Wind turbine performance degradation root cause diagnosis method

Abstract

The application discloses a wind turbine performance degradation traceability diagnosis method, and relates to the technical field of wind power generation. A personalized performance baseline model reflecting the mutual relationship between various operating parameters under the health state of each unit is established. In the operation process, the correlation characteristics between the current parameters are compared with the baseline, and a comprehensive deviation index is calculated. When the deviation exceeds the set threshold, the traceability diagnosis process based on the causal forest algorithm is started, the influence intensity of each operating parameter on the deviation is quantified, and the root parameter cluster causing the performance degradation is positioned according to the influence size and automatically sorted. This process solves the problems of the existing method in the aspects of multi-dimensional parameter correlation relationship modeling, early degradation identification at the parameter combination level, and cause and effect differentiation and root positioning.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, and in particular to a method for tracing and diagnosing the performance degradation of wind turbine generator sets. Background Technology

[0002] Wind turbines operate under harsh conditions such as high wind speeds, variable loads, and extreme temperature variations over extended periods, inevitably leading to performance degradation in core components such as the pitch system, gearbox, and generator. This degradation often doesn't manifest as a single parameter exceeding its limit in the early stages, but rather as subtle changes in the relationships between operating parameters. Examples include a slow shift in the pitch angle at specific wind speeds, a mismatch between speed and power output, and alterations in the effect of temperature on transmission efficiency. If these early signs are not detected and traced back to their root causes in a timely manner, it can further lead to power generation losses, accelerated component wear, and even unplanned outages.

[0003] Currently, the detection and diagnosis methods for abnormal wind turbine performance can be broadly categorized into the following types.

[0004] One type is the single-parameter threshold monitoring method. This method sets fixed upper and lower limits for key parameters such as wind speed, engine speed, and power. Once a parameter exceeds the threshold, an alarm is triggered. Its limitation is that it ignores the coupling relationship between parameters, making it difficult to identify early deterioration states where the parameter combination has deviated but the individual parameters are still within the threshold range. False alarms and false alarms are relatively common.

[0005] Another type is the performance evaluation method based on power curves. This method determines whether the unit's performance has degraded by comparing the actual power output with the expected power calculated based on wind speed. Compared to the single-parameter threshold method, it considers the correlation between wind speed and power, but it does not incorporate multi-dimensional parameters such as engine speed, pitch angle, and temperature into the analysis, resulting in a relatively limited diagnostic dimension. When the root cause of the power reduction lies in the pitch system or drivetrain, this method often cannot further trace the specific cause.

[0006] In recent years, causal inference methods (such as causal forest algorithms) have begun to attract attention in root cause analysis of complex industrial systems, and some studies have attempted to apply them to scenarios such as wind power corrosion risk analysis. However, most existing work focuses on the identification of specific failure modes and has not yet formed a complete closed-loop methodology from baseline modeling, deviation detection to causal tracing and root cause localization. In particular, effective technical means are still lacking in the construction of personalized baseline models, the quantification of deviations in multidimensional parameter associations, and the automatic tracing of root cause parameter clusters. Summary of the Invention

[0007] The main objective of this invention is to propose a method for tracing and diagnosing the performance degradation of wind turbine generator sets, aiming to solve the technical problems in the prior art of lacking systematic modeling of the correlation between multi-dimensional parameters and the difficulty in detecting early degradation at the parameter combination level.

[0008] To achieve the above objectives, this invention proposes a method for tracing and diagnosing the performance degradation of wind turbine generator sets, comprising: Collect multi-dimensional operating parameter data of each wind turbine generator in the target wind farm under normal operating conditions. The operating parameters include at least wind speed, rotational speed, pitch angle, ambient temperature, and power. Based on the operating parameter data under the normal operating conditions, a corresponding performance baseline model is constructed for each unit. The performance baseline model is used to characterize the correlation between various operating parameters of the unit under healthy conditions. Real-time collection of current operating parameter data for each unit; comparison of the current correlation between current operating parameters with the corresponding performance baseline model; calculation of the deviation index of the current operating state relative to the performance baseline model. In response to the deviation index exceeding a preset threshold, the source tracing and diagnosis process is triggered; Based on the causal forest algorithm, a causal inference model is constructed with the deviation index as the target variable, and the causal effect strength of each operating parameter on the deviation index is calculated. Based on the ranking of the causal effect intensity, the root cause parameter clusters leading to performance degradation are traced, and the source diagnosis results are output.

