A wind turbine multi-component abnormal propagation tracing and hierarchical early warning method and system

Abstract

The application discloses a kind of wind turbine multi-component abnormal propagation traceability and hierarchical early warning method and system, and relates to wind power generation equipment condition monitoring technical field.The method includes constructing key component health index sequence, dividing working condition unit, establishing the health benchmark and residual scale statistics of each component, calculating working condition normalized heat balance residual;Residual abnormal event is identified based on residual amplitude, change rate and duration;According to the relative target component abnormal response of source component abnormal event, determine the candidate propagation relationship;Through three constraint counterfactual propagation verification, the difference of target component propagation risk score after shielding or weakening the source component abnormal event is used as abnormal propagation contribution, confirm, sort and screen the candidate propagation edge;Build counterfactual verification abnormal propagation graph, identify abnormal propagation chain, calculate comprehensive risk score and output result.The application can distinguish synchronous correlation and real propagation relationship, improve the accuracy and interpretability of abnormal traceability and hierarchical early warning.

Description

Technical Field

[0001] This invention relates to the field of wind power equipment condition monitoring technology, and more specifically to a method and system for tracing the source and hierarchical early warning of anomalies in multiple components of a wind turbine. Background Technology

[0002] Currently, wind turbine condition monitoring mainly relies on SCADA systems to collect operating parameters such as temperature, power, and speed. Anomaly alarms are then triggered by establishing single-component temperature prediction models, residual analysis, or health scoring. Some methods combine drivetrain topology, statistical correlation, or time-delay analysis to attempt to determine the state correlation between multiple components. Existing technologies also employ dedicated sensors such as vibration and acoustic emission sensors to enhance local fault identification, but these typically require additional hardware deployment, resulting in higher engineering costs.

[0003] However, most existing SCADA-based methods still rely on independent threshold judgments for single components or overall health scoring, making it difficult to distinguish between normal operating condition fluctuations and genuine component anomalies. Furthermore, they lack the ability to identify the sequential relationships and propagation paths of abnormal events among multiple critical components. Some methods utilizing correlation or graphical models are prone to misinterpreting normal operating condition disturbances as abnormal propagation relationships when operating conditions such as wind speed, power, and ambient temperature change synchronously, and cannot confirm whether the increased risk of a target component is indeed caused by an abnormal event in a specific source component. In addition, existing early warning outputs are mostly isolated states such as "normal," "warning," and "fault," lacking explanations of anomaly source location, propagation path, and propagation contribution, making it difficult for maintenance personnel to determine priority inspection targets and handling sequences.

[0004] Therefore, how to accurately identify the true propagation relationship between abnormal events of multiple components based on existing SCADA data without adding additional dedicated sensors, and output graded early warning results with traceability and interpretation, is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of the above problems, the present invention proposes a method and system for tracing the source and classifying the early warning of anomalies in multiple components of wind turbine units, which overcomes or at least partially solves the above problems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, embodiments of the present invention provide a method for tracing the source and providing graded early warning of anomalies propagating in multiple components of a wind turbine, comprising the following steps: S1. Construct health indicator sequences for each of the multiple key components; S2. Divide the operating condition units according to wind speed, active power, ambient temperature and turbulence intensity, and establish the baseline statistics and residual scale statistics of the health indicators of each key component under each operating condition unit based on historical health samples. S3. Based on the current operating conditions, call the corresponding health indicator benchmark statistics and residual scale statistics, normalize the deviation between the current health indicators and the benchmark statistics, and obtain the normalized thermal balance residual of each key component. S4. Identify residual abnormalities of each key component based on the normalized thermal balance residuals under the aforementioned operating conditions; S5. Based on the leading relationship between the residual abnormal events of the source critical components and the residual abnormal responses of the target critical components, and in combination with multiple propagation judgment conditions, determine the candidate abnormal propagation relationship. S6. Perform counterfactual propagation verification on the candidate anomaly propagation relationship: Keep the current operating condition unit, the residual risk item of the target key component itself, and the residual abnormal events of other key components besides the source key component unchanged, only shield or weaken the residual abnormal events of the source key component, and use the difference between the propagation-related risk score of the target key component under the original conditions and the counterfactual conditions as the anomaly propagation contribution, and confirm the effective anomaly propagation edge based on the anomaly propagation contribution. S7. Using each key component as a node and the confirmed valid anomaly propagation edge as a directed edge, construct a counterfactual verification anomaly propagation graph to identify multi-component anomaly propagation chains. S8. Calculate the propagation chain risk score of the abnormal propagation chain to determine the main risk source components and the main abnormal propagation paths; S9. Construct a comprehensive risk score based on the residual risk of the key component itself, the input risk of the propagation chain, and the persistent risk of residual abnormality. Output a graded early warning result when the comprehensive risk score meets the continuous triggering condition.

[0008] Furthermore, the key components include at least two of the following: main bearing, gearbox, generator, and pitch system; The health indicator is composed of the difference between the monitoring parameters of key components and the reference parameters: The health index of the main bearing is composed of the difference between the main bearing temperature and the engine room temperature. The gearbox health index is composed of the difference between the gearbox bearing temperature and the gearbox oil temperature. The generator health index is composed of the difference between the temperature at the non-driving end and the temperature at the driving end of the generator. The health index of the pitch system consists of the difference between the temperature of the pitch control cabinet and the temperature of the wheel hub.

[0009] Furthermore, the normalized heat balance residual under the operating condition is determined as follows: The thermal balance residual is obtained by subtracting the current health index from the health thermal balance baseline statistic under the current operating condition unit. The thermal balance residual is normalized using the residual scale statistics under the current operating condition unit. The normalized thermal balance residuals are filtered or subjected to continuity constraints within a sliding time window to generate a normalized thermal balance residual sequence under steady-state conditions.

[0010] Furthermore, the identification of the residual abnormal event includes: When the normalized thermal balance residual amplitude of the critical component reaches or exceeds the amplitude threshold, and the residual change rate or cumulative change reaches or exceeds the corresponding threshold, and the duration of the state that simultaneously satisfies the amplitude condition and the change rate / cumulative change condition reaches the minimum duration step, a residual abnormal event is determined to have occurred. When the residual amplitude is consistently higher than the high amplitude amplification threshold, even if the rate of change does not reach the rate of change threshold, a residual abnormality event is still determined to have occurred. The amplitude threshold and the rate of change threshold are determined by the quantiles of historical healthy samples.

[0011] Furthermore, the multiple propagation determination conditions include: matching of the leader delay within a preset propagation delay range, consistency of residual change direction, hysteresis correlation within the sliding time window reaching a preset correlation threshold, leader relation repeatability score reaching a preset condition, propagation persistence score reaching a preset condition, and physical reachability meeting preset requirements. The candidate anomaly propagation relationship is only used as input for counterfactual propagation verification and is not directly used as the final anomaly propagation edge.

[0012] Furthermore, the counterfactual propagation verification includes: While keeping the current operating condition unit, the health index sequence of the target key component, its own residual risk items, and the residual abnormal events of other key components besides the source key component unchanged, the residual abnormal events of the source key component within the candidate propagation delay window are masked or proportionally weakened. Based on the weakened counterfactual residual input, the propagation-related risk score of the target key component is recalculated while keeping the residual risk term of the target key component unchanged. The difference between the propagation-related risk score under the original conditions and the propagation-related risk score under the counterfactual conditions is taken as the contribution of anomalous propagation. When the contribution of the abnormal propagation reaches or exceeds the preset contribution threshold, the candidate abnormal propagation relationship is confirmed as a valid abnormal propagation relationship, and the corresponding abnormal propagation edge is activated. When multiple source critical components point to the same target critical component, they are sorted from largest to smallest according to their contribution to abnormal propagation. Abnormal propagation edges whose contribution reaches the preset contribution threshold and are ranked in the top K positions or exceed the preset percentile threshold are retained.

