Wind turbine generator fault detection method and device and computer device

Abstract

The invention relates to a wind turbine generator fault detection method and device and a computer device. Relates to the field of wind power fault monitoring, and solves the problem of low unit operation efficiency caused by fault detection and response lag. The method comprises the steps that a fault data set and a normal data set are obtained according to historical operation data of the wind turbine generator, the fault data set comprises operation data associated with a fault before the fault occurs, and the normal data set comprises operation data in a normal working state; obtaining at least one difference feature between the fault data set and the normal data set; and generating a fault early warning rule according to the at least one difference feature. The technical scheme provided by the invention is suitable for wind generating set management, realizes early warning of faults such as vibration and the like, and provides support for converting the operation posture in time to prevent serious faults from influencing the operation efficiency.

Description

Technical Field

[0001] This disclosure relates to the field of wind power fault monitoring, and in particular to a method, apparatus and computer device for detecting faults in wind turbine generators. Background Technology

[0002] In recent years, the wind power industry has developed rapidly, with the scale of wind farm construction continuously expanding and project development cycles constantly being shortened. Under the pressure of cost control and construction period constraints, some wind turbine units have experienced problems such as reduced safety factors and insufficient control over manufacturing and assembly quality during the design phase, posing potential risks to the long-term stable operation of the units. At the same time, wind turbine units are rapidly iterating towards larger sizes, with rotor diameters and swept areas continuously increasing, leading to an increasingly unbalanced aerodynamic and mechanical load on the units.

[0003] Under severe wind resource conditions with complex wind conditions and rapid and frequent wind shear, traditional unit control strategies are difficult to adapt to the dynamically changing external conditions. This can easily lead to problems such as control response lag and load distribution imbalance, which in turn can cause typical faults such as abnormal vibration of the unit nacelle and increased impact on the transmission chain. As a result, the unit is forced to operate with limited power or to shut down unplanned, which significantly reduces the power generation efficiency and economic benefits of the wind farm. Summary of the Invention

[0004] To overcome the problems existing in related technologies, this disclosure provides a wind turbine fault detection method, device, and computer device. By analyzing historical operating data, data characteristics before the occurrence of a fault are obtained, and then a method is generated that can provide early warnings and detect wind turbine faults at an early stage. This solves the problem of low unit operating efficiency caused by the lag in fault detection and response, realizes early warning of faults such as vibration, and provides support for timely conversion of operating attitude to avoid serious faults affecting operating efficiency.

[0005] According to a first aspect of the embodiments of this disclosure, a wind turbine fault detection method is provided, comprising: Based on the historical operating data of the wind turbine, a fault dataset and a normal dataset are obtained. The fault dataset contains operating data associated with the fault before it occurred, and the normal dataset contains operating data under normal operating conditions. Obtain at least one difference feature between the fault dataset and the normal dataset; A fault warning rule is generated based on at least one of the aforementioned difference features.

[0006] Furthermore, the historical operating data includes at least one or more of the following parameters of the wind turbine: Wind speed, windward angle, impeller speed, pitch angle, pitch angle pitch speed, power, nacelle forward and backward vibration, nacelle left and right vibration, The steps for obtaining fault datasets and normal datasets based on historical operating data of wind turbine units include: Based on preset fault detection conditions, determine at least one fault time point; The fault dataset is constructed based on the historical operating data within the first time interval before each fault time point, and the normal dataset is constructed based on the historical operating data outside the fault dataset.

[0007] Furthermore, the step of obtaining at least one difference feature between the faulty dataset and the normal dataset includes: Obtain the fault value and normal value of each feature index in the fault dataset and the normal dataset contained in the preset feature index list, wherein the feature index list contains at least one feature index. The feature index whose difference between the fault value and the normal value meets the preset discrimination condition is used as the difference feature.

[0008] Furthermore, the step of obtaining the fault value of each feature index in the fault dataset and the normal value in the normal dataset contained in the preset feature index list includes: Based on the first time interval, the normal dataset is slidably sliced ​​into multiple first sliding windows, and the faulty dataset is slidably sliced ​​into multiple second sliding windows; Based on the list of feature indicators, feature extraction is performed on the first sliding window one by one to obtain the first feature data points corresponding to each first sliding window, and a normal feature dataset is constructed based on all the first feature data points. Based on the list of feature indicators, feature extraction is performed on the second sliding window one by one to obtain the second feature data points corresponding to each second sliding window, and a fault feature dataset is constructed based on all the second feature data points. Statistical analysis is performed on the normal feature dataset and the fault feature dataset to obtain the normal value and the fault value of each feature indicator in the feature indicator list.

[0009] Furthermore, the step of using the feature index whose difference between the fault value and the normal value meets a preset discrimination condition as the difference feature includes: Calculate the effect magnitude of the fault value and the normal value of each of the aforementioned characteristic indicators; The feature index whose effect size value meets the preset discrimination condition is used as the difference feature.

[0010] Furthermore, before the steps of sliding slice the normal dataset into multiple first sliding windows and sliding slice the fault dataset into multiple second sliding windows according to the first time interval, the method further includes: Obtain multiple lists of the aforementioned feature indicators, with different lists of the aforementioned feature indicators associated with different fault types, to indicate subsequent feature extraction according to the different fault types.

