Fault prediction method, device and electronic equipment for wind turbine generator set

By dividing the components of the wind turbine into events and calculating probabilities, and using a normal behavior prediction model for fault prediction, the problem of inaccurate fault prediction in existing technologies for wind turbines is solved, thereby improving the accuracy of prediction and power generation efficiency.

CN114077922BActive Publication Date: 2024-11-01HUANENG SHEYANG NEW ENERGY POWER GENERATION CO LTD +2
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202111205488.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-10-15
Publication Date
2024-11-01
Estimated Expiration
2041-10-15

AI Technical Summary

Technical Problem

Existing technologies cannot effectively and accurately predict the types and types of faults that may occur in wind turbines in the future, leading to a decrease in the power generation efficiency of the units.

Method used

By classifying the components of the wind turbine into events, the probability of occurrence of the events to be predicted is obtained, and the failure probability of the wind turbine is calculated using a normal behavior prediction model. Combined with weighted processing, failure prediction is achieved.

Benefits of technology

It improves the accuracy and reliability of wind turbine fault prediction, provides data support for operation and maintenance, and improves power generation efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114077922B_ABST
    Figure CN114077922B_ABST
Patent Text Reader

Abstract

The present application provides a fault prediction method, device and electronic equipment for a wind turbine, the method comprising: selecting at least one component to be predicted corresponding to the wind turbine, for each component to be predicted, performing at least one round of event division on the component to be predicted to obtain at least one graded event to be predicted, obtaining a first probability of failure of each graded event to be predicted, and obtaining a second probability of failure of the corresponding component to be predicted based on the first probability, determining a third probability of failure of the wind turbine based on the second probability, and performing fault prediction on the wind turbine based on the third probability. The present application can realize the fault prediction function of the wind turbine by calculating the probability combination of different events, and at the same time provide data support for the next operation and maintenance, facilitate the formulation of maintenance plans, and improve the power generation efficiency of the wind turbine.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of power generation, and in particular to a fault prediction method, device and electronic equipment for a wind turbine generator set. Background Art

[0002] At present, wind turbines have been widely used in power generation. However, wind turbines are bound to fail during operation, causing the unit to shut down, thereby reducing the power generation efficiency of the unit. In order to avoid the occurrence of wind turbine failures, related technologies often use methods to analyze and diagnose the current state of wind turbines based on historical data to estimate possible failures.

[0003] However, the methods used in the related art cannot effectively and accurately predict the possible faults and fault types of wind turbines in the future. Therefore, how to achieve true fault prediction and improve the accuracy, reliability and effectiveness of wind turbine fault prediction has become an urgent problem to be solved. Summary of the invention

[0004] The present application provides a fault prediction method, device and electronic equipment for a wind turbine generator set, which are used to achieve true fault prediction and improve the accuracy, reliability and effectiveness of fault prediction of a wind turbine generator set.

[0005] According to a first aspect of the present application, a fault prediction method for a wind turbine is provided, comprising: selecting at least one component to be predicted corresponding to the wind turbine; for each of the components to be predicted, performing at least one round of event division on the components to be predicted to obtain at least one graded event to be predicted; obtaining a first probability of failure of each of the graded events to be predicted, and based on the first probability, obtaining a second probability of failure of the corresponding component to be predicted; based on the second probability, determining a third probability of failure of the wind turbine, and performing fault prediction on the wind turbine based on the third probability.

[0006] In addition, the fault prediction method for a wind turbine generator set according to the above embodiment of the present application may also have the following additional technical features:

[0007] According to an embodiment of the present application, for each of the components to be predicted, at least one round of event division is performed on the components to be predicted to obtain at least one graded event to be predicted, including: obtaining at least one initial event to be predicted for the component to be predicted; starting from the initial event to be predicted, for any current event division round, event division is performed on the division result of the previous round, and the division result of the current event division round and the initial event to be predicted are used as the graded events to be predicted.

[0008] According to one embodiment of the present application, obtaining the first probability of failure of each of the graded events to be predicted includes: obtaining a preset duration and a predicted duration of failure of each of the graded events to be predicted within the preset duration; obtaining the ratio of the target duration to the preset duration, and using the ratio as the first probability.

[0009] According to one embodiment of the present application, obtaining the predicted duration of failure of each of the graded events to be predicted within the preset duration includes: obtaining a first operating parameter corresponding to each of the graded events to be predicted; inputting each of the first operating parameters into the corresponding normal behavior prediction model, outputting a second operating parameter prediction value corresponding to the graded event to be predicted, and obtaining a prediction residual based on the second operating parameter prediction value; in response to the prediction residual being greater than or equal to a preset residual threshold, obtaining a time step corresponding to the normal behavior prediction model, and obtaining the prediction duration based on the time step and the second operating parameter prediction value.

[0010] According to one embodiment of the present application, the prediction method also includes: inputting all of the graded events to be predicted and the corresponding first operating parameters into the same normal behavior model in sequence, so as to output the second operating parameter prediction value corresponding to each of the graded events to be predicted in sequence.

[0011] According to one embodiment of the present application, the prediction method also includes: obtaining a first weight and weighting all the first probabilities belonging to the same component to be predicted according to the first weight, and using the weighted result as the second probability; obtaining a second weight and weighting all the second probabilities according to the second weight, and using the weighted result as the third probability.

[0012] According to one embodiment of the present application, a training method for a normal behavior prediction model includes: obtaining any sample-graded event to be predicted of any component to be predicted corresponding to a wind turbine generator set, and obtaining a first operating parameter sample value corresponding to the sample-graded event to be predicted, wherein the sample-graded event to be predicted includes a labeled second operating parameter sample value; inputting the first operating parameter sample value into the corresponding initial normal behavior prediction model, and the initial normal behavior prediction model outputs a second operating parameter training value corresponding to the sample-graded event to be predicted; obtaining a training residual based on the second operating parameter training value and the second operating parameter sample value, and adjusting the model parameters of the initial normal behavior prediction model based on the training residual until the training end condition is met to obtain the target normal behavior prediction model.