[0013] Furthermore, the edge weight of the effective anomaly propagation edge is dominated by the counterfactual propagation contribution, and is constrained and modified by the leading delay matching score, direction consistency score, lag correlation score, repeatability score, propagation persistence score, and physical reachability mask. When the contribution of counterfactual propagation does not reach the preset contribution threshold, the abnormal propagation edge will not be activated even if the above score reaches the corresponding threshold. The physical reachability is determined based on at least one of the transmission chain connection relationship, thermal coupling path and control coupling relationship between key components. If the above relationship does not exist, the corresponding abnormal propagation edge will not be activated.

[0014] Furthermore, the propagation chain risk score is determined jointly based on the residual abnormality intensity of the critical component at the source of the propagation chain, the edge weight of each abnormal propagation edge in the propagation chain, the propagation duration, and the risk status of the target critical component. The abnormal intensity of the residual of the source key component is determined by the normalized thermal balance residual amplitude and its duration under the operating conditions of the source key component. The main abnormal propagation paths are determined by sorting them from high to low according to the propagation chain risk score; The main risk source component is the key component at the head of the abnormal propagation chain with the highest risk score in the propagation chain.

[0015] Furthermore, the comprehensive risk score is obtained by coupling a basic risk term, a cross-component propagation enhancement term, and a time-dependent amplification term: The basic risk item is the residual risk of the key components themselves; The cross-component propagation enhancement term is the propagation chain input risk, which is determined based on the edge weight of the activated abnormal propagation edge pointing to the target critical component, the counterfactual propagation contribution, and the corresponding propagation chain risk score. The time-duration amplification term is the residual abnormality persistence risk, which is determined based on the number of consecutive persistence steps of residual abnormality events of the target critical component; When there is no risk input into the transmission chain, the overall risk score is determined by the basic risk item and the time-dependent amplification item.

[0016] Secondly, embodiments of the present invention provide a system for tracing and classifying the propagation of anomalies in multiple components of a wind turbine generator, comprising: The data acquisition module is used to acquire existing SCADA operating data of the wind turbine. The health indicator construction module is used to construct a health indicator sequence for each key component based on the monitoring parameters and reference parameters corresponding to multiple key components. The operating condition unit division module is used to divide the operating status of wind turbine units into multiple operating condition units based on wind speed, active power, ambient temperature and turbulence intensity. The residual benchmark library construction module is used to establish benchmark statistics and residual scale statistics of health indicators for each key component under different operating conditions based on historical health samples. The normalized residual calculation module is used to calculate the normalized thermal balance residual of each key component based on the operating condition unit to which the current operating condition belongs. The stable residual generation module is used to perform sliding window processing on the normalized thermal balance residuals under the operating conditions to generate a stable residual sequence. The residual abnormal event identification module is used to identify residual abnormal events of each key component based on the stable residual sequence; The candidate propagation relationship identification module is used to determine candidate anomaly propagation relationships based on the leading relationship between the source critical component residual abnormal event and the target critical component residual abnormal response and multiple propagation judgment conditions. The counterfactual propagation verification module is used to perform counterfactual propagation verification on the candidate abnormal propagation relationship. The difference between the risk score related to the propagation of the target key component under the original conditions and the counterfactual conditions is used as the abnormal propagation contribution. The abnormal propagation edge is activated when the preset conditions are met. The Counterfact Verification Anomaly Propagation Graph Construction Module is used to construct a counterfact verification anomaly propagation graph with each key component as a node and the activated anomaly propagation edge as a directed edge, and to identify multi-component anomaly propagation chains. The source tracing and early warning module is used to calculate the risk score of the propagation chain to determine the main risk source components and the main abnormal propagation paths, and output the graded early warning results when the comprehensive risk score meets the continuous triggering conditions.

[0017] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a method and system for tracing the source and hierarchical early warning of anomalies in multiple components of wind turbine units, which has the following beneficial effects: 1. This invention divides operating condition units based on wind speed, active power, ambient temperature and turbulence intensity, and establishes health index benchmark statistics and residual scale statistics under the same operating condition unit. This can reduce the interference of normal operating condition fluctuations on anomaly identification and avoid misjudging normal thermal fluctuations as component anomalies.

[0018] 2. This invention uses the normalized thermal balance residual under operating conditions to characterize the degree of abnormal deviation of key components from the healthy thermal balance benchmark under the same operating conditions, and combines the residual amplitude, rate of change and duration to identify residual abnormal events, which can reduce false alarms caused by single-point noise or instantaneous disturbances.

[0019] 3. This invention identifies candidate anomaly propagation relationships based on leader delay, directional consistency, hysteresis correlation, repeatability, persistence, and physical reachability, so that the judgment of multi-component anomaly propagation no longer depends solely on fixed topology or static correlation.

[0020] 4. This invention distinguishes between synchronous correlation and true anomaly propagation relationship through three-constraint counterfactual propagation verification. Under the condition that the operating conditions, the residual risk items of the target component itself and the abnormal inputs of non-source components remain unchanged, only the residual abnormal events of the source component are changed, and the contribution of anomaly propagation is determined by the difference in propagation correlation risk scores, thereby improving the reliability of anomaly propagation side confirmation and risk source location.

[0021] 5. This invention identifies abnormal propagation chains based on counterfactual verification anomaly propagation graphs and outputs risk components, risk levels, main risk source components, abnormal propagation paths, propagation delays, propagation contributions, and priority inspection targets, providing more interpretable decision-making basis for wind farm operation and maintenance. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0023] Figure 1 This is a flowchart of the multi-component anomaly propagation tracing and hierarchical early warning method provided in this embodiment of the invention; Figure 2 This is a schematic diagram illustrating the construction of health indicators for key components and the division of operating condition units in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the construction of the health residual benchmark library and the calculation of the normalized heat balance residual under operating conditions provided in this embodiment of the invention; Figure 4 This is a schematic diagram of residual abnormal event identification provided in an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the identification and counterfactual propagation verification of the leader-response anomaly propagation relationship provided in this embodiment of the invention; Figure 6 This is a structural diagram of the multi-component anomaly propagation tracing and hierarchical early warning system provided in this embodiment of the invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] This invention discloses a method for tracing the source and providing graded early warning of anomalies in multiple components of a wind turbine, such as... Figure 1 As shown, it includes the following steps: S1. Construct health indicator sequences for each of the multiple key components; S2. Divide the operating condition units according to wind speed, active power, ambient temperature and turbulence intensity, and establish the baseline statistics and residual scale statistics of the health indicators of each key component under each operating condition unit based on historical health samples. S3. Based on the current operating conditions, call the corresponding health indicator benchmark statistics and residual scale statistics, normalize the deviation between the current health indicators and the benchmark statistics, and obtain the normalized thermal balance residual of each key component. S4. Identify residual abnormalities of each key component based on the normalized thermal balance residuals under the aforementioned operating conditions; S5. Based on the leading relationship between the residual abnormal events of the source critical components and the residual abnormal responses of the target critical components, and in combination with multiple propagation judgment conditions, determine the candidate abnormal propagation relationship. S6. Perform counterfactual propagation verification on the candidate anomaly propagation relationship: Keep the current operating condition unit, the residual risk item of the target key component itself, and the residual abnormal events of other key components besides the source key component unchanged, only shield or weaken the residual abnormal events of the source key component, and use the difference between the propagation-related risk score of the target key component under the original conditions and the counterfactual conditions as the anomaly propagation contribution, and confirm the effective anomaly propagation edge based on the anomaly propagation contribution. S7. Using each key component as a node and the confirmed valid anomaly propagation edge as a directed edge, construct a counterfactual verification anomaly propagation graph to identify multi-component anomaly propagation chains. S8. Calculate the propagation chain risk score of the abnormal propagation chain to determine the main risk source components and the main abnormal propagation paths; S9. Construct a comprehensive risk score based on the residual risk of the key component itself, the input risk of the propagation chain, and the persistent risk of residual abnormality. Output a graded early warning result when the comprehensive risk score meets the continuous triggering condition.