[0011] Furthermore, the step of generating fault warning rules based on at least one of the said difference features includes: The optimal fault range of the differential features is obtained by analyzing the values ​​of the differential features through a decision tree. Based on a single difference feature and the corresponding optimal fault interval, construct independent basic rules; Generate the fault warning rule that includes at least one of the independent basic rules.

[0012] Furthermore, the method also includes: Upon detecting an event that matches the fault warning rule, an application decision is generated, which includes control commands for the wind turbine.

[0013] According to a second aspect of the embodiments of this disclosure, a wind turbine fault detection device is provided, comprising: The dataset partitioning module is used to obtain fault datasets and normal datasets based on the historical operating data of the wind turbine. The fault datasets contain operating data associated with the fault before it occurred, and the normal datasets contain operating data under normal operating conditions. The feature filtering module is used to obtain at least one difference feature between the fault dataset and the normal dataset; The fault rule generation module is used to generate fault warning rules based on at least one of the aforementioned difference features.

[0014] According to a third aspect of the embodiments of this disclosure, a computer apparatus is provided, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to execute the aforementioned wind turbine fault detection method.

[0015] The technical solutions provided by the embodiments of this disclosure can include the following beneficial effects: Based on historical operating data of the wind turbine, a fault dataset and a normal dataset are obtained. The fault dataset includes operating data associated with the fault before it occurred, and the normal dataset includes operating data under normal operating conditions. Then, at least one difference feature between the fault dataset and the normal dataset is obtained, and a fault warning rule is generated based on at least one of the difference features. Generating fault warning rules based on data features prior to the fault provides a mechanism for timely intervention in faults and prevention of serious production accidents, solving the problem of low unit operating efficiency caused by delayed fault detection and response.

[0016] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0018] Figure 1 This is a flowchart illustrating a wind turbine fault detection method according to an exemplary embodiment.

[0019] Figure 2 This is a flowchart illustrating yet another wind turbine fault detection method according to an exemplary embodiment.

[0020] Figure 3 This is a flowchart illustrating yet another wind turbine fault detection method according to an exemplary embodiment.

[0021] Figure 4 This is a flowchart illustrating yet another wind turbine fault detection method according to an exemplary embodiment.

[0022] Figure 5 This is a flowchart illustrating yet another wind turbine fault detection method according to an exemplary embodiment.

[0023] Figure 6 This is a flowchart illustrating yet another wind turbine fault detection method according to an exemplary embodiment.

[0024] Figure 7 This is a flowchart illustrating yet another wind turbine fault detection method according to an exemplary embodiment.

[0025] Figure 8 This is a flowchart illustrating yet another wind turbine fault detection method according to an exemplary embodiment.

[0026] Figure 9 This is a block diagram illustrating a wind turbine fault detection device according to an exemplary embodiment.

[0027] Figure 10 This is a block diagram of a dataset partitioning module 901 illustrated according to an exemplary embodiment.

[0028] Figure 11 This is a block diagram of a feature filtering module 902 according to an exemplary embodiment.

[0029] Figure 12 This is a block diagram of a feature extraction submodule 1101 according to an exemplary embodiment.

[0030] Figure 13 This is a block diagram of a feature discrimination determination submodule 1102 according to an exemplary embodiment.

[0031] Figure 14 This is a block diagram of a fault rule generation module 903 according to an exemplary embodiment.

[0032] Figure 15 This is a block diagram illustrating yet another wind turbine fault detection device according to an exemplary embodiment. Detailed Implementation

[0033] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0034] Under severe wind resource conditions with complex wind conditions and rapid and frequent wind shear, traditional unit control strategies are difficult to adapt to the dynamically changing external conditions. This can easily lead to problems such as control response lag and load distribution imbalance, which in turn can cause typical faults such as abnormal vibration of the unit nacelle and increased impact on the transmission chain. As a result, the unit is forced to operate with limited power or to shut down unplanned, which significantly reduces the power generation efficiency and economic benefits of the wind farm.

[0035] In summary, the lack of accurate identification and early warning mechanisms for the early characteristics of nacelle vibration makes it difficult to efficiently detect, identify, and intervene in faults, which has become a prominent technical bottleneck restricting the safe and efficient operation of large wind turbine units.

[0036] To address the aforementioned issues, embodiments of this disclosure provide a wind turbine fault detection method, apparatus, and computer device. By analyzing historical operating data, the method obtains data characteristics prior to the occurrence of a fault, thereby generating a system capable of providing early warnings and detecting wind turbine faults in their early stages. This solves the problem of low turbine operating efficiency caused by delayed fault detection and response, and enables early warning of faults such as vibration. It also provides support for timely conversion of operating attitude to avoid serious faults affecting operating efficiency.

[0037] An exemplary embodiment of this disclosure provides a wind turbine fault detection method, the process of which uses this method to generate fault warning rules for early warning of wind turbine faults is as follows: Figure 1 As shown, it includes: Step 101: Obtain the fault dataset and normal dataset based on the historical operating data of the wind turbine.

