the health indicator is less than or equal to 1, then its value range is divided into 5 intervals, each interval having the same range length. If the value range of the health indicator is greater than 1 and less than or equal to 5, then its value range is divided into 4 intervals, each interval having the same range length. If the value range of the health indicator is greater than 5 and less than or equal to 10, then its value range is divided into 3 intervals, each interval having the same range length. If the value range of the health indicator is greater than 10, then its value range is divided into two intervals, each with the same range length.

6. The comprehensive evaluation method for the overall health status of a wind turbine based on fuzzy hierarchical analysis as described in claim 5, characterized in that, For each segmented interval, a level is assigned based on the magnitude of the numerical value, where: If the value of the health indicator is in the first range, it is defined as low level; if the value of the health indicator is in the middle range, it is defined as medium level; if the value of the health indicator is in the last range, it is defined as high level. The grade labels after the interval division are matched with the corresponding health indicators one by one, generating a segmented statistical record of the non-linear relationship between each pair of health indicators, and generating a corresponding frequency statistical value for each segment.

7. The comprehensive evaluation method for the overall health status of a wind turbine based on fuzzy hierarchical analysis as described in claim 4, characterized in that, Generate the health indicator association dataset, including: Read the value sequence of each health indicator in the original dataset of the health indicators, extract the single indicator distribution features from the value sequence to describe the fluctuation trend, offset amplitude and concentration position, and write the single indicator distribution features into the marginal distribution feature record table; Based on the distribution characteristics of each health indicator in the marginal distribution feature record table, pairwise comparisons are made between each health indicator to calculate the statistical correlation results used to characterize the linear relationship between the indicators, and each statistical correlation result is written into the indicator correlation record table. In the indicator association record table, an indicator association strength identifier is generated based on the range of numerical changes in the statistical association results, and each indicator association strength identifier is written into the association strength field. In the correlation strength field, the correlation strength identifiers of each health indicator are summarized according to their numbering order to generate a health indicator correlation dataset.

8. The comprehensive evaluation method for the overall health status of a wind turbine based on fuzzy hierarchical analysis as described in claim 7, characterized in that, The calculations used to characterize the statistical correlation results representing the linear relationship between health indicators include: Read the value sequence of each health indicator, perform discretization processing on the value sequence, and then write the processed data into the discretized indicator sequence record table; For each pair of health indicators in the discretized indicator sequence record table, the aligned data segments are extracted according to the time window sliding method to form a comparative analysis sample, and the sample is written into the indicator comparison sample field. Based on the aligned data segments in the comparison sample field of the aforementioned indicators, the average difference coefficient, the consistency rate of change direction, and the correlation coefficient of the rank sequence are calculated respectively, and the three types of statistical results are written into the indicator association analysis field; The three types of statistical results in the correlation analysis field of the indicator are weighted and synthesized to generate the final statistical correlation result.

9. The comprehensive evaluation method for the overall health status of a wind turbine based on fuzzy hierarchical analysis as described in claim 8, characterized in that, The original dataset for the health indicators is formed by: During the wind turbine's operating cycle, temperature, vibration, and load data of each component are collected according to a preset data sampling interval, and the original measured values ​​corresponding to each data source are written into the initial monitoring data field. The data in the initial monitoring data field is processed for time-series synchronization, and a uniformly structured sampling dataset is generated based on the mapping relationship between the measurement point number and the indicator type, and written into the time-aligned data record table; For the sampled data in the time-aligned data record table, outlier removal and missing value completion are performed, and the processing results are written into the data cleaning record table; All fields in the data cleaning record table are reordered according to the health indicator dimension, and a raw dataset of health indicators containing indicator number, indicator type, and measurement sequence is generated.

10. The comprehensive evaluation method for the overall health status of a wind turbine based on fuzzy hierarchical analysis as described in claim 9, characterized in that, Based on the original dataset of health indicators, a multi-level indicator structure for overall machine health evaluation is established, including: Based on the functional division relationship of the wind turbine system components, and according to the component to which the health indicators belong in the original data set of the health indicators, all health indicators are divided into primary structural units of generator system, main shaft system, pitch system, and yaw system, and the division results are written into the primary indicator structure mapping table. In the primary index structure mapping table, based on the monitoring parameter type of the index contained in each primary structural unit, they are further classified into temperature, vibration and load indexes, and written into the secondary index structure table; Based on the number, affiliated component, and type information of each indicator in the secondary indicator structure table, a multi-level structure path identifier is generated and written into the indicator structure path field. The indicator structure path field is bound to the unique number of each health indicator in the original dataset to form a multi-level indicator structure.

Images