[0013] According to one embodiment of the present application, obtaining the sample data corresponding to the sample graded event to be predicted includes: obtaining a preset collection time, and obtaining an intermediate first operating parameter sample value corresponding to the sample graded event to be predicted according to the preset collection time; and preprocessing the intermediate first operating parameter sample value to obtain the first operating parameter sample value.

[0014] According to one embodiment of the present application, the obtaining of intermediate sample data corresponding to the sample-graded event to be predicted according to the preset collection time includes: obtaining the initial first operating parameter sample value and the non-first operating parameter sample value corresponding to the sample-graded event to be predicted according to the preset collection time; selecting any initial first operating parameter sample value and any non-first operating parameter sample value, and obtaining the degree of correlation between any initial first operating parameter sample value and any first operating parameter sample value; in response to the degree of correlation satisfying a degree of correlation threshold, using any initial first operating parameter sample value as the intermediate first operating parameter sample value.

[0015] According to one embodiment of the present application, the training residual corresponding to the sample graded event to be predicted is output by the initial normal behavior prediction model, including: determining at least two feature extraction dimensions, and for each of the feature extraction dimensions, performing feature extraction on the first operating parameter sample value by the corresponding feature extraction network in the initial normal behavior prediction model to obtain multidimensional feature information corresponding to the first operating parameter sample value; fusing the feature information of all dimensions in the multidimensional feature information to obtain multidimensional fused feature information; and inputting the multidimensional fused feature information into the prediction network in the initial normal behavior prediction model to output the second operating parameter training value.

[0016] In addition, the fault prediction device for a wind turbine generator set according to the above embodiment of the present application may also have the following additional technical features:

[0017] According to one embodiment of the present application, the first acquisition module is also used to: acquire at least one initial event to be predicted for the component to be predicted; starting from the initial event to be predicted, for any current event division round, perform event division on the division result of the previous round, and use the division result of the current event division round and the initial event to be predicted as the graded event to be predicted.

[0018] According to one embodiment of the present application, the first acquisition module and the second acquisition module are also used to: acquire at least one initial event to be predicted for the component to be predicted; starting from the initial event to be predicted, for any current event division round, perform event division on the division result of the previous round, and use the division result of the current event division round and the initial event to be predicted as the graded event to be predicted.

[0019] According to one embodiment of the present application, the second acquisition module is also used to: obtain the first operating parameter corresponding to each of the graded events to be predicted; input each of the first operating parameters into the corresponding normal behavior prediction model, output the second operating parameter prediction value corresponding to the graded event to be predicted, and obtain the prediction residual according to the second operating parameter prediction value; in response to the prediction residual being greater than or equal to a preset residual threshold, obtain the time step corresponding to the normal behavior prediction model, and obtain the prediction duration according to the time step and the second operating parameter prediction value.

[0020] According to one embodiment of the present application, the second acquisition module is also used to: input all of the graded events to be predicted and the corresponding first operating parameters into the same normal behavior model in sequence, so as to output the second operating parameter prediction value corresponding to each of the graded events to be predicted in sequence.

[0021] According to one embodiment of the present application, the fault prediction module is also used to: obtain a first weight and, based on the first weight, perform weighted processing on all the first probabilities belonging to the same component to be predicted, and use the weighted result as the second probability; obtain a second weight and, based on the second weight, perform weighted processing on all the second probabilities, and use the weighted result as the third probability.

[0022] According to the third aspect of the present application, a normal behavior prediction model training device is provided, including: a first acquisition module, used to acquire any sample-graded event to be predicted of any component to be predicted corresponding to a wind turbine generator set, and acquire a first operating parameter sample value corresponding to the sample-graded event to be predicted, wherein the sample-graded event to be predicted includes a labeled second operating parameter sample value; an output module, used to input the first operating parameter sample value into the corresponding initial normal behavior prediction model, and the initial normal behavior prediction model outputs the second operating parameter training value corresponding to the sample-graded event to be predicted; a second acquisition module, used to acquire a training residual according to the second operating parameter training value and the second operating parameter sample value, and adjust the model parameters of the initial normal behavior prediction model according to the training residual until the training end condition is met to obtain the target normal behavior prediction model.

[0023] In addition, a normal behavior prediction model training device according to the above embodiment of the present application may also have the following additional technical features:

[0024] According to an embodiment of the present application, the first acquisition module is further used to: obtain a preset acquisition time, and obtain the intermediate first operating parameter sample value corresponding to the sample classification to be predicted event according to the preset acquisition time; pre-process the intermediate first operating parameter sample value to obtain the first operating parameter sample value

[0025] According to one embodiment of the present application, the first acquisition module is also used to: obtain the initial first operating parameter sample value and the non-first operating parameter sample value corresponding to the sample classification event to be predicted according to the preset collection time; select any initial first operating parameter sample value and any non-first operating parameter sample value, and obtain the degree of correlation between any initial first operating parameter sample value and any first operating parameter sample value; in response to the degree of correlation satisfying the degree of correlation threshold, use any initial first operating parameter sample value as the intermediate first operating parameter sample value.

[0026] According to one embodiment of the present application, the output module is also used to: determine at least two feature extraction dimensions, and for each of the feature extraction dimensions, have the corresponding feature extraction network in the initial normal behavior prediction model perform feature extraction on the first operating parameter sample value to obtain multidimensional feature information corresponding to the first operating parameter sample value; fuse the feature information of all dimensions in the multidimensional feature information to obtain multidimensional fused feature information; input the multidimensional fused feature information into the prediction network in the initial normal behavior prediction model to output the second operating parameter training value.

[0027] In order to achieve the above-mentioned purpose, the fourth aspect of the present application proposes an electronic device, characterized in that it includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the program, it implements the fault prediction method for the wind turbine set as described in the first aspect or the training method for the normal behavior prediction model as described in the second aspect.