[0038] The fault dataset contains operational data associated with the fault prior to its occurrence, while the normal dataset contains operational data under normal operating conditions.

[0039] In this step, historical operational data is collected, the events in which failures occur are identified, and the operational data associated with the failure before the failure time is collected as the failure dataset, while the other data is collected as the normal dataset.

[0040] According to one exemplary implementation, historical operating data for at least one full month is acquired from the wind turbine's Supervisory Control and Data Acquisition (SCADA) system, with a sampling frequency of 1 Hz. The acquired parameters include wind speed, yaw angle, rotor speed (CI_RotorSpeed) or generator speed (CI_PcsMeasuredGeneratorSpeed), pitch angle (CI_PitchPositionA1), pitch angle pitch rate (AtechPitchRate1), power (CI_IprReactivePower), and nacelle fore-and-aft vibration (SubVibNacelle). ForeAftAcceleration), cabin left and right vibration (SubVibNacelleSideSideAcceleration), status words, etc.

[0041] According to one exemplary implementation, historical operational data can be preprocessed before being divided into fault datasets and normal datasets. Preprocessing includes: 1. Outlier removal: Delete invalid data from power-limited operation, sensor malfunctions, and downtime.

[0042] 2. Handling of Null Values: For positions that should have values ​​but have not received valid data, these positions are considered null values ​​(missing or blank) and can be deleted. Alternatively, for data where all sampled points have completely equal values ​​with no variation (standard deviation of 0), deletion can also be used.

[0043] Step 102: Obtain at least one difference feature between the fault dataset and the normal dataset.

[0044] In this step, further analysis is performed on the fault dataset and the normal dataset. For example, the values ​​of feature indicators in the fault dataset and the normal dataset are extracted, and the degree of difference in the values ​​in different datasets is used to determine whether the corresponding feature indicators can be used as differential features.

[0045] Step 103: Generate a fault warning rule based on at least one of the aforementioned difference features.

[0046] In this step, fault warning rules are generated based on one or more difference features that can distinguish the fault dataset from the normal dataset. These features cover one or more of the difference features. Since the fault warning rules are formulated based on data associated with the fault before it occurred, they can reflect the characteristics before or in the early stages of the fault. This allows for early warning and identification of faults before or in their early stages, providing data support for timely intervention.

[0047] The wind turbine fault detection method provided in the embodiments of this disclosure performs feature analysis on the fault dataset before the fault occurs and the normal dataset under normal operating conditions to obtain multiple difference features that can reflect the difference between the fault dataset and the normal dataset. Then, fault detection rules are generated based on the difference features, which solves the problem of low unit operating efficiency caused by fault detection and response lag.

[0048] An exemplary embodiment of this disclosure also provides a method for detecting faults in wind turbine generators, the steps of which include obtaining fault datasets and normal datasets using this method are as follows: Figure 2 As shown, it includes: Step 201: Determine at least one fault time point based on the preset fault detection conditions.

[0049] Historical operating data includes at least one or more of the following parameters of the wind turbine: Wind speed, windward angle, impeller speed, pitch angle, pitch angle pitch speed, power, nacelle forward and backward vibration, nacelle left and right vibration.

[0050] In this step, fault events that meet the preset fault detection conditions are acquired, and the time point of the fault occurrence is determined. Historical operational data covers a long collection period, potentially a whole month or longer. Therefore, historical operational data may cover multiple fault events.

[0051] According to one exemplary implementation, in this step, different fault detection conditions can be set for different fault types. In subsequent processing, dataset partitioning, difference rule acquisition, and fault warning rule generation are performed for different fault types. Alternatively, associations can be established between different fault types to generate fault warning rules covering multiple fault types.

[0052] According to one exemplary implementation, "absolute vibration value > 0.12" can be used as a fault detection condition, and the time point that meets the condition can be extracted as the fault time point.

[0053] Step 202: Construct the fault dataset based on the historical operating data within the first time interval before each fault time point, and construct the normal dataset based on the historical operating data outside the fault dataset.

[0054] In this step, after determining the time point of the failure, data associated with the failure before or in its early stages can be obtained to generate failure datasets and normal datasets. The specific process is as follows: Figure 3 As shown, it includes: Step 301: Obtain the fault value of each feature index in the fault dataset and the normal value in the normal dataset for each feature index contained in the preset feature index list.

[0055] The list of feature indicators includes at least one feature indicator.

[0056] According to one exemplary implementation, the feature index list points to one or more core parameters, and each core parameter corresponds to one or more feature indices. The feature index list clarifies all the feature indices of all core parameters that need to be collected. The feature indices can be the values ​​of the core parameters themselves, that is, indicating the values ​​of the core parameter collected in normal datasets or fault datasets; or they can be statistical values, such as the mean, standard deviation (SD), maximum value (Max), minimum value (Min), and fluctuation ratio (ti) of a core parameter.

[0057] According to an exemplary implementation, the fluctuation ratio ti is calculated using the following expression: ti=SD / Mean.