[0026] This invention divides operating condition units based on wind speed, active power, ambient temperature, and turbulence intensity, and establishes health index benchmark statistics and residual scale statistics under the same operating condition unit. This can reduce the interference of normal operating condition fluctuations on anomaly identification and avoid misjudging normal thermal fluctuations as component anomalies.

[0027] This invention provides a method for tracing and classifying early warning of multi-component anomaly propagation in wind turbines based on normalized thermal balance residual event chains and counterfactual propagation verification. This method uses existing SCADA operating data from the wind turbine as a foundation, and achieves multi-component anomaly propagation tracing and classified early warning through normalized residual modeling, residual anomaly event identification, leader-response candidate propagation relationship identification, three-constraint counterfactual propagation verification, and propagation chain risk calculation. This method does not require additional vibration sensors, acoustic emission sensors, electromagnetic noise sensors, or other dedicated condition monitoring sensors; it is entirely based on existing data resources from the wind turbine, offering advantages such as low deployment cost, strong interpretability, and high engineering practicality.

[0028] In one embodiment, such as Figure 1 As shown, the method includes the following steps S1 to S9. It should be noted that the step numbers above are only used to describe the technical logic of the embodiments of the present invention and do not limit the steps to be executed in a strict order. Without affecting the establishment of the operating condition residual benchmark, residual event identification, anomaly propagation verification, and hierarchical early warning output, each step can be merged, split, or executed in parallel according to the needs of online calculation or offline analysis.

[0029] In step S1: Construct the health indicator sequences for each of the multiple key components.

[0030] In this step, SCADA operation data of the wind turbine is acquired, and a health index sequence for each key component is constructed based on the monitoring parameters and reference parameters corresponding to multiple key components.

[0031] Acquire existing SCADA operating data of wind turbine without deploying additional vibration sensors, acoustic emission sensors, electromagnetic noise sensors or other dedicated condition monitoring sensors; like Figure 2 As shown, the SCADA operating data includes wind speed, active power, ambient temperature, turbulence intensity, nacelle temperature, main bearing temperature, gearbox bearing temperature, gearbox oil temperature, generator non-drive end temperature, generator drive end temperature, pitch control cabinet temperature, and hub temperature.

[0032] Preferably, the key components include at least two of the main bearing, gearbox, generator, and pitch system.

[0033] The acquired SCADA operation data undergoes time alignment, outlier removal, missing value imputation, smoothing, and normalization to ensure data quality and time consistency. Let sampling time t, key component number i, and the set of key components be:

[0034] Where N represents the number of critical components.

[0035] For the i-th critical component, select the monitoring parameters related to that critical component. and reference parameters The difference between the two is used as a health indicator:

[0036] The health indicators for the main bearing, gearbox, generator, and pitch system are respectively expressed as follows:

[0037] in, , , , These represent the health indicators of the main bearing, gearbox, generator, and pitch system, respectively. Indicates the main bearing temperature. Indicates cabin temperature. Indicates the temperature of the gearbox bearings. Indicates gearbox oil temperature. This indicates the temperature at the non-driving end of the generator. Indicates the temperature at the generator drive end. This indicates the temperature of the pitch control cabinet. This indicates the hub temperature.

[0038] To characterize the time-varying health status of critical components, the first... i The construction length of each key component is L Health indicator history window:

[0039] in, L Indicates the length of the history window.

[0040] In step S2: the operating condition units are divided according to wind speed, active power, ambient temperature and turbulence intensity, and the baseline statistics and residual scale statistics of the health indicators of each key component under each operating condition unit are established based on historical health samples.

[0041] The operating condition vector is constructed based on wind speed, active power, ambient temperature, and turbulence intensity. The operating condition vector at time t is denoted as:

[0042] in, Indicates wind speed. Indicates active power. Indicates ambient temperature. This indicates the intensity of turbulence.

[0043] Wind speed, active power, ambient temperature, and turbulence intensity are each divided into several intervals, and a joint operating condition unit is formed by combining multiple operating condition variable intervals. Let the first interval be... q The working condition unit is as follows:

[0044] in, Indicates the first a Wind speed range Indicates the first b A range of active power. Indicates the first c A range of ambient temperatures, Indicates the first d A range of turbulence intensity.

[0045] If time t Operating condition vector Falling into the working condition unit Then record:

[0046] in, Indicates time t The operating condition unit number.

[0047] The operating condition unit can be determined using equal-width intervals, quantile intervals, clustering, or manual experience intervals. For operating condition units with a sample size lower than a preset threshold, they can be merged with adjacent operating condition units, or compensated using the benchmark statistics of health indicators and residual scale statistics of adjacent operating condition units.

[0048] like Figure 3 As shown in the diagram, the system is divided into offline and online phases. The offline phase involves selecting historical health samples from historical operational data. These samples are categorized based on factors such as no fault records, no downtime records, no alarm records, stable operation periods after maintenance, or manually labeled healthy operation periods. (Note: The last sentence appears to be incomplete and possibly refers to a separate, unrelated section.) q The first working condition unit i The health sample set for each key component is as follows:

[0049] in, This represents the set of healthy sample moments.

[0050] Calculate the first based on the aforementioned healthy sample set. i The key component is in the first q Benchmark statistics of health indicators under individual working conditions:

[0051] And calculate the residual scaling statistic of the i-th critical component under the q-th operating condition:

[0052] The coefficient 1.4826 is used to convert the absolute deviation of the median into a robust scaling estimate consistent with the standard deviation.

[0053] When the number of healthy samples is insufficient under a certain working condition unit, the compensation benchmark can be calculated using global healthy samples:

[0054]

[0055] in, This is the baseline statistic for the health indicator of the i-th critical component calculated based on the global health sample. Let g represent the residual scaling statistic of the i-th critical component calculated based on the global health sample; And order:

[0056] This results in the formation of a healthy residual benchmark library:

[0057] in, K This indicates the number of operating condition units. The health residual benchmark library is used to provide residual benchmarks for the same operating condition.

[0058] In step S3: Entering the online stage, the current operating condition is identified based on the current SCADA data, and the corresponding health indicator benchmark statistics and residual scale statistics are called. The deviation between the current health indicators and the benchmark statistics is normalized to obtain the normalized thermal balance residual of each key component.

[0059] According to the operating condition unit to which the current operating condition belongs From the health residual benchmark library Calling the first i The baseline statistics of key components under the current operating conditions, namely the aforementioned baseline statistics of health indicators. And scaling statistics, namely the aforementioned residual scaling statistics. .

[0060] For thei For each key component, first calculate the thermal balance residual of the current health index relative to the health benchmark under the same operating conditions:

[0061] The thermal balance residual is not a direct exceedance of the original temperature or simple temperature difference of the key component, but rather a deviation of the health index of the key component from the healthy thermal balance baseline statistics under the same operating condition unit. It is used to characterize the degree of abnormal thermal imbalance after removing thermal fluctuations under normal operating conditions.

[0062] The thermal balance residuals are normalized using the residual scaling statistics under the current operating condition, resulting in the normalized thermal balance residuals for the current operating condition:

[0063] in, To prevent positive numbers with a denominator of zero, This is the residual scale statistic under the current operating condition unit.

[0064] Through the above processing Used for subsequent generation of stable residual sequences and identification of residual anomalies.

[0065] In one embodiment, the health residual benchmark database is periodically updated according to a sliding time window. During updates, only samples meeting the health status screening criteria are used, and the statistics of the old and new benchmarks are smoothly merged.

[0066] in, and These represent the baseline statistic and residual scale statistic of the health indicator calculated in the new window, respectively. Represents the smoothing coefficient, and .