[0028] In order to achieve the above-mentioned objectives, the fourth aspect of the present application proposes a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the fault prediction method for the wind turbine set described in the first aspect or the training method for the normal behavior prediction model described in the second aspect.

[0029] In order to achieve the above-mentioned objectives, the fifth aspect of the present application proposes a computer program product, including a computer program, which, when executed by a processor, implements the fault prediction method for the wind turbine set described in the first aspect or the training method for the normal behavior prediction model described in the second aspect.

[0030] The technical solution provided by the embodiments of the present application includes at least the following beneficial effects:

[0031] The present application provides a fault prediction method for a wind turbine set, which can realize the fault prediction function of the wind turbine set by calculating the probability combination of different events, and at the same time provide data support for the next operation and maintenance, facilitate the formulation of maintenance plans, and improve the power generation efficiency of the wind turbine set.

[0032] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present application.

[0034] Figure 1 A schematic diagram of a flow chart of a fault prediction method for a wind turbine generator set provided in an embodiment of the present application;

[0035] Figure 2 A schematic flow chart of another method for predicting wind turbine faults provided in an embodiment of the present application;

[0036] Figure 3 A schematic flow chart of another method for predicting wind turbine faults provided in an embodiment of the present application;

[0037] Figure 4 A schematic flow chart of another method for predicting wind turbine faults provided in an embodiment of the present application;

[0038] Figure 5 A schematic flow chart of another method for predicting wind turbine faults provided in an embodiment of the present application;

[0039] Figure 6 A flowchart of a method for training a normal behavior prediction model provided in an embodiment of the present application;

[0040] Figure 7 A flowchart of another normal behavior prediction model training method provided in an embodiment of the present application;

[0041] Figure 8 A flowchart of another normal behavior prediction model training method provided in an embodiment of the present application;

[0042] Fig. 9 A flowchart of another normal behavior prediction model training method provided in an embodiment of the present application;

[0043] Fig.10 A schematic diagram of the structure of a fault prediction device for a wind turbine generator set provided in an embodiment of the present application;

[0044] Fig.11 A schematic diagram of the structure of a training device for a normal behavior prediction model provided in an embodiment of the present application;

[0045] Fig.12 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0046] The following is a description of exemplary embodiments of the present application in conjunction with the accompanying drawings, including various details of the embodiments of the present application to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present application. Similarly, for the sake of clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0047] The fault prediction method of the wind turbine generator set of the present application is described in detail below using an embodiment.

[0048] Figure 1 A schematic flow chart of a fault prediction method for a wind turbine generator set provided in an embodiment of the present application.

[0049] like Figure 1 As shown, the fault prediction method for a wind turbine generator set proposed in this embodiment specifically includes the following steps:

[0050] S101. Select at least one component to be predicted corresponding to a wind turbine generator system.

[0051] The component to be predicted may be any component corresponding to the wind turbine, for example, a generator, a gearbox, a yaw system, a pitch control system, and the like.

[0052] S102: For each component to be predicted, perform at least one round of event division on the component to be predicted to obtain at least one graded event to be predicted.

[0053] In the embodiment of the present application, after each component to be predicted is obtained, at least one round of event division may be performed for each component to be predicted to obtain at least one graded event to be predicted.

[0054] It should be noted that the present application does not limit the specific method and rounds of event division, which can be selected according to actual conditions.

[0055] For example, when the component to be predicted is a generator, a round of division may be performed on the generator (component to be predicted) to take the bearing temperature, rotation speed, voltage, current, etc. of the generator as graded events to be predicted.

[0056] S103: Obtain a first probability of failure of each graded event to be predicted, and obtain a second probability of failure of a corresponding component to be predicted based on the first probability.

[0057] It should be noted that the present application does not limit the specific method for obtaining the first probability of failure of each graded event to be predicted, and it can be set according to actual conditions.

[0058] Optionally, the first probability of failure of each graded event to be predicted may be obtained according to the probability prediction model.

[0059] Furthermore, after the first probability is obtained, a second probability of a corresponding component to be predicted failing may be obtained based on the first probability.

[0060] It should be noted that the present application does not limit the specific method for obtaining the second probability of failure of the corresponding component to be predicted based on the first probability, and it can be set according to actual conditions.

[0061] Optionally, a first weight may be set according to each graded event to be predicted, the first probability may be weighted, and the weighted result may be used as the second probability of the corresponding component to be predicted to fail. The first weight may be set according to actual conditions.

[0062] S104: Determine a third probability of a wind turbine generator set failure based on the second probability, and perform a failure prediction on the wind turbine generator set based on the third probability.

[0063] In the embodiment of the present application, after the second probability is obtained, a third probability of a failure of the wind turbine generator set may be determined based on the second probability.

[0064] It should be noted that the specific method of predicting faults of wind turbines according to the third probability is not limited in the present application and can be set according to actual conditions.

[0065] Optionally, after obtaining the second probability, the second probability may be weighted according to the second weight of each component to be predicted of the generator set to obtain a third probability. The second weight may be set according to actual conditions.

[0066] Further, after the third probability is obtained, fault prediction of the wind turbine generator system may be performed according to the third probability.

[0067] The wind turbine fault prediction method provided by the present application selects at least one component to be predicted corresponding to the wind turbine, and for each component to be predicted, performs at least one round of event division on the component to be predicted, so as to obtain at least one graded event to be predicted, obtain the first probability of failure of each graded event to be predicted, and obtain the second probability of failure of the corresponding component to be predicted according to the first probability, and determine the third probability of failure of the wind turbine according to the second probability, and predict the failure of the wind turbine according to the third probability. Therefore, the present application realizes the wind turbine fault prediction function by calculating the probability combination of different events, improves the accuracy and reliability of wind turbine fault prediction, and provides data support for the next operation and maintenance, facilitates relevant personnel to formulate maintenance plans, and improves the power generation efficiency of the wind turbine.

[0068] As a possible implementation, Figure 2 As shown, based on the above steps, in the above step S102, for each component to be predicted, at least one round of event division is performed on the component to be predicted to obtain at least one graded event to be predicted, including the following steps:

[0069] S201: Obtain at least one initial event to be predicted for a component to be predicted.