[0009] In one embodiment, the step of constructing a corresponding performance baseline model for each unit based on the operating parameter data under the normal operating conditions includes: Using Gaussian process regression or random forest regression, with the wind speed and ambient temperature as input variables and the rotational speed, pitch angle and power as output variables, a multi-output regression model for each unit under healthy conditions is established as the performance baseline model. The performance baseline model also includes confidence intervals for each output variable.

[0010] In one embodiment, the step of comparing the current correlation between current operating parameters with the corresponding performance baseline model and calculating the deviation index of the current operating state relative to the performance baseline model includes: Input the current combination of operating parameters into the performance baseline model, and calculate the standardized residual vector between the actual output value and the model prediction value; The standardized residual vectors are aggregated into a comprehensive deviation index based on Mahalanobis distance or Euclidean distance.

[0011] In one embodiment, after the step of calculating the deviation index of the current operating state relative to the performance baseline model, the method further includes: The deviation index is smoothed using a time sliding window, and the degradation rate is calculated based on the temporal change trend of the smoothed deviation index.

[0012] In one embodiment, the step of constructing a causal inference model with the deviation index as the target variable based on the causal forest algorithm includes: Construct a training dataset containing historical normal operating state samples and current deterioration state samples, using the deviation of each operating parameter as a feature variable and the deviation index as a target variable; Multiple causal decision trees are constructed using the causal forest algorithm, and each causal decision tree is grown based on random subsampling and random feature subsets; Based on the causal effect estimates of each of the causal decision trees, the average causal effect strength of each feature variable on the target variable is calculated, and the confidence interval of each average causal effect strength is calculated.

[0013] In one embodiment, the step of tracing the root cause parameter clusters leading to performance degradation based on the ranking of the causal effect strengths includes: Arrange the causal effect strength of each operating parameter in descending order; The first few parameters whose cumulative causal effect contribution rate reaches a preset threshold are extracted as the root cause parameter cluster; the root cause parameter cluster includes a single parameter or a combination of parameters.

[0014] In one embodiment, in response to the root cause parameter set including a pitch angle parameter, the step of tracing the root cause parameter set leading to performance degradation further includes: Obtain the deviation between the actual value of the pitch angle parameter at the current ambient temperature and the baseline value at the corresponding temperature in the performance baseline model; The deviation value is compared with the allowable deviation range after temperature correction. When the deviation value exceeds the allowable deviation range, the degradation mode is determined to be a temperature-related pitch angle deviation mode.

[0015] In one embodiment, before the step of comparing the current correlation between current operating parameters with the corresponding performance baseline model, the method further includes: For multiple turbines in the same wind farm that are in the same wind speed range and the same ambient temperature range, calculate the deviation index of each turbine. The deviation indicators of each unit are relatively sorted. When the deviation indicator of one unit exceeds the preset position threshold in the relative sort, the unit is marked as a priority source tracing and diagnosis object.

[0016] In one embodiment, the step of outputting the source tracing diagnostic results further includes: In response to the confirmation that the unit performance has returned to normal, the normal operating parameter data under the current operating conditions is used as an incremental sample to incrementally update the parameters of the performance baseline model; During the incremental update process, the incremental samples are anomaly identified based on the model prediction error, and samples identified as anomalies are automatically removed.

[0017] In one embodiment, the step of outputting the source tracing diagnostic results further includes: A diagnostic report is generated based on the source tracing and diagnostic results. The diagnostic report includes at least the performance degradation warning time, the degradation severity level, the causal effect contribution ranking of each parameter in the root cause parameter cluster, and the corresponding recommended maintenance strategy. The diagnostic report is sent to the human-computer interaction interface for display.