[0067] It should be noted here that the left and right sides are smoothly blended. and It does not refer to the exact same concept. Its essence is the in-place update, i.e., assignment, which is common in the computer field. Its expression can be described in layman's terms as: new baseline = (1-λ) × old baseline + λ × new window.

[0068] For the first i Normalized thermal balance residuals of key components After performing sliding time window stabilization, the normalized thermal equilibrium residual sequence under stable operating conditions is obtained. .

[0069]

[0070] in,W This indicates the length of the sliding time window, and median represents the median function. Indicates the sampling time index within the sliding time window, superscript s This represents the residual sequence after stabilization.

[0071] In step S4: Based on the normalized thermal balance residual of the operating condition, identify residual abnormalities of each key component.

[0072] like Figure 4 As shown, the figure is divided into two parts. One part is a schematic diagram of the change of the normalized thermal balance residual under the operating condition over time, and the other part is the residual abnormal event judgment condition. In this embodiment, the judgment is made by combining the residual amplitude, the residual change rate and the minimum number of continuous triggering steps.

[0073] Before the event occurs, the normalized thermal balance residual under the operating condition is within the range of normal fluctuation. When the event occurs, its starting time is defined as the residual amplitude threshold. The period from the starting time of the event to the end time is defined as the continuous abnormal event window. The normalized thermal balance residual under the operating condition at the end time is also defined as the residual amplitude threshold. During the continuous abnormal event window, the normalized thermal balance residual under the operating condition reaches its maximum value. The stage of its continuous increase is defined as the range of increasing rate of change, until the first decrease.

[0074] For the i A key component, defining its time... t The residual amplitude is:

[0075] Definition of the first i The residual change of each key component between adjacent time points is:

[0076] Let the first i The residual amplitude threshold for each key component is The threshold for the rate of change of residuals is The minimum number of continuous triggering steps is K i In one embodiment, the residual magnitude threshold and the residual rate of change threshold can be determined by the quantiles of the healthy samples:

[0077] in, and They represent quantile functions, and This is the preset quantile level.

[0078] When the i Key components in continuous windows Internal satisfaction:

[0079] And it meets one of the following conditions:

[0080] or

[0081] Then determine the first i Several key components experienced residual abnormalities within their corresponding time windows. Indicates the first Key components at time The residual change Indicates the first Key components at time The residual amplitude, It is a high-amplitude magnification factor, and This setting can retain persistently high-amplitude anomalies, even if their rate of change is not significant, and will not be misjudged as normal.

[0082] The first i Key components at time t Whether a disability abnormality event occurred is recorded as follows:

[0083] Furthermore, the start time of each residual abnormal event is recorded. Event End Time Duration Peak amplitude Average amplitude , change direction and event intensity .

[0084] The duration is:

[0085] The peak amplitude is:

[0086] The average amplitude of the event is:

[0087] The direction of change is:

[0088] Event intensity can be expressed as:

[0089]

[0090] in, , , Let be the weighting coefficient, and satisfy:

[0091] Through the above processing, the normalized thermal equilibrium residual sequence under stable operating conditions can be transformed into a residual abnormal event sequence, providing an event basis for subsequent identification of candidate anomaly propagation relationships and verification of counterfactual propagation.

[0092] In step S5: Based on the leading relationship between the residual abnormal events of the source critical component and the residual abnormal responses of the target critical component, and combined with multiple propagation judgment conditions, the candidate abnormal propagation relationship is determined.

[0093] like Figure 5 As shown in the left half, for any two key components i and j Based on the occurrence time, propagation delay, residual change direction, recurrence, propagation persistence, and physical reachability of residual abnormal events, candidate anomaly propagation relationships are identified.

[0094] Let the first i The first key component is a candidate source key component, the second... j The key components are candidate target key components. The residual abnormal response time of the target key component refers to the moment when the normalized thermal balance residual of the target key component under stable operating conditions first reaches the preset response threshold, or the start time of the residual abnormal event of the target key component. When there are multiple residual abnormal response times of the target key component within the same analysis window, the earliest response time that satisfies the preset propagation delay range with the residual abnormal event of the source key component is selected as the residual abnormal response time of the target key component. If the residual abnormal event of the source key component occurs earlier than the residual abnormal response time of the target key component, there may be a leader-response relationship between the two.

[0095] Define the propagation delay between the abnormal event of the source critical component and the abnormal response of the target critical component as:

[0096] in, Indicates the start time of the abnormal event of the source critical component. This indicates the moment when the target's critical component responds abnormally.

[0097] When the propagation delay satisfies:

[0098] At that time, it was believed that the firsti The key component is relative to the first j Several key components meet the time-precedented conditions. Among them, and These represent the minimum propagation delay and the maximum propagation delay, respectively.

[0099] Furthermore, determine the consistency of the direction of residual change. If the following conditions are met:

[0100] It is then assumed that the source critical component and the target critical component satisfy the directional consistency condition.

[0101] Calculate the hysteresis correlation between the two. For the sliding time window... The lagged correlation score is defined as:

[0102] Where τ represents the lag step size and Δ represents the sliding time window.

[0103] Take the maximum hysteresis correlation score within the preset propagation delay range:

[0104] When the following conditions are met:

[0105] At that time, it is considered that the two satisfy the condition of lagged correlation. Among them, This indicates a preset relevant threshold.

[0106] Further calculate the repeatability score of the precedence relation:

[0107] in, This indicates the number of times that an abnormal event of the source critical component matches an abnormal response of the target critical component within a preset propagation delay range. This indicates the total number of abnormal events in the source critical components.

[0108] Further, a propagation persistence score is calculated. This score characterizes the extent to which the leader-response relationship between the source critical component residual abnormality event and the target critical component residual abnormality response persists within a continuous sliding window. The propagation persistence score can be expressed as:

[0109] in, Indicates the first i Key components up to the first j The propagation persistence score of each key component This represents the number of windows that satisfy the conditions of leader delay, direction consistency, and hysteresis within a continuous sliding window. ν To prevent the denominator from being zero and to adjust for positive numbers that maintain sensitivity.

[0110] Simultaneously, a physical accessibility mask is constructed based on the transmission chain connections, thermal coupling paths, and control coupling relationships among the key components of the wind turbine:

[0111] when At that time, it was not confirmed that the first i Key components up to the first j Propagation relationships of candidate anomalies for key components.

[0112] In one embodiment, the candidate anomaly propagation relationship score can be expressed as:

[0113] in, This indicates the leader delay matching score. The score represents the consistency of direction. This represents the maximum lag correlation score. Indicates repeated scores, to Let be the weighting coefficient, and satisfy:

[0114] When the candidate anomaly propagation relationship score satisfies:

[0115] At that time, determine the first i Key components up to the first j There are candidate anomaly propagation relationships among several key components, denoted as:

[0116] in, Indicates the first i Residual anomalies in one critical component may precede and affect the first. j The residual abnormal responses of key components. The candidate anomaly propagation relationship is only used as input for counterfactual propagation verification in step S6, and is not directly used as the final anomaly propagation edge.

[0117] In step S6: Counterfactual propagation verification.

[0118] like Figure 5As shown in the right half, a three-constraint counterfactual propagation verification is performed on the candidate anomaly propagation relationship. Specifically, while keeping the current operating condition unit, the residual risk term of the target critical component itself, and the residual anomalies of other critical components besides the source critical component unchanged, only the residual anomalies of the source critical component within the candidate propagation delay window are masked or proportionally weakened, and the anomaly propagation contribution is determined by the difference in the propagation-related risk scores of the target critical component.

[0119] First, calculate the target key components. j The transmission-related risk score under the original conditions is denoted as... The aforementioned propagation-related risk score is only used for calculating the contribution of counterfactual propagation and is not equivalent to the final comprehensive risk score for graded early warning.