[0070] It should be noted that, in order to obtain at least one initial event to be predicted for the component to be predicted, the component to be predicted may be split to obtain at least one initial event to be predicted.

[0071] For example, the generator set may be split to obtain at least one initial event to be predicted, such as temperature, speed, voltage, and current.

[0072] S202: Starting from the initial event to be predicted, for any current event division round, perform event division on the division result of the previous round, and use the division result of the current event division round and the initial event to be predicted as the graded event to be predicted.

[0073] In an embodiment of the present application, after obtaining the initial component to be predicted, any current event can be divided into rounds, the division results of the previous round can be divided into events, and the division results of the current event division round and the initial event to be predicted can be used as graded events to be predicted.

[0074] For example, if the initial event to be predicted is of the temperature type, event X1 can be obtained as the graded event to be predicted; if the initial event to be predicted is of the speed type, event X2 can be obtained as the graded event to be predicted; if the initial event to be predicted is of the voltage and current type, event X n As a graded event to be predicted.

[0075] As a possible implementation, Figure 3 As shown, based on the above steps, the specific process of obtaining the first probability of failure of each graded event to be predicted in the above step S103 includes the following steps:

[0076] S301, obtaining a preset duration and a predicted duration of a failure of each graded event to be predicted within the preset duration.

[0077] It should be noted that the preset duration can be set according to actual conditions.

[0078] For example, it can be set to 5h or 10h.

[0079] Among them, the predicted duration of the fault in the graded prediction time is the duration from the beginning to the end of the fault. If the same fault event occurs multiple times within multiple preset time periods in history, the sum of the time is taken as the predicted duration of the fault of the graded event to be predicted.

[0080] It should be noted that, in the present application, the second operating parameter training value can be obtained in a variety of ways. Optionally, each graded event to be predicted can be predicted independently, in which case each graded event to be predicted corresponds to a different normal behavior prediction model; optionally, each graded event to be predicted can be predicted jointly, in which case all graded events to be predicted correspond to the same normal behavior prediction model.

[0081] As a possible implementation method, each graded event to be predicted may be predicted independently, and further, the prediction duration may be obtained according to the second operating parameter training value.

[0082] like Figure 4 As shown, based on the above steps, the specific process of obtaining the predicted duration of each graded event to be predicted within the preset duration in the above step S301 includes the following steps:

[0083] S401: Obtain a first operating parameter corresponding to each graded event to be predicted.

[0084] In the embodiment of the present application, after obtaining the graded time to be predicted, the first operating parameter corresponding to each graded event to be predicted can be obtained, wherein the first operating parameter refers to non-event variables such as wind speed and power.

[0085] S402: Input each first operating parameter into the corresponding normal behavior prediction model, output the second operating parameter prediction value corresponding to the graded event to be predicted, and obtain the prediction residual according to the second operating parameter training value.

[0086] The second operating parameter prediction value refers to event-type variables such as generator temperature.

[0087] In an embodiment of the present application, after obtaining the predicted value of the second operating parameter, the difference between the predicted value of the second operating parameter (predicted value) and the sample value of the second operating parameter (true value) can be obtained as a prediction residual.

[0088] It should be noted that the normal behavior prediction model is any trained convergent model. Preferably, a tree model such as an Xgboost model can be selected for training until convergence.

[0089] S403: In response to the prediction residual being greater than or equal to a preset residual threshold, a time step corresponding to the normal behavior prediction model is obtained, and a prediction duration is obtained according to the time step and the predicted value of the second operating parameter.

[0090] The time step refers to the difference between two previous and subsequent time points.

[0091] It should be noted that the preset residual threshold value can be set according to actual conditions, for example, according to the historical operation of the generator set, or according to the historical data of the normal behavior prediction model.

[0092] In the embodiment of the present application, after the time step and the prediction residual are obtained, the predicted duration can be obtained according to the time step and the prediction residual.

[0093] It should be noted that the present application does not limit the specific method of obtaining the prediction duration based on the time step and the prediction residual, and it can be selected according to actual conditions.

[0094] For example, if the time step is 10 and the prediction residual is 10, the prediction duration can be obtained as 10*10=100.

[0095] As another possible implementation manner, a joint prediction may be performed on each graded event to be predicted, and further, a prediction duration may be obtained according to the prediction residual.

[0096] Optionally, all the graded events to be predicted and the corresponding first operating parameters may be sequentially input into the same normal behavior prediction model to sequentially output the predicted value of the second operating parameter corresponding to each graded event to be predicted.

[0097] In this case, a multi-objective prediction method is adopted, that is, multiple graded events to be predicted share a model to perform multi-objective prediction to obtain the predicted value of the second operating parameter corresponding to each graded event to be predicted.

[0098] S302: Obtain a ratio of the target duration to the preset duration, and use the ratio as a first probability.

[0099] The target duration is the total duration of the failure event of the graded event to be predicted.

[0100] In the embodiment of the present application, after the target duration and the predicted duration are obtained, the ratio of the target duration to the preset duration can be used as the first probability.

[0101] For example, the total prediction duration of the classified predicted events in multiple historical observation periods T can be expressed as t(all), and the classified predicted events X in multiple historical observation periods T can be expressed as i The total predicted duration of failure is expressed as t(x i In this case, the first probability P(X i ) is expressed as P(X i )=t(x i ) / t(all).

[0102] Furthermore, in order to ensure the accuracy of the prediction result, the first probability and the second probability may be weighted.

[0103] As a possible way to achieve this, Figure 5 As shown, the specific steps include:

[0104] S501: Obtain a first weight and perform weighted processing on all first probabilities belonging to the same component to be predicted according to the first weight, and use the weighted result as a second probability.

[0105] It should be noted that different weights may be set for the components to be predicted according to the degree of influence of each component to be predicted on the generator set.