[0018] In the technical solution of this invention, a personalized performance baseline model reflecting the interrelationships between various operating parameters under healthy conditions is first established for each unit. During operation, the correlation characteristics between current parameters are compared with this baseline to calculate a comprehensive deviation index. When the deviation exceeds a set threshold, a source tracing and diagnosis process based on a causal forest algorithm is initiated to quantify the influence intensity of each operating parameter on the deviation and automatically sort them according to the magnitude of the influence, thereby locating the root cause parameter clusters leading to performance degradation. This process addresses the shortcomings of existing methods in multidimensional parameter correlation modeling, early degradation identification at the parameter combination level, and causal differentiation and root cause location.

[0019] Specifically, because the baseline model is independently constructed based on the historical health data of each unit, the interference of differences in manufacturing processes and operating environments between units on the diagnostic results is effectively weakened, making deviation detection more targeted and the results more reliable. Simultaneously, by comparing the correlation between current operating parameters and the baseline as a whole and aggregating them into a single comprehensive deviation index, the monitoring method is upgraded from traditional single-parameter threshold alarms to multi-parameter correlation deviation measurements, enabling the detection of early signs of degradation at the parameter combination level before individual parameters exceed their limits. Furthermore, by introducing a causal forest algorithm, a causal inference model is constructed with deviation as the target variable, quantifying the causal effect strength of each parameter. Without relying on a pre-set fault knowledge base, it can automatically distinguish from the operating data which parameter changes are the cause of degradation and which are merely accompanying results, ultimately achieving accurate location and ranking of root cause parameter clusters. This method helps shorten the fault investigation time for operation and maintenance personnel, reduces power generation losses caused by unplanned outages, and provides reliable data support for precise operation and maintenance and repair decisions in wind farms. Attached Figure Description

[0020] 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0021] Figure 1 This is a flowchart of a method for tracing and diagnosing the performance degradation of wind turbine generator sets according to an embodiment of the present invention.

[0022] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0023] 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 a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0024] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a specific posture. If the specific posture changes, the directional indicators will also change accordingly.

[0025] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the use of "and / or" or "and / or" throughout the text includes three parallel solutions. For example, "A and / or B" includes solution A, solution B, or a solution where both A and B are satisfied simultaneously. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0026] In recent years, causal inference methods (such as causal forest algorithms) have begun to attract attention in root cause analysis of complex industrial systems, and some studies have attempted to apply them to scenarios such as wind power corrosion risk analysis. However, most existing work focuses on the identification of specific failure modes and has not yet formed a complete closed-loop methodology from baseline modeling, deviation detection to causal tracing and root cause localization. In particular, effective technical means are still lacking in the construction of personalized baseline models, the quantification of deviations in multidimensional parameter associations, and the automatic tracing of root cause parameter clusters.

[0027] Please combine Figure 1 To address the above problems, this invention proposes a method for tracing and diagnosing the performance degradation of wind turbine generator sets, comprising: S100: Collect multi-dimensional operating parameter data of each wind turbine generator in the target wind farm under normal operating conditions. The operating parameters include at least wind speed, rotational speed, pitch angle, ambient temperature, and power. Collect multi-dimensional operating parameter data for each wind turbine in the target wind farm under normal operating conditions. The collected operating parameters should at least cover wind speed, engine speed, pitch angle, ambient temperature, and power. This step provides the data foundation for subsequent baseline model construction and deviation analysis. S200: Based on the operating parameter data under the normal operating conditions, construct a corresponding performance baseline model for each unit. The performance baseline model is used to characterize the correlation between various operating parameters of the unit under healthy conditions. Based on the operating parameter data collected under normal operating conditions, a separate performance baseline model is established for each unit. This model is used to characterize the correlation between various operating parameters of the unit under healthy conditions. Specifically, a regression method capable of handling multiple input and multiple output relationships can be used. For example, wind speed and ambient temperature can be used as input variables, and engine speed, pitch angle, and power can be used as output variables to establish a multi-output regression model. Simultaneously, the confidence intervals for each output variable are determined to allow for subsequent assessment of whether actual observed values ​​significantly deviate from the normal range. S300: Collects the current operating parameter data of each unit in real time, compares the current correlation between the current operating parameters with the corresponding performance baseline model, and calculates the deviation index of the current operating state relative to the performance baseline model. The system collects real-time data on the unit's current operating parameters and compares the relationships between these parameters with the corresponding performance baseline model. Specifically, the current operating parameters can be combined and input into the baseline model to obtain the predicted values ​​and confidence intervals of each output variable. The standardized residual vector between the actual output value and the predicted value is then calculated. Finally, this residual vector is aggregated into a comprehensive deviation index using an appropriate distance metric to reflect the overall degree of deviation of the current operating status from the healthy baseline. S400: In response to the deviation index exceeding a preset threshold, the source tracing and diagnosis process is triggered; The deviation index obtained in step S300 is compared with a preset threshold. When the deviation index exceeds the threshold, it is considered that the unit's operating status has shown a noteworthy abnormal deviation, triggering the subsequent source tracing and diagnosis process; if it does not exceed the threshold, monitoring continues. S500: Based on the causal forest algorithm, a causal inference model is constructed with the deviation index as the target variable, and the causal effect strength of each operating parameter on the deviation index is calculated. After the deviation exceeds the threshold, a causal inference model with the deviation index as the target variable is constructed. Specifically, a training dataset is built using historical normal operating state samples and current deteriorated state samples, with the deviation of each operating parameter as the feature variable and the deviation index as the target variable. Multiple causal decision trees are constructed using the causal forest algorithm, with each tree growing based on random subsampling and random feature subsets. The causal effect estimates of each tree on each feature variable are summarized to obtain the average causal effect strength of each operating parameter on the deviation index, and the corresponding confidence interval is calculated. S600: Based on the ranking of the causal effect intensity, trace the root cause parameter cluster that leads to performance degradation and output the source diagnosis results.