[0120] For candidate anomaly propagation relationship Construct counterfactual conditions. The three-constraint counterfactual conditions include the following three constraints: First, the current operating condition unit... c ( t First, the target key components remain unchanged; second, the target key components remain unchanged. j The health indicator sequence, the stable operating condition normalized heat balance residual sequence, and the inherent residual risk term remain unchanged; third, except for key components... i Other critical components, including residual abnormalities, event intensity, and residual input, remain unchanged. Under the above constraints, only the source critical components... i Residual abnormal events within the candidate propagation delay window are masked or proportionally weakened.

[0121] When using the shielding method, let:

[0122] When using a weakening approach, let:

[0123] in, Indicating the counterfactual condition of the first i Abnormal event indicator variables for key components, Indicating the counterfactual condition of the first i Normalized thermal balance residuals under stable operating conditions of key components This indicates the reduction ratio.

[0124] In one embodiment, the reduction ratio When the value approaches 1, it indicates no shielding; when it approaches 0, it indicates shielding.

[0125] Under counterfactual conditions, and assuming the residual risk term of the target critical component remains unchanged, recalculate the target critical component. j The transmission-related risk score is denoted as The propagation-related risk score is either the propagation chain input risk score of the target critical component or a comprehensive risk score that includes the propagation chain input risk, used to characterize the shielding or weakening of the source critical component. i Changes in the propagation risk received by the target critical components after a residual abnormality event.

[0126] During the counterfactual propagation verification process, the residual risk term of the target critical component itself remains frozen. The shielding or weakening of the abnormal event of the source critical component only changes the propagation chain input risk received by the target critical component. Therefore, the difference between the propagation-related risk score under the original condition and the counterfactual condition is used to characterize the contribution of the abnormal event of the source critical component to the propagation risk of the target critical component.

[0127] Define candidate anomaly propagation relationships The contribution of counterfactual communication is:

[0128] When the following conditions are met:

[0129] At that time, confirm the candidate abnormal propagation relationship. For effective abnormal propagation relationships, and to activate the relationship by the first i The key component points to the first j Anomaly propagation edges for key components are activated; otherwise, the anomaly propagation edge is not activated. This indicates the preset propagation contribution threshold.

[0130] Furthermore, for the same target key components j If multiple candidate source key components exist, then a candidate propagation contribution set is constructed. C j ( t ), and sorted from largest to smallest according to the counterfactual propagation contribution of each candidate source's key components. Only when Reaching the preset propagation contribution threshold, and Key components pointing to the same target j The candidate propagation contribution set is ranked first K top The number of digits, or exceeding the preset quantile threshold. Q CF Only then is it activated by i point to j Abnormal propagation edge.

[0131] Furthermore, the edge weight synthesis function can be expressed in a weighted form:

[0132] in, This represents the normalized counterfactual propagation contribution. to Let be the weighting coefficient, and satisfy:

[0133] when If the preset propagation contribution threshold is not reached, or the contribution ranking does not meet the filtering criteria, then... .

[0134] In step S7: Using each key component as a node and the confirmed valid anomaly propagation edge as a directed edge, construct a counterfactual verification anomaly propagation graph to identify multi-component anomaly propagation chains.

[0135] Using each key component as a node and the activated anomaly propagation edge that has been verified through counterfactual propagation and satisfies the contribution ranking condition as directed edges, construct a counterfactual verification anomaly propagation graph:

[0136] in, Indicates time t The set of anomalous propagation edges that are activated after counterfactual propagation verification. Let represent the edge weight matrix of the counterfactual verification anomaly propagation graph.

[0137] The counterfactual verification anomaly propagation graph is dynamically formed by the anomaly propagation edges activated after counterfactual verification, rather than a preset fixed topology graph.

[0138] Based on the counterfactual verification anomaly propagation graph, the anomaly propagation chain is identified, and the main risk source components and main anomaly propagation paths are determined; Anomaly propagation diagram in counterfactual verification In the context of anomaly propagation, if there exists a set of consecutive activated anomaly propagation edges:

[0139] This forms an abnormal propagation chain:

[0140] in, p Indicates an abnormal propagation chain. Indicates key components of the transmission chain. Indicates the key components of the transmission chain target. m This represents the number of edges in the propagation chain.

[0141] In step S8: calculate the propagation chain risk score of the abnormal propagation chain, and determine the main risk source components and the main abnormal propagation paths.

[0142] In this step, after identifying the abnormal propagation chain based on the counterfactual verification anomaly propagation graph, the risk score of the propagation chain is calculated according to the residual abnormality intensity of the source component of the propagation chain, the edge weight of each abnormal propagation edge in the propagation chain, the propagation duration, and the risk status of the target component, thereby determining the main risk source component and the main abnormal propagation path.

[0143] The aggregate value of the edge weights of the propagation chain can be expressed as:

[0144] in, This represents the abnormal propagation edge weight between two adjacent critical components in the propagation chain.

[0145] The residual abnormal strength of the source component is expressed as:

[0146] The risk status of the target component is represented as follows:

[0147] The duration score of the propagation chain is represented as:

[0148] in, This indicates the number of steps in the propagation chain that persist within a continuous time window. This is the adjustment coefficient.

[0149] In a preferred embodiment, the propagation chain risk score is determined using a product form:

[0150] This product form makes the propagation chain risk significantly increase only when the source component's abnormal strength, propagation edge weight, target component's risk state, and propagation duration are all high.

[0151] The propagation chain risk scores of all anomalous propagation chains are ranked:

[0152] in, Indicates time t The set of abnormal propagation chains, This indicates the abnormal propagation chain with the highest risk score.

[0153] Will The key component at the beginning of the chain has been identified as the main source of risk:

[0154] Will This itself has been identified as the primary path of anomaly propagation.

[0155] If multiple anomalous propagation chains with similar risk scores exist, multiple candidate risk source components and candidate anomalous propagation paths can be output simultaneously, and sorted from high to low according to the propagation chain risk score and the cumulative value of counterfactual propagation contribution.

[0156] in, Indicates the propagation chain p The cumulative counterfactual propagation contribution of each abnormal propagation edge in the middle.

[0157] In step S9: a comprehensive risk score is constructed based on the residual risk of the key component itself, the input risk of the propagation chain, and the persistent risk of residual abnormality. When the comprehensive risk score meets the persistent triggering condition, a graded early warning result is output.

[0158] A comprehensive risk score is constructed based on the residual risk of key components themselves, the input risk of the propagation chain, and the persistent risk of abnormal residuals. The multi-component graded early warning results are output according to the continuous triggering conditions.

[0159] For the j The residual risk of a key component is represented as follows:

[0160] Where μ1, μ2 and μ3 are non-negative weight coefficients, and μ1+μ2+μ3=1.

[0161] The risk of transmission chain input is represented as:

[0162] in, Indicates time t All of the first j Each key component is a set of anomaly propagation chains for the target component.

[0163] The persistent risk of residual abnormalities is represented as follows:

[0164] in, Indicates the first j A critical component defect abnormality event at time t The number of consecutive steps, This is the adjustment coefficient.

[0165] In a preferred embodiment, the first j Key components at time t The overall risk score is obtained by coupling the basic risk of the residuals, the propagation chain input enhancement term, and the residual abnormality amplification term:

[0166] Where, β c γ represents the input enhancement coefficient of the propagation chain. c Indicates the residual abnormality amplification factor The warning threshold and alarm threshold are set as follows: and And satisfy:

[0167] When the comprehensive risk score continuously meets the early warning threshold and the number of consecutive trigger steps reaches... K warn When the overall risk score continuously meets the alarm threshold and the number of consecutive trigger steps reaches a certain threshold, an early warning status is output; when the overall risk score continuously meets the alarm threshold and the number of consecutive trigger steps reaches a certain threshold, an early warning status is output. K alarm When the alarm status is met, the alarm status is output; when the warning status or the continuous triggering condition corresponding to the alarm status is not met, the normal status is output.