[0106] It should be noted that after obtaining the first weight W i Afterwards, all first probabilities belonging to the same component to be predicted may be weighted according to the first weight, and the weighted result may be used as the second probability.

[0107] Among them, the calculation formula of the second probability P(fault) is:

[0108]

[0109] Among them, P(Xi) is the probability of each event occurring within multiple historical observation periods T, and Wi is the weight of the event on the failure of the component.

[0110] S502: Obtain a second weight and perform weighted processing on all second probabilities according to the second weight, and use the weighted result as a third probability.

[0111] It should be noted that after the second probability is obtained, the probability can be weighted according to the second weight, and the weighted result is used as the third probability.

[0112] Furthermore, after the third probability is obtained, fault prediction can be performed on the wind turbine generator set.

[0113] Therefore, the fault prediction method for wind turbines proposed in the present application divides the generator set into different components to be predicted, and then divides the components to be predicted into graded components to be predicted. By calculating the probability combination of different events, the probability of failure of the components to be predicted in the future is judged, and the operating parameters that may cause the failure are given, thereby providing data support for the next operation and maintenance, facilitating the formulation of maintenance plans, and improving the power generation efficiency of the generator set.

[0114] The following example illustrates the training method of the normal behavior prediction model of the present application.

[0115] Figure 6 A flowchart of a method for training a normal behavior prediction model provided in an embodiment of the present application.

[0116] S601. Obtain any sample-graded event to be predicted of any component to be predicted corresponding to the wind turbine generator set, and obtain a first operating parameter sample value corresponding to the sample-graded event to be predicted, wherein the sample-graded event to be predicted includes a marked second operating parameter sample value.

[0117] Among them, the first operating parameter sample value corresponding to the sample-classified event to be predicted can be any non-event variable, such as wind speed, power, etc.

[0118] It should be noted that, in order to further improve the training effect, after obtaining the first operating parameter sample value, the first operating parameter sample value may be pre-processed such as normalization.

[0119] S602: Input the first operating parameter sample value into the corresponding initial normal behavior prediction model, and the initial normal behavior prediction model outputs the second operating parameter training value corresponding to the sample graded event to be predicted.

[0120] S603: Obtain training residuals according to the second operating parameter training value and the second operating parameter sample value, and adjust model parameters of the initial normal behavior prediction model according to the training residuals until a training end condition is met to obtain a target normal behavior prediction model.

[0121] In the embodiment of the present application, after obtaining the sample-level event to be predicted and the sample value of the first operating parameter, the sample value of the first operating parameter can be input into the corresponding initial normal behavior prediction model to output the second operating parameter training value corresponding to the sample-level event to be predicted. Further, the training residual can be obtained according to the second operating parameter training value and the labeled second operating parameter sample value. Among them, the initial normal behavior prediction model is an unconverged model to be trained.

[0122] Furthermore, the model parameters of the initial normal behavior prediction model can be adjusted according to the training residuals to obtain the optimal model parameters through multiple iterative optimization of the model parameters, and then the model can be retrained with the first operating parameter sample values ​​and the optimal model parameters to obtain the optimal model, that is, the target normal behavior prediction model.

[0123] The training stop condition can be set according to the actual situation. For example, the training stop condition can be set to the training residual being less than a preset training residual threshold; or the training stop condition can be set to the number of times the model parameters are adjusted reaching a preset number threshold.

[0124] The training method of the normal behavior prediction model provided in the present application ensures the training effect of the normal behavior prediction model, improves the accuracy, reliability and effectiveness of the output normal behavior prediction results, and lays the foundation for accurately predicting wind turbine faults using the normal behavior prediction model.

[0125] As a possible implementation, Figure 7 As shown, based on the above steps, the specific process of obtaining the sample data corresponding to the sample-classified event to be predicted in the above step S601 includes the following steps:

[0126] S701. Obtain a preset collection time, and obtain an intermediate first operating parameter sample value corresponding to a sample-graded event to be predicted according to the preset collection time.

[0127] It should be noted that the specific method for obtaining the preset collection time is not limited in the present application and can be set according to actual conditions.

[0128] Optionally, the preset collection time may be directly obtained from a user interaction interface of the generator set.

[0129] S702: Preprocess the intermediate first operating parameter sample value to obtain a first operating parameter sample value.

[0130] It should be noted that after obtaining the intermediate first operating parameter sample value, the intermediate first operating parameter sample value may be preprocessed to obtain the first operating parameter sample value. The specific method of preprocessing may be selected according to actual conditions.

[0131] For example, the validity of sensor data can be verified to eliminate abnormal sensor data; the non-operating state of the unit can be cleaned to eliminate the unit shutdown data and unit fault data; the special state of the unit can be cleaned to eliminate the unit power limit data and pitch angle limit data.

[0132] As a possible implementation, Figure 8 As shown, based on the above steps, the specific process of obtaining the intermediate first operating parameter sample value corresponding to the sample-level event to be predicted according to the preset collection time in the above step S701 includes the following steps:

[0133] S801. According to the preset collection time, obtain the initial first operating parameter sample value and the non-first operating parameter sample value corresponding to the sample-classified event to be predicted.

[0134] S802: Select any initial first operating parameter sample value and any non-first operating parameter sample value, and obtain the degree of correlation between any initial first operating parameter sample value and any non-first operating parameter sample value.

[0135] It should be noted that the correlation coefficient between any initial first operating parameter sample value and any non-first operating parameter sample value can be calculated to obtain the degree of association between any initial first operating parameter sample value and any non-first operating parameter sample value, wherein the correlation coefficient calculation formula is:

[0136]

[0137] Where, the first operating parameter sample value is y, the non-first operating parameter sample value is x, Cov(x,y) is the covariance of the two variables, σ x is the standard deviation of the variable, σ y is the standard deviation of the variable.

[0138] S803: In response to the correlation degree satisfying the correlation degree threshold, any initial first operating parameter sample value is used as an intermediate first operating parameter sample value.