[0028] Based on the causal effect strength of each parameter obtained in step S500, the parameters are sorted, and the parameter or parameter combination that contributes the most to performance degradation is determined according to the sorting result. This parameter is then output as the root cause parameter cluster. This root cause parameter cluster can be a single parameter or a combination of multiple parameters. The diagnostic results can provide a basis for subsequent operation and maintenance decisions.

[0029] In the technical solution of this invention, a personalized performance baseline model reflecting the interrelationships between various operating parameters under healthy conditions is first established for each unit. During operation, the correlation characteristics between current parameters are compared with this baseline to calculate a comprehensive deviation index. When the deviation exceeds a set threshold, a source tracing and diagnosis process based on a causal forest algorithm is initiated to quantify the influence intensity of each operating parameter on the deviation and automatically sort them according to the magnitude of the influence, thereby locating the root cause parameter clusters leading to performance degradation. This process addresses the shortcomings of existing methods in multidimensional parameter correlation modeling, early degradation identification at the parameter combination level, and causal differentiation and root cause location.

[0030] Specifically, because the baseline model is independently constructed based on the historical health data of each unit, the interference of differences in manufacturing processes and operating environments between units on the diagnostic results is effectively weakened, making deviation detection more targeted and the results more reliable. Simultaneously, by comparing the correlation between current operating parameters and the baseline as a whole and aggregating them into a single comprehensive deviation index, the monitoring method is upgraded from traditional single-parameter threshold alarms to multi-parameter correlation deviation measurements, enabling the detection of early signs of degradation at the parameter combination level before individual parameters exceed their limits. Furthermore, by introducing a causal forest algorithm, a causal inference model is constructed with deviation as the target variable, quantifying the causal effect strength of each parameter. Without relying on a pre-set fault knowledge base, it can automatically distinguish from the operating data which parameter changes are the cause of degradation and which are merely accompanying results, ultimately achieving accurate location and ranking of root cause parameter clusters. This method helps shorten the fault investigation time for operation and maintenance personnel, reduces power generation losses caused by unplanned outages, and provides reliable data support for precise operation and maintenance and repair decisions in wind farms.

[0031] In one embodiment, step S200 includes: S210: Using Gaussian process regression or random forest regression, with the wind speed and ambient temperature as input variables, and the rotational speed, pitch angle and power as output variables, a multi-output regression model of each unit under healthy conditions is established as the performance baseline model. The performance baseline model also includes confidence intervals for each of the output variables.