[0168] In one embodiment, the multi-component hierarchical early warning result includes the risk component, risk level, comprehensive risk score, main risk source component, abnormal propagation path, propagation delay, propagation contribution, and priority inspection object. When multiple abnormal propagation paths exist, the system outputs the top-ranked abnormal propagation path according to the propagation chain risk score and the cumulative value of the counterfactual propagation contribution.

[0169] Through the above steps, this invention can identify the abnormal propagation relationships between multiple key components based on existing SCADA operating data of wind turbines without adding additional dedicated sensors, and output graded early warning results with traceable explanations. Compared with methods that only output health scores or single-point alarm results, this invention can further explain where the abnormality starts, along what path it propagates, and which components should be checked first, thereby improving the engineering practicality and operation and maintenance decision-making value of multi-component abnormality early warning for wind turbines.

[0170] Application Example 1: Source tracing and graded early warning of pitch system anomalies propagating to main bearings and gearbox; To further illustrate the specific application process of the wind turbine multi-component anomaly propagation tracing and hierarchical early warning method described in this invention, this application example selects existing SCADA operating data of a wind turbine for analysis. The original SCADA data includes wind speed, active power, ambient temperature, turbulence intensity, and relevant temperature parameters of the main bearing, gearbox, generator, and pitch system. After performing time alignment, outlier processing, missing value supplementation, and daily-scale statistics on the SCADA operating data, multi-component anomaly propagation tracing and hierarchical early warning are performed according to steps S1 to S9.

[0171] During the continuous monitoring period from August 25 to August 29, 2024, the pitch system first exhibited a persistent residual anomaly, followed by abnormal responses satisfying the leader-response conditions detected in the main bearing and gearbox. Taking August 27, 2024 as a representative date, the wind speed was 9.9908 m / s, the active power was 5025.8935 kW, the ambient temperature was 27.8224 ℃, and the turbulence intensity characterization value was 3.1697.

[0172] In step S1, a health index sequence is constructed based on the differences between the monitoring parameters and reference parameters of each key component, and the health indexes are then subjected to sliding smoothing. After processing, the health indices for the main bearing, gearbox, generator, and pitch system on August 27, 2024, were 24.1503, 26.3792, 10.8000, and 2.7367, respectively.

[0173] In step S2, the current operating condition is determined based on wind speed, active power, ambient temperature, and turbulence intensity. The corresponding operating condition unit for August 27, 2024 is W2_P2_T2_B2. A health residual benchmark library is established based on 44 historical health samples under this operating condition unit. The benchmark statistics of health indicators for the main bearing, gearbox, generator, and pitch system are 20.6245, 21.3708, 3.8700, and 2.5467, respectively, and the corresponding residual scale statistics are 2.6319, 4.8098, 3.0418, and 0.3842, respectively.

[0174] In step S3, the difference between the current health index and the baseline statistics of the health index under the corresponding operating condition unit is calculated, and normalization is performed using the corresponding residual scaling statistics. The calculated normalized thermal balance residuals for the main bearing, gearbox, generator, and pitch system are 1.3397, 1.0413, 2.2783, and 0.4945, respectively. After stabilization via a sliding time window, the normalized thermal balance residuals for the stable operating condition are 1.2445, 0.9037, 1.6575, and 2.6459, respectively.

[0175] In step S4, the pitch system began generating a residual anomaly on August 25, 2024, which continued until August 27, 2024, spanning three sampling steps. The stable residual peak value of this anomaly was 2.8541, the maximum residual change was 2.3943, the residual change direction was positive, and the dimensionless anomaly intensity was 1.5809. The corresponding residual amplitude threshold was 2.2652, and the residual change rate threshold was 1.1243. Therefore, this time period met the joint criteria of residual amplitude, change rate, and duration, and was identified as a residual anomaly event of the pitch system.

[0176] In step S5, using the pitch system as a candidate source critical component, a lead-response analysis is performed on the subsequent residual responses of other critical components. This results in two candidate anomaly propagation relationships: "pitch system → main bearing" and "pitch system → gearbox". The average propagation delay for "pitch system → main bearing" is 1.5 sampling steps, with a maximum hysteresis correlation coefficient of 0.2375, a directional consistency score of 1.0000, a repeatability score of 0.5000, a propagation persistence score of 0.7143, and a candidate propagation score of 0.6498. The average propagation delay for "pitch system → gearbox" is 2.0 sampling steps, with a maximum hysteresis correlation coefficient of 0.2199, a directional consistency score of 1.0000, a repeatability score of 0.5000, a propagation persistence score of 0.8462, and a candidate propagation score of 0.6486. All of these candidate propagation relationships satisfy the physical reachability condition and are only used as input for subsequent counterfactual propagation verification, not directly as the final anomaly propagation edge.

[0177] In step S6, counterfactual propagation verification is performed on the above candidate propagation relationships. While keeping the current operating condition unit, the residual risk terms of the target critical components themselves, and the abnormal inputs of other critical components besides the pitch system unchanged, only the abnormal event input of the pitch system is masked, and the propagation-related risks of the main bearing and gearbox are recalculated. The results are shown in Table 1.

[0178] Table 1. Results of Counterfactual Propagation Verification and Propagation Chain Risk Calculation

[0179] As shown in Table 1, after shielding the abnormal input to the pitch system, the propagation-related risk of the main bearing decreased from 0.6377 to 0.3942, with an abnormal propagation contribution of 0.2436; the propagation-related risk of the gearbox decreased from 0.5952 to 0.3240, with an abnormal propagation contribution of 0.2712. The abnormal propagation contributions of both candidate relationships met the preset contribution conditions; therefore, "pitch system → main bearing" and "pitch system → gearbox" are confirmed as valid abnormal propagation edges.

[0180] In step S7, a counterfactual verification anomaly propagation graph is constructed using the main bearing, gearbox, generator, and pitch system as nodes and the valid anomaly propagation edges confirmed by the above counterfactual verification as directed edges, and two anomaly propagation chains, "pitch system → main bearing" and "pitch system → gearbox", are identified.

[0181] In step S8, the propagation chain risk is calculated based on the residual anomaly intensity of the source critical component, the anomaly propagation edge weight, the risk status of the target critical component, and the propagation duration. On August 27, 2024, the propagation chain risk scores for "pitch system → main bearing" and "pitch system → gearbox" were 0.1350 and 0.1107, respectively, thus identifying the pitch system as the current primary risk source component. As the anomaly propagation process continued, by August 29, 2024, the risk scores of the two propagation chains increased to 0.2052 and 0.2079, respectively, indicating that the primary anomaly propagation path can be dynamically updated based on the anomaly persistence and the risk status of the target component.

[0182] In step S9, a comprehensive risk score is calculated based on the residual risk of the critical components themselves, the input risk of the propagation chain, and the risk of persistent residual abnormalities. On August 27, 2024, the comprehensive risk score of the pitch system was 0.8462, meeting the continuous alarm triggering condition and outputting an ALARM level; the comprehensive risk scores of the main bearing, gearbox, and generator were 0.4128, 0.3366, and 0.4235, respectively, and did not yet meet the continuous early warning triggering condition.

[0183] Therefore, even before independent early warnings are generated for the target critical components, this invention can already identify the abnormal propagation impact of the pitch system on the main bearing and gearbox through leader-response relationships and counterfactual propagation verification, and determine the pitch system as the main source of risk. Accordingly, the pitch system can be prioritized for inspection, while the main bearing and gearbox are listed as key components for subsequent attention, thus providing a basis for tracing the source of wind turbine anomalies and the order of operation and maintenance inspections.

[0184] Application Example 2: Elimination of directional propagation relationships under abnormal conditions of synchronous residuals; To further illustrate the ability of this invention to distinguish between the propagation relationship of synchronization anomalies and directional anomalies, this application example uses continuous operating data from December 13, 2024 to December 17, 2024 for analysis.