[0139] It should be noted that a correlation threshold is pre-set in the present application. If the correlation between any initial first operating parameter sample value and any non-first operating parameter sample value satisfies the correlation threshold, any initial first operating parameter sample value is used as an intermediate first operating parameter sample value, and the intermediate first operating parameter sample value is used as an input variable of the normal behavior prediction model. The correlation threshold can be set according to actual conditions.

[0140] As a possible implementation, Fig. 9 As shown, based on the above steps, the specific process of the initial normal behavior prediction model outputting the second operating parameter training value corresponding to the sample graded event to be predicted in the above step S602 includes the following steps:

[0141] S901. Determine at least two feature extraction dimensions, and for each feature extraction dimension, use a corresponding feature extraction network in an initial normal behavior prediction model to perform feature extraction on a first operating parameter sample value to obtain multidimensional feature information corresponding to the first operating parameter sample value.

[0142] It should be noted that the specific method for obtaining the multi-dimensional feature information corresponding to the first operating parameter sample value is not limited in the present application and can be set according to actual conditions.

[0143] Optionally, given a time window, the statistical features of each input variable within the time window are extracted, such as the maximum value, the minimum value, and the ratio of the standard deviation to the mean. In this case, the feature information corresponding to the statistical dimension can be obtained; given a time window, the trend features within the time window are extracted, such as the upward trend, downward trend, etc. of the data variable. In this case, the feature information corresponding to the trend dimension can be obtained.

[0144] Optionally, multiple input variables can be constructed as a data matrix, and feature extraction can be performed using a convolutional neural network. Multi-dimensional feature information can be obtained through multiple convolution, pooling, convolution, and pooling operations.

[0145] S902: Fusing feature information of dimensions in the multidimensional feature information to obtain multidimensional fused feature information.

[0146] It should be noted that after obtaining the multidimensional feature information, the feature information of the dimensions in the multidimensional feature information may be fused to obtain multidimensional fused feature information, and the multidimensional fused feature information may be used as the input of the normal behavior prediction model.

[0147] S903: Input the multi-dimensional fusion feature information into the prediction network in the initial normal behavior prediction model, using the second operating parameter training value.

[0148] It should be noted that the multi-dimensional fusion feature information is input into the prediction network in the initial normal behavior prediction model, and the second operation parameter training value can be output.

[0149] The training method of the normal behavior prediction model proposed in the present application further ensures the accuracy and reliability of the output results of the target normal behavior prediction model, and further ensures the accuracy and stability of the fault prediction of the wind turbine.

[0150] In order to implement the above embodiment, this embodiment provides a fault prediction device for a wind turbine generator set. Fig.10 A schematic structural diagram of a fault prediction device for a wind turbine set provided in an embodiment of the present application.

[0151] like Fig.10 As shown, the fault prediction device 1000 of the wind turbine generator system includes: a selection module 1001 , a first acquisition module 1002 , a second acquisition module 1003 and a prediction module 1004 .

[0152] A selection module 1001 is used to select at least one component to be predicted corresponding to a wind turbine generator set;

[0153] A first acquisition module 1002 is used to perform at least one round of event division on each of the components to be predicted, so as to obtain at least one graded event to be predicted;

[0154] The second acquisition module 1003 is used to acquire a first probability of failure of each of the graded events to be predicted, and acquire a second probability of failure of the corresponding component to be predicted based on the first probability;

[0155] The prediction module 1004 is used to determine a third probability of failure of the wind turbine generator set according to the second probability, and perform failure prediction on the wind turbine generator set according to the third probability.

[0156] According to one embodiment of the present application, the first acquisition module 1002 is also used to: obtain at least one initial event to be predicted for the component to be predicted; starting from the initial event to be predicted, for any current event division round, perform event division on the division result of the previous round, and use the division result of the current event division round and the initial event to be predicted as the graded event to be predicted.

[0157] According to one embodiment of the present application, the first acquisition module 1002 is also used to: obtain at least one initial event to be predicted for the component to be predicted; starting from the initial event to be predicted, for any current event division round, perform event division on the division result of the previous round, and use the division result of the current event division round and the initial event to be predicted as the graded event to be predicted.

[0158] According to one embodiment of the present application, the second acquisition module 1003 is also used to: obtain the first operating parameter corresponding to each of the graded events to be predicted; input each of the first operating parameters into the corresponding normal behavior prediction model, output the second operating parameter prediction value corresponding to the graded event to be predicted, and obtain the prediction residual according to the second operating parameter prediction value; in response to the prediction residual being greater than or equal to a preset residual threshold, obtain the time step corresponding to the normal behavior prediction model, and obtain the prediction duration according to the time step and the second operating parameter prediction value.

[0159] According to one embodiment of the present application, the second acquisition module 1002 is also used to: sequentially input all of the graded events to be predicted and the corresponding first operating parameters into the same normal behavior model, so as to sequentially output the second operating parameter prediction value corresponding to each of the graded events to be predicted.

[0160] According to one embodiment of the present application, the fault prediction module 1004 is also used to: obtain a first weight and, based on the first weight, perform weighted processing on all the first probabilities belonging to the same component to be predicted, and use the weighted result as the second probability; obtain a second weight and, based on the second weight, perform weighted processing on all the second probabilities, and use the weighted result as the third probability.

[0161] The wind turbine fault prediction device provided in the present application realizes the wind turbine fault prediction function by calculating the probability combination of different events, thereby improving the accuracy and reliability of wind turbine fault prediction, and at the same time provides data support for the next operation and maintenance, facilitating relevant personnel to formulate maintenance plans, thereby improving the power generation efficiency of the wind turbine.

[0162] In order to implement the above embodiment, this embodiment provides a training device for a normal behavior prediction model. Fig.11 A schematic diagram of the structure of a training device for a normal behavior prediction model provided in an embodiment of the present application.

[0163] like Fig.11 As shown, the training device 1100 of the normal behavior prediction model includes: a first acquisition module 1101, an output module 1102, and a second acquisition module 1103.