[0058] According to one exemplary implementation, the characteristic index further includes characteristic indices derived from amplitude operating conditions, including: Maximum amplitude increase (ECD), amplitude increase acceleration (Acc_ECDdiff), amplitude decrease acceleration (InvAcc_ECDdiff).

[0059] Acc_ECDdiff can be calculated using the following expression: Acc_ECDdiff=ECD / ECDsec ECDsec represents the time taken for the amplitude to increase.

[0060] InvAcc_ECDdiff can be calculated using the following expression: InvAcc_ECDdiff=InvECD / InvECDsec, Where InvECD is the maximum decrease in amplitude, and InvECDsec is the time taken for the amplitude to decrease.

[0061] According to one exemplary implementation, the feature index further includes vector decomposition features. For example, performing two-dimensional vector decomposition on the original two-dimensional vectors of speed and yaw angle yields orthogonal components Speed_x (representing the x-axis component of the speed) and Speed_y (representing the y-axis component of the speed), calculated using the following formula: Speed_x=Speed cos(Yaw), Speed_y=Speed sin(Yaw).

[0062] In this step, features are extracted from both the normal and faulty datasets based on the feature indicator list to obtain the faulty and normal values ​​for each feature indicator. The specific process is as follows: Figure 4 As shown, it includes: Step 401: Based on the first time interval, the normal dataset is slidably sliced ​​into multiple first sliding windows, and the fault dataset is slidably sliced ​​into multiple second sliding windows.

[0063] In this step, sliding slices are applied to both the normal and faulty datasets. The step size of the sliding slice is set and adjusted based on factors such as business requirements, data integrity, and computational efficiency in the application scenario. For example, when the business scenario involves monitoring and alarming wind turbine units, high-frequency (e.g., millisecond-level) detection and alarming are generally not required, so the step size can be set to a slightly longer period, such as 10-30 seconds. Regarding data integrity, since the historical operating data used has a large time span, setting the step size too short will result in too many sliding windows, affecting computational efficiency and potentially compromising the integrity of information within a single sliding window.

[0064] According to one exemplary implementation, the step size of the sliding slice is a first time interval, that is, the window time span of the first and second sliding windows obtained after slicing. Considering the needs of the application scenario and the need for compatibility with the data acquisition cycle of the device itself, the first time interval can be set to 30 seconds. With a window length of 30 seconds, the slice slides in chronological order to complete the window division of normal operation data and fault operation data, and then perform statistical analysis on the data within the window.

[0065] Step 402: Extract features from the first sliding window one by one according to the feature index list, obtain the first feature data points corresponding to each first sliding window, and construct a normal feature dataset based on all the first feature data points.

[0066] In this step, feature extraction is performed on each of the first sliding windows. Feature extraction can be performed according to a preset list of feature indicators.

[0067] Based on the feature extraction results of a single first sliding window, a corresponding first feature data point is generated. That is, each first sliding window corresponds to one first feature data point. The first feature data point contains the extraction results of all feature indicators within the current first sliding window.

[0068] After extracting features from all the first sliding windows, all the first feature data points can be integrated to construct a normal feature dataset.

[0069] Step 403: Extract features from the second sliding window one by one according to the feature index list, obtain the second feature data points corresponding to each second sliding window, and construct a fault feature dataset based on all the second feature data points.

[0070] In this step, feature extraction is performed on each of the second sliding windows. Feature extraction can be performed according to a preset list of feature indicators.

[0071] Based on the feature extraction results of a single second sliding window, a corresponding second feature data point is generated. That is, each second sliding window corresponds to one second feature data point. The second feature data point contains the extraction results of all feature indicators within the current second sliding window.

[0072] After extracting features from all the second sliding windows, all the second feature data points can be integrated to construct a fault feature dataset.

[0073] It should be noted that steps 402 and 403 do not have a strict time sequence; they can be performed sequentially or in parallel.

[0074] Step 404: Perform statistical analysis on the normal feature dataset and the fault feature dataset to obtain the normal value and the fault value of each feature indicator in the feature indicator list.

[0075] In this step, for the normal feature dataset, statistical analysis is performed on the values ​​of the same feature index at each first feature data point; for the fault dataset, statistical analysis is performed on the values ​​of the same feature index at each second feature data point.

[0076] Statistical analysis methods include, but are not limited to, taking the average value. The normal value is the average of the characteristic index values ​​across all first characteristic data points, and the fault value is the average of the characteristic index values ​​across all second characteristic data points.

[0077] According to one exemplary implementation, the task of generating fault warning rules can be divided. For example, based on fault type, multiple lists of feature indicators are obtained, with different lists of feature indicators associated with different fault types, to indicate subsequent feature extraction according to different fault types.

[0078] In subsequent processing, different fault types are processed in parallel to generate their respective fault warning rules. Furthermore, fault warning rules for different fault types can be combined to generate fault warning rules that cover more complex application scenarios.

[0079] Step 302: The feature index whose difference between the fault value and the normal value meets the preset discrimination condition is taken as the difference feature.