[0032] A multi-output regression model was constructed using Gaussian process regression or random forest regression, with wind speed and ambient temperature as inputs and engine speed, propeller pitch angle, and power as outputs. Confidence intervals for each output variable were also provided. The advantage of this approach is that the baseline model not only characterizes the deterministic relationships between parameters under healthy conditions but also quantifies the uncertainty of the prediction results. This provides a statistical basis for subsequently determining whether actual observations truly deviate from the normal range, avoiding the risk of misjudgment caused by relying solely on single-point differences.

[0033] In one embodiment, step S300 includes: S310: Input the current combination of operating parameters into the performance baseline model, and calculate the standardized residual vector between the actual output value and the model prediction value; S320: Aggregate the standardized residual vectors into a comprehensive deviation index based on Mahalanobis distance or Euclidean distance.

[0034] First, the standardized residual vector between the actual output and the predicted value is calculated using the baseline model. Then, it is aggregated into a comprehensive deviation index using Mahalanobis distance or Euclidean distance. The benefits are that standardization eliminates the influence of differences in the dimensions and fluctuation amplitudes of different parameters on the deviation calculation, making the contribution of each parameter to the deviation comparable. Furthermore, using Mahalanobis distance further considers the correlation between parameters, avoiding double counting or underestimation of deviation due to the inherent correlation structure between parameters. This allows the comprehensive deviation index to more accurately reflect the degree of deviation of the overall operating status from the healthy baseline.

[0035] In one embodiment, after step S300, the method further includes: S301: Smooth the deviation index according to the time sliding window, and calculate the degradation rate based on the time-series change trend of the smoothed deviation index.

[0036] The deviation index is smoothed using a time-sliding window, and the degradation rate is calculated based on the smoothed time-series change trend. The smoothing process suppresses the interference of short-term random fluctuations and noise on the deviation index, making the degradation trend clearer and more discernible. Calculating the degradation rate on this basis helps maintenance personnel determine whether performance degradation is slow or rapid, providing a quantitative reference for determining the urgency of diagnosis and repair.

[0037] In one embodiment, step S500 includes: S510: Construct a training dataset containing historical normal operating state samples and current deterioration state samples, using the deviation of each operating parameter as a feature variable and the deviation index as a target variable; S520: Multiple causal decision trees are constructed using the causal forest algorithm, and each causal decision tree is grown based on random subsampling and random feature subsets; S530: Based on the causal effect estimates of each of the causal decision trees, calculate the average causal effect strength of each of the feature variables on the target variable, and calculate the confidence interval of each of the average causal effect strengths.

[0038] The training data uses parameter bias as a feature and deviation index as the target, ensuring that the causal forest's analysis objective always focuses on "which parameter's bias is driving up the overall deviation," avoiding interference from irrelevant factors. Multiple causal decision trees are grown separately through randomly sampled samples and feature subsets. This reduces overfitting of individual trees to specific samples and allows different trees to capture differences in parameter influence across different operating conditions, resulting in a more robust average causal effect strength. Confidence intervals help filter out truly statistically significant parameters, preventing parameters with high estimation noise from being misjudged as root causes of degradation, thus ensuring the reliability of subsequent ranking and root cause parameter cluster extraction results.

[0039] In one embodiment, step S600 includes: S610: Arrange the causal effect intensity of each operating parameter in descending order; S620: Extract the first few parameters whose cumulative causal effect contribution rate reaches a preset threshold as the root cause parameter cluster; the root cause parameter cluster includes a single parameter or a combination of parameters.

[0040] The causal effect strength is sorted in descending order, and the top few parameters that reach a preset threshold for cumulative causal effect contribution rate are extracted as the root parameter cluster. The beneficial effect is that the sorting process can intuitively reflect the relative impact of each parameter on performance degradation, while extracting the root parameter cluster based on cumulative contribution rate automatically controls the diagnostic scope while ensuring explanatory power. This avoids omitting important factors and prevents the output of a large number of weakly correlated parameters, making the diagnostic results more focused.

[0041] In one embodiment, in response to the root parameter set including a pitch angle parameter, step S600 further includes: S601: Obtain the deviation between the actual value of the pitch angle parameter at the current ambient temperature and the baseline value at the corresponding temperature in the performance baseline model; S602: Compare the deviation value with the temperature-corrected allowable deviation range. When the deviation value exceeds the allowable deviation range, the degradation mode is determined to be a temperature-related pitch angle deviation mode.