[0164] The first acquisition module 1101 is used to

[0165] Obtain any sample-graded event to be predicted of any component to be predicted corresponding to the wind turbine generator set, and obtain a first operating parameter sample value corresponding to the sample-graded event to be predicted, wherein the sample-graded event to be predicted includes a marked second operating parameter sample value;

[0166] An output module 1102 is used to input the first operating parameter sample value into the corresponding initial normal behavior prediction model, and the initial normal behavior prediction model outputs the second operating parameter training value corresponding to the sample graded event to be predicted;

[0167] The second acquisition module 1103 is used to obtain the training residual according to the second operating parameter training value and the second operating parameter sample value, and adjust the model parameters of the initial normal behavior prediction model according to the training residual until the training end condition is met to obtain the target normal behavior prediction model.

[0168] According to one embodiment of the present application, the first acquisition module 1101 is also used to: obtain a preset collection time, and obtain an intermediate first operating parameter sample value corresponding to the sample classification event to be predicted according to the preset collection time; preprocess the intermediate first operating parameter sample value to obtain the first operating parameter sample value.

[0169] According to one embodiment of the present application, the first acquisition module 1101 is also used to: obtain the initial first operating parameter sample value and the non-first operating parameter sample value corresponding to the sample classification event to be predicted according to the preset collection time; select any initial first operating parameter sample value and any non-first operating parameter sample value, and obtain the degree of correlation between any initial first operating parameter sample value and any first operating parameter sample value; in response to the degree of correlation satisfying the degree of correlation threshold, use any initial first operating parameter sample value as the intermediate first operating parameter sample value.

[0170] According to one embodiment of the present application, the output module 1102 is further configured to:

[0171] Determine at least two feature extraction dimensions, and for each of the feature extraction dimensions, perform feature extraction on the first operating parameter sample value by a corresponding feature extraction network in the initial normal behavior prediction model to obtain multidimensional feature information corresponding to the first operating parameter sample value;

[0172] Fusing feature information of all dimensions in the multidimensional feature information to obtain multidimensional fused feature information;

[0173] The multi-dimensional fusion feature information is input into the prediction network in the initial normal behavior prediction model to output the second operating parameter training value.

[0174] The training device of the normal behavior prediction model proposed in the present application ensures the training effect of the normal behavior prediction model, improves the accuracy, reliability and effectiveness of the output normal behavior prediction results, and lays the foundation for accurately predicting wind turbine faults using the normal behavior prediction model.

[0175] In order to implement the above embodiment, the present application also proposes an electronic device 2000, such as Fig.12 As shown, it includes: a memory 210, a processor 220, and a computer program stored in the memory 210 and executable on the processor 220. When the processor executes the program, the fault prediction method for the wind turbine set as described in the first aspect or the training method for the normal behavior prediction model as described in the second aspect is implemented.

[0176] In order to implement the above embodiments, the present application proposes a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the fault prediction method for the wind turbine set described in the first aspect or the training method for the normal behavior prediction model described in the second aspect.

[0177] In order to implement the above embodiments, the present application also proposes a computer program product, including a computer program, which, when executed by a processor, implements the fault prediction method for wind turbines described in the first aspect or the training method for the normal behavior prediction model described in the second aspect.

[0178] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this application can be executed in parallel, sequentially or in different orders, as long as the expected results of the technical solution disclosed in this application can be achieved, and this document is not limited here.

[0179] The above specific implementations do not constitute a limitation on the protection scope of this application. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of this application should be included in the protection scope of this application.

Claims

1. A fault prediction method for wind turbines, include: Selecting at least one component to be predicted corresponding to the wind turbine generator set; For each of the components to be predicted, at least one round of event division is performed on the components to be predicted to obtain at least one graded event to be predicted, wherein the graded event to be predicted includes an initial event to be predicted of at least one of a temperature class, a rotation speed class, and a voltage and current class; Obtaining a first probability of failure of each of the graded events to be predicted, and obtaining a second probability of failure of the corresponding component to be predicted based on the first probability; Determine a third probability of failure of the wind turbine generator set according to the second probability, and perform failure prediction on the wind turbine generator set according to the third probability; The obtaining of a first probability of a failure occurring in each of the graded events to be predicted comprises: Obtaining a preset duration and a predicted duration of a failure of each of the graded events to be predicted within the preset duration; Obtaining a ratio of a target duration to the preset duration, and using the ratio as the first probability, wherein the target duration is a total predicted duration of a failure event of a graded event to be predicted; The step of obtaining the predicted duration of each of the graded events to be predicted to fail within the preset duration includes: Acquire a first operating parameter corresponding to each of the graded events to be predicted, wherein the first operating parameter is a non-event variable; Input each of the first operating parameters into the corresponding normal behavior prediction model, output the second operating parameter prediction value corresponding to the graded event to be predicted, and obtain the prediction residual according to the second operating parameter sample value, wherein the second operating parameter is an event class variable; In response to the prediction residual being greater than or equal to a preset residual threshold, a time step corresponding to the normal behavior prediction model is obtained, and the prediction duration is obtained according to the time step and the second operating parameter prediction value.

2. The prediction method according to claim 1, in, The step of performing at least one round of event division on each of the components to be predicted to obtain at least one graded event to be predicted includes: Acquire at least one initial event to be predicted for the component to be predicted; Starting from the initial event to be predicted, for any current event division round, event division is performed on the division result of the previous round, and the division result of the current event division round and the initial event to be predicted are used as the hierarchical event to be predicted.

3. The prediction method according to claim 1, in, Also includes: All of the graded events to be predicted and the corresponding first operating parameters are input into the same normal behavior model in sequence, so as to output the predicted value of the second operating parameter corresponding to each of the graded events to be predicted in sequence.

4. The prediction method according to claim 1, in, Also includes: Obtaining a first weight and, according to the first weight, weighting all the first probabilities belonging to the same component to be predicted, and using the weighted result as the second probability; A second weight is obtained and all the second probabilities are weighted according to the second weight, and the weighted result is used as the third probability.