[0080] In this step, the degree of difference between the fault values ​​and normal values ​​of each characteristic indicator is obtained. By assessing the degree of difference, it can be determined whether the corresponding characteristic indicator reflects the difference between the normal operating state and the fault warning or initial state. The specific process is as follows: Figure 5 As shown, it includes: Step 501: Calculate the effect of the fault value and the normal value of each of the aforementioned characteristic indicators.

[0081] In this step, the effect magnitude of the fault value and the normal value of each of the aforementioned characteristic indicators is calculated.

[0082] According to an exemplary embodiment, the pooled standard deviation is calculated using the following expression: , Where s is the pooled standard deviation. n 1 represents the number of the first feature data points. n 2 represents the number of the second feature data points. s 1 represents the standard deviation among all the first feature data points.s 2 represents the standard deviation among all second feature data points.

[0083] Calculate the effect size of the characteristic index using the following expression: , in, For effect size mean diff The difference between the fault value and the normal value.

[0084] Step 502: The feature index whose effect size value meets the preset discrimination condition is taken as the difference feature.

[0085] In this step, feature indicators are selected based on effect size. According to one exemplary implementation, different values ​​of effect size indicate different grading standards for discrimination; for example, an effect size less than 0.2 is considered to indicate a very small difference, while an effect size greater than 0.8 is considered to indicate a large difference.

[0086] To select characteristic indicators with high discriminative power and representativeness of faults, in this step, the discriminative power condition is set to an effect size greater than 1, i.e., taking... Features with a value greater than 1 are used as differential features.

[0087] A higher discriminant value indicates that fewer feature indicators will be selected, reducing the complexity of subsequent decision tree partitioning and fault warning rule generation, and effectively avoiding misjudgment.

[0088] An exemplary embodiment of this disclosure also provides a wind turbine fault detection method, wherein the process of generating fault early warning rules based on at least one of the aforementioned differential features is as follows: Figure 6 As shown, it includes: Step 601: Analyze the values ​​of the differential features using a decision tree to obtain the optimal fault range of the differential features.

[0089] In this step, for the differential features, a decision tree algorithm is used to split the data intervals of each differential feature, dividing the value of the feature index into multiple continuous data intervals. The number of first feature data points and second feature data points covered in each numerical interval is counted to determine the data interval that best represents the discriminative degree of the differential feature, which is then used as the optimal fault interval.

[0090] Step 602: Construct independent basic rules based on the individual difference features and the corresponding optimal fault intervals.

[0091] In this step, each difference feature and its corresponding optimal fault interval constitute an independent basic rule, which is the smallest unit of the fault warning rule.

[0092] Step 603: Generate the fault warning rule that includes at least one of the independent basic rules.

[0093] In this step, one or more independent basic rules are combined to generate fault warning rules. The multiple independent basic rules contained in the fault warning rules are related by "AND" or "OR".

[0094] According to one exemplary implementation, each fault type undergoes a separate process for acquiring differential features and generating fault warning rules, as described in this disclosure. Subsequently, the independent basic rules or fault warning rules from different fault types can be further integrated into entirely new fault warning rules.

[0095] An exemplary embodiment of this disclosure also provides a method for detecting faults in wind turbine generators, the process of which includes early warning of faults and timely intervention, as follows: Figure 7 As shown, it includes: Step 701: Obtain the fault dataset and normal dataset based on the historical operating data of the wind turbine.

[0096] Step 702: Obtain at least one difference feature between the fault dataset and the normal dataset.

[0097] Step 703: Generate a fault warning rule based on at least one of the aforementioned difference features.

[0098] The implementation principles of steps 701 to 703 are the same as those of steps 101 to 103, and will not be repeated here.

[0099] Step 704: If an event that matches the fault warning rule is detected, an application decision is generated.

[0100] In this step, after generating fault warning rules, the wind turbine is monitored according to these rules. Upon detecting an event that matches the fault warning rules, an application decision is generated to intervene in the current operating status of the wind turbine.

[0101] Application decisions may include control commands for the wind turbine, such as adjusting the pitch angle or limiting power operation.

[0102] Application decisions can also include alarm push notifications, such as triggering maintenance alarms or pushing alarm information to associated platforms, accounts, and terminals.

[0103] According to an exemplary implementation, when performing event detection, the period of the detected data is a first time interval. If the data in the first time interval window, which is the current moment or a recent moment, meets the fault prediction rules, it is determined that an event has occurred, and an application decision can be generated.

[0104] If the characteristic indicators of the real-time historical operating data in the first time interval do not meet any of the fault warning rules, no warning will be triggered, and the wind turbine will operate according to the original control logic and safety chain.

[0105] An exemplary embodiment of this disclosure also provides a method for detecting faults in wind turbine generators, the process of using this method to provide early warning for wind turbine generators is as follows: Figure 8 As shown, it includes: Step 801: Obtain historical operation data.

[0106] In this step, SCADA data for a specific generating unit from January to April 2024 is obtained. This includes second-level data from 13 data points: AtekPitchRate1, CI_PitchPositionA1, CI_WindSpeed2, CI_WindSpeed1, CI_PcsMeasuredGeneratorSpeed, CI_RotorSpeed, CI_PcsMeasuredElectricalTorque, CI_IprReactivePower, CI_YawError2, CI_YawError1, CI_SubVibNacelleForeAftAcceleration, CI_SubVibNacelleSideSideAcceleration, and status words. The data is then processed to select the dataset showing the unit operating normally.