[0185] On December 13, 2024, the wind speed was 2.8530 m / s, the active power was 0 kW, and the ambient temperature was [missing information]. The temperature was 8.6663 ℃, and the turbulence intensity was 0.8795, corresponding to the operating condition unit W0_P0_T0_B0. This operating condition unit contains 64 historical healthy samples; therefore, the baseline statistics and residual scale statistics of the health indicators corresponding to this operating condition unit were used for residual calculation.

[0186] In steps S1 to S3, after data processing, the main bearing health index is 6.8085, the corresponding baseline statistic for the health index is 17.0223, and the residual scale statistic is 4.0458. The thermal balance residual is thus obtained. 10.2138, Normalized heat balance residual under operating conditions 2.5246 and the normalized heat balance residual under steady-state conditions 2.1280; Gearbox health indicators are 0.8500 corresponds to a baseline statistic of 8.3713 for the health indicator and a residual scaling statistic of 3.8591, from which the heat balance residual is obtained. 9.2213, Normalized Heat Balance Residual under Operating Conditions 2.3895 and the normalized heat balance residual under steady-state operating conditions 2.2217.

[0187] The negative residuals mentioned above indicate that the current health indicators are lower than the health benchmark under the corresponding working condition unit. The negative sign is used to indicate the direction of deviation of the residual abnormality, and does not indicate that the corresponding risk score is negative.

[0188] In step S4, both the main bearing and the gearbox experienced residual anomalies on December 13, 2024, which lasted until December 17, 2024, for a duration of 5 sampling steps. Specifically, the absolute value of the peak value of the stable residual anomaly in the main bearing was 2.5248, and the maximum residual change was 1.6900; the absolute value of the peak value of the stable residual anomaly in the gearbox was 2.3895, and the maximum residual change was 1.0954. The abnormal changes in both components were negative.

[0189] Although both the main bearing and the gearbox exhibited persistent residual anomalies within the same timeframe, and both are physically reachable in the wind turbine's drivetrain, the starting time of both component anomalies was December 13, 2024. There is no forward propagation delay relationship where one component's anomaly precedes the other's. Therefore, in step S5, a directional anomaly propagation relationship is determined not only based on overlapping anomaly times, consistent residual change directions, or physical connections, but also by the possibility of direct anomaly propagation.

[0190] After candidate propagation relationship screening in step S5 and counterfactual propagation verification in step S6, no valid abnormal propagation edges for "main bearing → gearbox" or "gearbox → main bearing" are activated in the current analysis window. Therefore, no directional abnormal propagation chain between the main bearing and the gearbox is formed in steps S7 and S8, and the input risk of the propagation chain between them is 0.

[0191] In step S9, the main bearing and gearbox undergo independent risk assessment based on their respective residual risks and residual abnormality persistence states. The graded early warning results at some time points are shown in Table 2.

[0192] Table 2. Graded early warning results under synchronous residual abnormal conditions

[0193] Because this invention uses continuous triggering conditions for tiered early warning, the warning level depends not only on the magnitude of the comprehensive risk score at the current moment, but also on the number of consecutive steps of the abnormal state. As the abnormal state persists, both the main bearing and the gearbox meet the corresponding continuous alarm conditions and output the ALARM level.

[0194] However, since there is no valid anomaly propagation edge between the two components that has been verified by counterfactual propagation, the system will not incorrectly identify the main bearing as the source of anomaly in the gearbox, nor will it incorrectly identify the gearbox as the source of anomaly in the main bearing. Instead, it will output its own anomaly risk status.

[0195] Therefore, when multiple key components exhibit synchronous residual anomalies at the same or similar times, this invention does not directly establish a directional anomaly propagation edge based solely on statistical correlation, consistency of residual change direction, or physical connection relationship between components. Instead, it further confirms the anomaly propagation relationship through leading time delay constraints and counterfactual propagation verification, thereby reducing the possibility of synchronous anomalies being misjudged as directional anomaly propagation.

[0196] Based on the same inventive concept, embodiments of the present invention also provide a system for tracing and classifying the propagation of anomalies in multiple components of a wind turbine, such as... Figure 6 As shown, it includes: The data acquisition module is used to acquire existing SCADA operating data of the wind turbine. The health indicator construction module is used to construct a health indicator sequence for each key component based on the monitoring parameters and reference parameters corresponding to multiple key components. The operating condition unit division module is used to divide the operating status of wind turbine units into multiple operating condition units based on wind speed, active power, ambient temperature and turbulence intensity. The residual benchmark library construction module is used to establish benchmark statistics and residual scale statistics of health indicators for each key component under different operating conditions based on historical health samples. The normalized residual calculation module is used to calculate the normalized thermal balance residual of each key component based on the operating condition unit to which the current operating condition belongs. The stable residual generation module is used to perform sliding window processing on the normalized thermal balance residuals under the operating conditions to generate a stable residual sequence. The residual abnormal event identification module is used to identify residual abnormal events of each key component based on the stable residual sequence; The candidate propagation relationship identification module is used to determine candidate anomaly propagation relationships based on the leading relationship between the source critical component residual abnormal event and the target critical component residual abnormal response and multiple propagation judgment conditions. The counterfactual propagation verification module is used to perform counterfactual propagation verification on the candidate abnormal propagation relationship. The difference between the risk score related to the propagation of the target key component under the original conditions and the counterfactual conditions is used as the abnormal propagation contribution. The abnormal propagation edge is activated when the preset conditions are met. The Counterfact Verification Anomaly Propagation Graph Construction Module is used to construct a counterfact verification anomaly propagation graph with each key component as a node and the activated anomaly propagation edge as a directed edge, and to identify multi-component anomaly propagation chains. The source tracing and early warning module is used to calculate the risk score of the propagation chain to determine the main risk source components and the main abnormal propagation paths, and output the graded early warning results when the comprehensive risk score meets the continuous triggering conditions.

[0197] Since the principle behind the problem solved by this system is similar to that of the aforementioned methods, the implementation of this system can be found in the implementation of the aforementioned methods, and the repetitive parts will not be repeated.

[0198] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for tracing the source and providing graded early warning of anomalies in multiple components of a wind turbine, characterized in that, Includes the following steps: S1. Construct a sequence of health indicators for each of several key components, wherein the key components include at least two of the following: main bearing, gearbox, generator, and pitch system; S2. Divide the operating condition units according to wind speed, active power, ambient temperature and turbulence intensity, and establish the baseline statistics and residual scale statistics of the health indicators of each key component under each operating condition unit based on historical health samples. S3. Based on the current operating conditions, call the corresponding health indicator benchmark statistics and residual scale statistics, normalize the deviation between the current health indicators and the benchmark statistics, and obtain the normalized thermal balance residual of each key component. S4. Identify residual abnormalities of each key component based on the normalized thermal balance residuals under the aforementioned operating conditions; S5. Based on the leading relationship between the residual abnormal events of the source critical components and the residual abnormal responses of the target critical components, and in combination with multiple propagation judgment conditions, determine the candidate abnormal propagation relationship. S6. Perform counterfactual propagation verification on the candidate anomaly propagation relationship: Keep the current operating condition unit, the residual risk item of the target key component itself, and the residual abnormal events of other key components besides the source key component unchanged, only shield or weaken the residual abnormal events of the source key component, and use the difference between the propagation-related risk score of the target key component under the original conditions and the counterfactual conditions as the anomaly propagation contribution, and confirm the effective anomaly propagation edge based on the anomaly propagation contribution. S7. Using each key component as a node and the confirmed valid anomaly propagation edge as a directed edge, construct a counterfactual verification anomaly propagation graph to identify multi-component anomaly propagation chains. S8. Calculate the propagation chain risk score of the abnormal propagation chain to determine the main risk source components and the main abnormal propagation paths; S9. Construct a comprehensive risk score based on the residual risk of the key component itself, the input risk of the propagation chain, and the persistent risk of residual abnormality. Output a graded early warning result when the comprehensive risk score meets the continuous triggering condition.