[0042] For cases where the root cause parameter cluster includes the pitch angle parameter, a temperature-corrected allowable deviation range is introduced for secondary discrimination. The advantage of this approach is that the pitch angle itself is significantly affected by ambient temperature; without temperature correction, normal temperature-adaptive changes can easily be misjudged as abnormal deviations. By comparing the temperature-corrected allowable deviation range, it is possible to more accurately identify whether a temperature-related pitch angle deviation truly exists, improving the diagnostic accuracy of this specific degradation mode and reducing false alarms.

[0043] In one embodiment, before the step of comparing the current correlation between current operating parameters with the corresponding performance baseline model, the method further includes: S290: For multiple turbines in the same wind farm that are in the same wind speed range and the same ambient temperature range, calculate the deviation index of each turbine. S291: The deviation indicators of each unit are relatively sorted. When the deviation indicator of one unit exceeds the preset position threshold in the relative sorting, the unit is marked as a priority source tracing and diagnosis object.

[0044] Before triggering the absolute threshold judgment for a single unit, multiple turbines within the same wind farm that are in the same wind speed and ambient temperature range are first ranked by their relative deviation, and the units with significantly abnormal rankings are marked first. The beneficial effect is that relative comparison can maintain strong discriminative ability even when the overall operating conditions of the wind farm undergo systematic changes, avoiding the impact on the accuracy of single-unit anomaly identification caused by a general increase or decrease in absolute deviation due to overall shifts in the external environment. It also provides a priority basis for the scheduling of operation and maintenance resources for multiple units.

[0045] In one embodiment, the step of outputting the source tracing diagnostic results further includes: S710: In response to confirming that the unit performance has returned to normal, the normal operating parameter data under the current operating conditions is used as an incremental sample to incrementally update the parameters of the performance baseline model; S720: During the incremental update process, the incremental samples are judged for anomalies based on the model prediction error, and the samples judged to be abnormal are automatically removed.

[0046] After confirming that the unit's performance has returned to normal, the baseline model is incrementally updated using current normal operating data, and outlier samples are automatically removed during the update process based on prediction errors. This ensures that the baseline model can adaptively adjust to the slow performance evolution of the unit itself and long-term changes in the operating environment, maintaining its effectiveness as a health reference; simultaneously, automatically removing outlier samples prevents degraded data from contaminating the baseline model, ensuring that the baseline always represents the unit's true health status.

[0047] In one embodiment, the step of outputting the source tracing diagnostic results further includes: S790: Generate a diagnostic report based on the source tracing and diagnostic results. The diagnostic report shall include at least the performance degradation warning time, the degradation severity level, the causal effect contribution ranking of each parameter in the root cause parameter cluster, and the corresponding recommended maintenance strategy. S791: Send the diagnostic report to the human-computer interaction interface for display.

[0048] The source tracing and diagnostic results are compiled into a diagnostic report that includes the warning time, the severity level of degradation, the causal contribution ranking of the root cause parameter clusters, and recommended maintenance strategies, and then pushed to the human-computer interaction interface. The beneficial effects are that the structured report can transform complex causal inferences into information that maintenance personnel can directly understand and implement, reducing the time cost of manual analysis and decision-making. Simultaneously, the interface visualization of the diagnostic results improves the timeliness and accuracy of maintenance responses.

[0049] The above are merely optional embodiments of the present invention and do not limit the patent scope of the present invention. All equivalent structural transformations made under the concept of the present invention using the contents of the specification and drawings of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A method for tracing and diagnosing the performance degradation of wind turbine generator sets, characterized in that, include: Collect multi-dimensional operating parameter data of each wind turbine generator in the target wind farm under normal operating conditions. The operating parameters include at least wind speed, rotational speed, pitch angle, ambient temperature, and power. Based on the operating parameter data under the normal operating conditions, a corresponding performance baseline model is constructed for each unit. The performance baseline model is used to characterize the correlation between various operating parameters of the unit under healthy conditions. Real-time collection of current operating parameter data for each unit; comparison of the current correlation between current operating parameters with the corresponding performance baseline model; calculation of the deviation index of the current operating state relative to the performance baseline model. In response to the deviation index exceeding a preset threshold, the source tracing and diagnosis process is triggered; Based on the causal forest algorithm, a causal inference model is constructed with the deviation index as the target variable, and the causal effect strength of each operating parameter on the deviation index is calculated. Based on the ranking of the causal effect intensity, the root cause parameter clusters leading to performance degradation are traced, and the source diagnosis results are output.