[0107] Step 802: Obtain the fault dataset and normal dataset based on the historical operating data of the wind turbine.

[0108] Using 0.12g as the vibration threshold, and taking the vibration before and after CI_SubVibNacelleForeAftAcceleration as an example, the dataset is partitioned. Feature analysis of the windowed data points is performed, using the feature index list for the AtechPitchRate1 dataset as an example. The feature index list includes: The formulas are: AtechPitchRate1_max_30s, AtechPitchRate1_ECDdiff_30s, AtechPitchRate1_std_30s, AtechPitchRate1_min_30s, AtechPitchRate1_Acc_ECDdiff_30s, AtechPitchRate1_mean_30s, AtechPitchRate1_InvECDsec_30s, AtechPitchRate1_ti_30s, AtechPitchRate1_ECDsec_30s, AtechPitchRate1_InvECDdiff_30s, and AtechPitchRate1_InvAcc_ECDdiff_30s, where "_30s" indicates that the step size for the window sliding calculation is 30s.

[0109] Step 803: Obtain at least one difference feature between the fault dataset and the normal dataset.

[0110] The average value and effect size of the characteristic indicators were calculated, and the characteristic indicators with an effect size greater than 1 were selected as differential characteristics, as shown in Table 1.

[0111] Table 1. Mean values ​​and effect sizes of differential characteristics

[0112] Decision trees are used to obtain the number of normal and fault occurrences for different differential features in different numerical ranges, which can be represented by the number of the first feature data points and the second feature data points involved. Some examples are shown in Table 2.

[0113] Table 2. Statistics on the percentage of data points showing differences in different numerical ranges.

[0114] Step 804: Generate a fault warning rule based on at least one of the aforementioned difference features.

[0115] The following is an example of a fault warning rule: A certain fault rule is as follows: (AtechPitchRate1_ECDdiff_30s<=0.07) &(AtechPitchRate1_ECDdiff_30s>0.03) &(CI_PcsMeasuredGeneratorSpeed_Acc_ECDdiff_30s<= 1.75) &(CI_PcsMeasuredGeneratorSpeed_Acc_ECDdiff_30s>1.25) &(CI_WindSpeed1_ECDdiff_30s>3.60) &(CI_PitchPositionA1_ECDdiff_30s>0.12) Step 805: Formulate the fault rule mechanism logic.

[0116] This step generates application decisions to implement the fault rule mechanism logic. This includes control commands for the wind turbine and push alarms.

[0117] For example, when the data within the past 30 seconds meets any of the warning rules, the system automatically executes control actions (such as adjusting the pitch angle, limiting power operation, or triggering a maintenance alarm).

[0118] Step 806: System Deployment and Real-time Control.

[0119] Fault warning rules can be integrated into the wind turbine main control system, and the decision-making mechanism can be embedded into the SCADA system or edge computing device.

[0120] The data flow is transmitted as follows: Sensor → Edge Gateway (real-time calculation of 30-second sliding window features) → SCADA Server (application of decision rules) → Main Controller.

[0121] This disclosure provides a method for detecting faults in wind turbine units. The resulting early warning mechanism can provide real-time early vibration warnings for operating units, enabling the units to change their operating posture in advance to avoid strong vibrations, thereby reducing power reduction or shutdown of the units, increasing power generation, reducing the failure rate, and making the fault cause logic explainable.

[0122] An exemplary embodiment of this disclosure also provides a wind turbine fault detection device, the structure of which is as follows: Figure 9 As shown, it includes: The dataset partitioning module 901 is used to obtain a fault dataset and a normal dataset based on the historical operating data of the wind turbine. The fault dataset contains the operating data associated with the fault before the fault occurred, and the normal dataset contains the operating data under normal operating conditions. Feature filtering module 902 is used to obtain at least one difference feature between the fault dataset and the normal dataset; The fault rule generation module 903 is used to generate fault warning rules based on at least one of the aforementioned difference features.

[0123] Furthermore, the historical operating data includes at least one or more of the following parameters of the wind turbine: Wind speed, windward angle, impeller speed, pitch angle, pitch angle pitch speed, power, nacelle forward and backward vibration, nacelle left and right vibration, The structure of the dataset partitioning module 901 is as follows: Figure 10 As shown, it includes: The fault location submodule 1001 is used to determine at least one fault time point based on preset fault detection conditions. The data partitioning submodule 1002 is used to construct the fault dataset based on the historical operating data within the first time interval before each fault time point, and to construct the normal dataset based on the historical operating data outside the fault dataset.

[0124] Furthermore, the structure of the feature filtering module 902 is as follows: Figure 11 As shown, it includes: Feature extraction submodule 1101 is used to obtain the fault value and normal value of each feature index in the fault dataset and the normal dataset for each feature index contained in the preset feature index list, wherein the feature index list contains at least one feature index. The feature discrimination determination submodule 1102 is used to take the feature index whose difference between the fault value and the normal value meets the preset discrimination condition as the difference feature.