5. The prediction method according to claim 1, in, The training of the normal behavior prediction model includes: Obtain any sample-graded event to be predicted of any component to be predicted corresponding to the wind turbine generator set, and obtain a first operating parameter sample value corresponding to the sample-graded event to be predicted, wherein the sample-graded event to be predicted includes a marked second operating parameter sample value; Inputting the first operating parameter sample value into the corresponding initial normal behavior prediction model, and having the initial normal behavior prediction model output a second operating parameter training value corresponding to the sample graded event to be predicted; Obtaining a training residual according to the second operating parameter training value and the second operating parameter sample value, and adjusting a model parameter of the initial normal behavior prediction model according to the training residual until a training end condition is met to obtain a target normal behavior prediction model; The obtaining of the first operating parameter sample value corresponding to the sample-graded event to be predicted includes: Obtaining a preset collection time, and obtaining an intermediate first operating parameter sample value corresponding to the sample-classified event to be predicted according to the preset collection time; The outputting of the second operating parameter training value corresponding to the sample graded event to be predicted by the initial normal behavior prediction model includes: Determine at least two feature extraction dimensions, and for each of the feature extraction dimensions, perform feature extraction on the first operating parameter sample value by a corresponding feature extraction network in the initial normal behavior prediction model to obtain multidimensional feature information corresponding to the first operating parameter sample value; Fusing feature information of all dimensions in the multidimensional feature information to obtain multidimensional fused feature information; The multi-dimensional fusion feature information is input into the prediction network in the initial normal behavior prediction model to output the second operating parameter training value.

6. The prediction method according to claim 5, in, The step of obtaining the intermediate first operating parameter sample value corresponding to the sample-graded event to be predicted according to the preset collection time includes: According to the preset collection time, respectively obtain the initial first operating parameter sample value and the non-first operating parameter sample value corresponding to the sample-level event to be predicted; Selecting any initial first operating parameter sample value and any non-first operating parameter sample value, and obtaining a correlation degree between the any initial first operating parameter sample value and the any non-first operating parameter sample value; In response to the correlation degree satisfying a correlation degree threshold, any one of the initial first operating parameter sample values ​​is used as the intermediate first operating parameter sample value.

7. A fault prediction device for a wind turbine generator set, It is characterized in that include: A selection module, used for selecting at least one component to be predicted corresponding to the wind turbine generator set; A first acquisition module is used to perform at least one round of event division on each of the components to be predicted, so as to obtain at least one graded event to be predicted, wherein the graded event to be predicted includes an initial event to be predicted of at least one of a temperature class, a rotation speed class, and a voltage and current class; A second acquisition module is used to acquire a first probability of failure of each of the graded events to be predicted, and acquire a second probability of failure of the corresponding component to be predicted based on the first probability; A prediction module, configured to determine a third probability of failure of the wind turbine generator set according to the second probability, and perform failure prediction on the wind turbine generator set according to the third probability; The second acquisition module is further used for: Obtaining a preset duration and a predicted duration of a failure of each of the graded events to be predicted within the preset duration; Obtaining a ratio of a target duration to the preset duration, and using the ratio as the first probability, wherein the target duration is a total predicted duration of a failure event of a graded event to be predicted; The second acquisition module is further used for: Acquire a first operating parameter corresponding to each of the graded events to be predicted, wherein the first operating parameter is a non-event variable; Input each of the first operating parameters into the corresponding normal behavior prediction model, output the second operating parameter prediction value corresponding to the graded event to be predicted, and obtain the prediction residual according to the second operating parameter sample value, wherein the second operating parameter is an event class variable; In response to the prediction residual being greater than or equal to a preset residual threshold, a time step corresponding to the normal behavior prediction model is obtained, and the prediction duration is obtained according to the time step and the second operating parameter prediction value.

8. The fault prediction device for a wind turbine generator set according to claim 7, in, The first acquisition module is further used for: Acquire at least one initial event to be predicted for the component to be predicted; Starting from the initial event to be predicted, for any current event division round, event division is performed on the division result of the previous round, and the division result of the current event division round and the initial event to be predicted are used as the hierarchical event to be predicted.

9. The fault prediction device for a wind turbine generator set according to claim 7 or 8, in, The second acquisition module is further used for: Acquire at least one initial event to be predicted for the component to be predicted; Starting from the initial event to be predicted, for any current event division round, event division is performed on the division result of the previous round, and the division result of the current event division round and the initial event to be predicted are used as the hierarchical event to be predicted.

10. The fault prediction device for a wind turbine generator set according to claim 9, in, The second acquisition module is further used for: All of the graded events to be predicted and the corresponding first operating parameters are input into the same normal behavior model in sequence, so as to output the predicted value of the second operating parameter corresponding to each of the graded events to be predicted in sequence.

11. The fault prediction device for a wind turbine generator set according to claim 7, in, The prediction module is further used for: Obtaining a first weight and, according to the first weight, weighting all the first probabilities belonging to the same component to be predicted, and using the weighted result as the second probability; A second weight is obtained and all the second probabilities are weighted according to the second weight, and the weighted result is used as the third probability.

12. An electronic device, It is characterized in that include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the fault prediction method for a wind turbine set according to any one of claims 1 to 6 is implemented.

13. A non-transitory computer-readable storage medium storing computer instructions, in, The computer instructions are used to enable the computer to execute the fault prediction method for a wind turbine according to any one of claims 1-6.

14. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the fault prediction method for a wind turbine according to any one of claims 1 to 6.

Citation Information

Patent Citations

  • A method and an arrangement for monitoring the condition of a rotating system

    CN104067011A

  • Variable pitch system fault prediction method based on improved depth random forest algorithm

    CN110298485A

  • Fan generator fault monitoring and diagnosing method based on XGBoost algorithm

    CN112035716A

  • Fault and service life prediction methods and devices of wind generating set

    CN113239534A

  • Failure sign determination method, failure sign determination device and failure sign determination program

    JP2018005393A