2. The method for tracing the source and providing graded early warning of anomalies in multiple components of a wind turbine as described in claim 1, characterized in that, The key components include at least two of the following: main bearing, gearbox, generator, and pitch system; The health indicator is composed of the difference between the monitoring parameters of key components and the reference parameters: The health index of the main bearing is composed of the difference between the main bearing temperature and the engine room temperature. The gearbox health index is composed of the difference between the gearbox bearing temperature and the gearbox oil temperature. The generator health index is composed of the difference between the temperature at the non-driving end and the temperature at the driving end of the generator. The health index of the pitch system consists of the difference between the temperature of the pitch control cabinet and the temperature of the wheel hub.

3. The method for tracing the source and providing graded early warning of anomalies in multiple components of a wind turbine as described in claim 1, characterized in that, The normalized heat balance residual under the operating condition is determined as follows: The thermal balance residual is obtained by subtracting the current health index from the health thermal balance baseline statistic under the current operating condition unit. The thermal balance residual is normalized using the residual scale statistics under the current operating condition unit. The normalized thermal balance residuals are filtered or subjected to continuity constraints within a sliding time window to generate a normalized thermal balance residual sequence under steady-state conditions.

4. The method for tracing the source and providing graded early warning of anomalies in multiple components of a wind turbine as described in claim 1, characterized in that, The identification of residual abnormal events includes: When the normalized thermal balance residual amplitude of the critical component reaches or exceeds the amplitude threshold, and the residual change rate / cumulative change reaches or exceeds the corresponding threshold, and the duration of the state in which the amplitude condition and the change rate / cumulative change condition are simultaneously satisfied reaches the minimum duration step, a residual abnormal event is determined to have occurred. When the residual amplitude is consistently higher than the high amplitude amplification threshold, even if the rate of change does not reach the rate of change threshold, a residual abnormality event is still determined to have occurred. The amplitude threshold and the rate of change threshold are determined by the quantiles of historical healthy samples.

5. The method for tracing the source and providing graded early warning of anomalies in multiple components of a wind turbine as described in claim 1, characterized in that, The multiple propagation determination conditions include: matching of the leader delay within the preset propagation delay range, consistency of the residual change direction, the hysteresis correlation within the sliding time window reaching a preset correlation threshold, the leader relation repeatability score reaching a preset condition, the propagation persistence score reaching a preset condition, and physical reachability meeting preset requirements. The candidate anomaly propagation relationship is only used as input for counterfactual propagation verification and is not directly used as the final anomaly propagation edge.

6. The method for tracing the source and providing graded early warning of anomalies in multiple components of a wind turbine as described in claim 1, characterized in that, The counterfactual propagation verification includes: While keeping the current operating condition unit, the health index sequence of the target key component, its own residual risk items, and the residual abnormal events of other key components besides the source key component unchanged, the residual abnormal events of the source key component within the candidate propagation delay window are masked or proportionally weakened. Based on the weakened counterfactual residual input, the propagation-related risk score of the target key component is recalculated while keeping the residual risk term of the target key component unchanged. The difference between the propagation-related risk score under the original conditions and the propagation-related risk score under the counterfactual conditions is taken as the contribution of anomalous propagation. When the contribution of the abnormal propagation reaches or exceeds the preset contribution threshold, the candidate abnormal propagation relationship is confirmed as a valid abnormal propagation relationship, and the corresponding abnormal propagation edge is activated. When multiple source critical components point to the same target critical component, they are sorted from largest to smallest according to their contribution to abnormal propagation. Abnormal propagation edges whose contribution reaches the preset contribution threshold and are ranked in the top K positions or exceed the preset percentile threshold are retained.

7. The method for tracing the source and providing graded early warning of anomalies in multiple components of a wind turbine as described in claim 1, characterized in that, The edge weight of the effective anomaly propagation edge is dominated by the counterfactual propagation contribution, and is constrained and modified by the leading delay matching score, direction consistency score, lag correlation score, repeatability score, propagation persistence score and physical reachability mask. When the contribution of counterfactual propagation does not reach the preset contribution threshold, the abnormal propagation edge will not be activated even if the above score reaches the corresponding threshold. The physical reachability is determined based on at least one of the transmission chain connection relationship, thermal coupling path and control coupling relationship between key components. If the above relationship does not exist, the corresponding abnormal propagation edge will not be activated.

8. The method for tracing the source and providing graded early warning of anomalies in multiple components of a wind turbine as described in claim 1, characterized in that, The propagation chain risk score is determined jointly based on the residual abnormality intensity of the critical component of the propagation chain source, the edge weight of each abnormal propagation edge in the propagation chain, the propagation duration, and the risk status of the target critical component. The abnormal intensity of the residual of the source key component is determined by the normalized thermal balance residual amplitude and its duration under the operating conditions of the source key component. The main abnormal propagation paths are determined by sorting them from high to low according to the propagation chain risk score; The main risk source component is the key component at the head of the abnormal propagation chain with the highest risk score in the propagation chain.

9. The method for tracing the source and providing graded early warning of anomalies in multiple components of a wind turbine as described in claim 1, characterized in that, The comprehensive risk score is obtained by coupling a basic risk term, a cross-component propagation enhancement term, and a time-dependent amplification term: The basic risk item is the residual risk of the key components themselves; The cross-component propagation enhancement term is the propagation chain input risk, which is determined based on the edge weight of the activated abnormal propagation edge pointing to the target critical component, the counterfactual propagation contribution, and the corresponding propagation chain risk score. The time-duration amplification term is the residual abnormality persistence risk, which is determined based on the number of consecutive persistence steps of residual abnormality events of the target critical component; When there is no risk input into the transmission chain, the overall risk score is determined by the basic risk item and the time-dependent amplification item.

10. A multi-component anomaly propagation tracing and hierarchical early warning system for wind turbines, characterized in that, include: The data acquisition module is used to acquire existing SCADA operating data of the wind turbine. The health indicator construction module is used to construct a health indicator sequence for each key component based on the monitoring parameters and reference parameters corresponding to multiple key components. The operating condition unit division module is used to divide the operating status of wind turbine units into multiple operating condition units based on wind speed, active power, ambient temperature and turbulence intensity. The residual benchmark library construction module is used to establish benchmark statistics and residual scale statistics of health indicators for each key component under different operating conditions based on historical health samples. The normalized residual calculation module is used to calculate the normalized thermal balance residual of each key component based on the operating condition unit to which the current operating condition belongs. The stable residual generation module is used to perform sliding window processing on the normalized thermal balance residuals under the operating conditions to generate a stable residual sequence. The residual abnormal event identification module is used to identify residual abnormal events of each key component based on the stable residual sequence; The candidate propagation relationship identification module is used to determine candidate anomaly propagation relationships based on the leading relationship between the source critical component residual abnormal event and the target critical component residual abnormal response and multiple propagation judgment conditions. The counterfactual propagation verification module is used to perform counterfactual propagation verification on the candidate abnormal propagation relationship. The difference between the risk score related to the propagation of the target key component under the original conditions and the counterfactual conditions is used as the abnormal propagation contribution. The abnormal propagation edge is activated when the preset conditions are met. The Counterfact Verification Anomaly Propagation Graph Construction Module is used to construct a counterfact verification anomaly propagation graph with each key component as a node and the activated anomaly propagation edge as a directed edge, and to identify multi-component anomaly propagation chains. The source tracing and early warning module is used to calculate the risk score of the propagation chain to determine the main risk source components and the main abnormal propagation paths, and output the graded early warning results when the comprehensive risk score meets the continuous triggering conditions.

Images