[0125] Furthermore, the structure of the feature extraction submodule 1101 is as follows: Figure 12 As shown, it includes: The sliding slice submodule 1201 is used to slide slice the normal dataset into multiple first sliding windows and the fault dataset into multiple second sliding windows according to the first time interval; The first feature extraction submodule 1202 is used to extract features from the first sliding window one by one according to the feature index list, obtain the first feature data points corresponding to each first sliding window, and construct a normal feature dataset based on all the first feature data points. The second feature extraction submodule 1203 is used to extract features from the second sliding window one by one according to the feature index list, obtain the second feature data points corresponding to each second sliding window, and construct a fault feature dataset based on all the second feature data points. The statistical analysis submodule 1204 is used to perform statistical analysis on the normal feature dataset and the fault feature dataset to obtain the normal value and the fault value of each feature indicator in the feature indicator list.

[0126] Furthermore, the feature extraction submodule 1101 also includes: The group management submodule 1205 is used to obtain multiple lists of the feature indicators, with different lists of the feature indicators associated with different fault types, so as to indicate that feature extraction is performed according to the different fault types.

[0127] Furthermore, the structure of the feature discrimination determination submodule 1102 is as follows: Figure 13 As shown, it includes: The effect size calculation submodule 1301 is used to calculate the effect size of the fault value and the normal value of each of the characteristic indicators; The discrimination determination submodule 1302 is used to take the feature index whose effect size value meets the preset discrimination condition as the difference feature.

[0128] Furthermore, the structure of the fault rule generation module 903 is as follows: Figure 14 As shown, it includes: The value interval decision submodule 1401 is used to perform value analysis on the difference feature through a decision tree to obtain the optimal fault interval of the difference feature; Independent rule construction submodule 1402 is used to construct independent basic rules based on a single difference feature and the corresponding optimal fault interval; The rule fusion submodule 1403 is used to generate the fault warning rule containing at least one of the independent basic rules.

[0129] Furthermore, the structure of the aforementioned wind turbine fault detection device is as follows: Figure 15 As shown, it also includes: The fault warning intervention module 904 is used to generate an application decision when an event that meets the fault warning rules is detected. The application decision includes control instructions for the wind turbine.

[0130] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0131] The aforementioned devices can be integrated into the main control unit of the wind turbine, embedding the decision-making mechanism into the SCADA system or edge computing device.

[0132] The data flow is transmitted in the following direction: Sensor → Edge Gateway (for feature extraction) → SCADA Server (for decision rule application) → Main Controller.

[0133] An exemplary embodiment of this disclosure also provides a computer apparatus, including: processor; Memory used to store processor-executable instructions; The processor is configured to execute the wind turbine fault detection method provided in the embodiments of this disclosure.

[0134] An exemplary embodiment of this disclosure also provides a non-transitory computer-readable storage medium, wherein when instructions in the storage medium are executed by a computer's processor, the computer is able to perform the wind turbine fault detection method provided in the embodiments of this disclosure.

[0135] This disclosure provides a method, apparatus, and computer device for wind turbine fault detection. Based on historical operating data of the wind turbine, a fault dataset and a normal dataset are obtained. The fault dataset includes operating data associated with the fault prior to its occurrence, and the normal dataset includes operating data under normal operating conditions. At least one difference feature between the fault dataset and the normal dataset is then obtained, and a fault warning rule is generated based on this difference feature. Generating fault warning rules based on data features prior to the fault provides a mechanism for timely intervention and prevention of serious production accidents, addressing the problem of low turbine operating efficiency caused by delayed fault detection and response.

[0136] The established early warning mechanism can provide real-time early warning of vibration for operating units, and can intervene in the operating status of wind turbine units in a timely manner, so that the wind turbine units can change their operating attitude in advance to avoid strong vibration, thereby reducing the power reduction or shutdown of the units, thus increasing power generation, reducing the failure rate, and the logic of the failure cause can be explained.

[0137] Those skilled in the art will also understand that the various illustrative logical blocks and steps listed in the embodiments of this disclosure can be implemented by electronic hardware, computer software, or a combination of both. Whether such functionality is implemented in hardware or software depends on the specific application and the overall system design requirements. Those skilled in the art can implement the described functionality using various methods for each specific application, but such implementation should not be construed as exceeding the scope of protection of the embodiments of this disclosure.

[0138] Furthermore, the term “exemplary” is used herein to mean serving as an example, instance, or illustration. Any aspect or design described herein as “exemplary” is not necessarily to be construed as advantageous compared to other aspects or designs. Rather, the use of the term “exemplary” is intended to present the concept in a concrete manner. As used herein, the term “or” is intended to mean an inclusive “or” rather than an exclusive “or.” That is, unless otherwise specified or clear from the context, “X applies A or B” is intended to mean any of the natural inclusive arrangements. That is, “X applies A or B” satisfies any of the foregoing instances if X applies A; X applies B; or both X applies A and B. Additionally, unless otherwise specified or clear from the context to refer to the singular form, the articles “a” and “an” as used in this application and the appended claims are generally understood to mean “one or more.”