2. The method according to claim 1, characterized in that, The step of constructing a corresponding performance baseline model for each unit based on the operating parameter data under the normal operating conditions includes: Using Gaussian process regression or random forest regression, with the wind speed and ambient temperature as input variables and the rotational speed, pitch angle and power as output variables, a multi-output regression model for each unit under healthy conditions is established as the performance baseline model. The performance baseline model also includes confidence intervals for each output variable.

3. The method according to claim 1, characterized in that, The step of comparing the current correlation between current operating parameters with the corresponding performance baseline model and calculating the deviation index of the current operating state relative to the performance baseline model includes: Input the current combination of operating parameters into the performance baseline model, and calculate the standardized residual vector between the actual output value and the model prediction value; The standardized residual vectors are aggregated into a comprehensive deviation index based on Mahalanobis distance or Euclidean distance.

4. The method according to claim 1, characterized in that, After the step of calculating the deviation index of the current operating state relative to the performance baseline model, the method further includes: The deviation index is smoothed using a time sliding window, and the degradation rate is calculated based on the temporal change trend of the smoothed deviation index.

5. The method according to claim 1, characterized in that, The steps for constructing a causal inference model based on the causal forest algorithm, with the deviation index as the target variable, include: Construct a training dataset containing historical normal operating state samples and current deterioration state samples, using the deviation of each operating parameter as a feature variable and the deviation index as a target variable; Multiple causal decision trees are constructed using the causal forest algorithm, and each causal decision tree is grown based on random subsampling and random feature subsets; Based on the causal effect estimates of each of the causal decision trees, the average causal effect strength of each feature variable on the target variable is calculated, and the confidence interval of each average causal effect strength is calculated.

6. The method according to claim 1, characterized in that, The step of tracing the root cause parameter clusters leading to performance degradation based on the ranking results of the causal effect intensity includes: Arrange the causal effect strength of each operating parameter in descending order; The first few parameters whose cumulative causal effect contribution rate reaches a preset threshold are extracted as the root cause parameter cluster; the root cause parameter cluster includes a single parameter or a combination of parameters.

7. The method according to claim 6, characterized in that, In response to the fact that the root cause parameter set includes a pitch angle parameter, the step of tracing the root cause parameter set that leads to performance degradation further includes: Obtain the deviation between the actual value of the pitch angle parameter at the current ambient temperature and the baseline value at the corresponding temperature in the performance baseline model; The deviation value is compared with the allowable deviation range after temperature correction. When the deviation value exceeds the allowable deviation range, the degradation mode is determined to be a temperature-related pitch angle deviation mode.

8. The method according to claim 1, characterized in that, Before the step of comparing the current correlation between the current operating parameters with the corresponding performance baseline model, the method further includes: For multiple turbines in the same wind farm that are in the same wind speed range and the same ambient temperature range, calculate the deviation index of each turbine. The deviation indicators of each unit are relatively sorted. When the deviation indicator of one unit exceeds the preset position threshold in the relative sort, the unit is marked as a priority source tracing and diagnosis object.

9. The method according to any one of claims 1 to 8, characterized in that, Following the step of outputting the source tracing and diagnostic results, the following also includes: In response to the confirmation that the unit performance has returned to normal, the normal operating parameter data under the current operating conditions is used as an incremental sample to incrementally update the parameters of the performance baseline model; During the incremental update process, the incremental samples are anomaly identified based on the model prediction error, and samples identified as anomalies are automatically removed.

10. The method according to any one of claims 1 to 8, characterized in that, Following the step of outputting the source tracing and diagnostic results, the following also includes: A diagnostic report is generated based on the source tracing and diagnostic results. The diagnostic report includes at least the performance degradation warning time, the degradation severity level, the causal effect contribution ranking of each parameter in the root cause parameter cluster, and the corresponding recommended maintenance strategy. The diagnostic report is sent to the human-computer interaction interface for display.

Images