[0139] Similarly, although this disclosure has been shown and described with respect to one or more implementations, equivalent variations and modifications will occur to those skilled in the art upon reading and understanding this specification and the accompanying drawings. This disclosure includes all such modifications and variations and is limited only by the scope of the claims. In particular, with respect to the various functions performed by the components described above (e.g., elements, resources, etc.), unless otherwise indicated, the terminology used to describe such components is intended to correspond to any component (functionally equivalent) that performs the specific function of the described component, even if structurally not equivalent to the disclosed structure. Furthermore, although specific features of this disclosure may have been disclosed with respect to only one of several implementations, such features may be combined with one or more other features of other implementations, as may be desired and advantageous to any given or particular application. Moreover, with regard to the terms “comprising,” “owning,” “having,” “having,” or variations thereof as used in the detailed description or claims, such terms are intended to be inclusive in a manner similar to the term “including.”

[0140] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0141] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for detecting faults in wind turbine generators, characterized in that, include: Based on the historical operating data of the wind turbine, a fault dataset and a normal dataset are obtained. The fault dataset contains operating data associated with the fault before it occurred, and the normal dataset contains operating data under normal operating conditions. Obtain at least one difference feature between the fault dataset and the normal dataset; A fault warning rule is generated based on at least one of the aforementioned difference features.

2. The wind turbine fault detection method according to claim 1, characterized in that, The historical operating data includes at least one or more of the following parameters of the wind turbine: Wind speed, windward angle, impeller speed, pitch angle, pitch angle pitch speed, power, nacelle forward and backward vibration, nacelle left and right vibration, The steps for obtaining fault datasets and normal datasets based on historical operating data of wind turbine units include: Based on preset fault detection conditions, determine at least one fault time point; The fault dataset is constructed based on the historical operating data within the first time interval before each fault time point, and the normal dataset is constructed based on the historical operating data outside the fault dataset.

3. The wind turbine fault detection method according to claim 2, characterized in that, The step of obtaining at least one difference feature between the faulty dataset and the normal dataset includes: Obtain the fault value and normal value of each feature index in the fault dataset and the normal dataset contained in the preset feature index list, wherein the feature index list contains at least one feature index. The feature index whose difference between the fault value and the normal value meets the preset discrimination condition is used as the difference feature.

4. The wind turbine fault detection method according to claim 3, characterized in that, The step of obtaining the fault value and normal value of each feature index in the fault dataset and the normal dataset for each feature index included in the preset feature index list includes: Based on the first time interval, the normal dataset is slidably sliced ​​into multiple first sliding windows, and the faulty dataset is slidably sliced ​​into multiple second sliding windows; Based on the list of feature indicators, feature extraction is performed on the first sliding window one by one to obtain the first feature data points corresponding to each first sliding window, and a normal feature dataset is constructed based on all the first feature data points. Based on the list of feature indicators, feature extraction is performed on the second sliding window one by one to obtain the second feature data points corresponding to each second sliding window, and a fault feature dataset is constructed based on all the second feature data points. Statistical analysis is performed on the normal feature dataset and the fault feature dataset to obtain the normal value and the fault value of each feature indicator in the feature indicator list.

5. The wind turbine fault detection method according to claim 3, characterized in that, The step of using the feature index whose difference between the fault value and the normal value meets the preset discrimination condition as the difference feature includes: Calculate the effect magnitude of the fault value and the normal value of each of the aforementioned characteristic indicators; The feature index whose effect size value meets the preset discrimination condition is used as the difference feature.

6. The wind turbine fault detection method according to claim 3, characterized in that, Before the steps of sliding slice the normal dataset into multiple first sliding windows and sliding slice the fault dataset into multiple second sliding windows according to the first time interval, the method further includes: Obtain multiple lists of the aforementioned feature indicators, with different lists of the aforementioned feature indicators associated with different fault types, to indicate subsequent feature extraction according to the different fault types.

7. The wind turbine fault detection method according to claim 1, characterized in that, The step of generating fault warning rules based on at least one of the difference features includes: The optimal fault range of the differential features is obtained by analyzing the values ​​of the differential features through a decision tree. Based on a single difference feature and the corresponding optimal fault interval, construct independent basic rules; Generate the fault warning rule that includes at least one of the independent basic rules.

8. The wind turbine fault detection method according to claim 1, characterized in that, The method further includes: Upon detecting an event that matches the fault warning rule, an application decision is generated, which includes control commands for the wind turbine.

9. A wind turbine fault detection device, characterized in that, include: The dataset partitioning module is used to obtain fault datasets and normal datasets based on the historical operating data of the wind turbine. The fault datasets contain operating data associated with the fault before it occurred, and the normal datasets contain operating data under normal operating conditions. The feature filtering module is used to obtain at least one difference feature between the fault dataset and the normal dataset; The fault rule generation module is used to generate fault warning rules based on at least one of the aforementioned difference features.

10. A computer device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to perform the wind turbine fault detection method as described in any one of claims 1 to 8.

Images