A wind turbine generator fault early warning method, system, device and storage medium

Abstract

The application discloses a wind turbine fault early warning method, system, device and storage medium. The wind turbine fault early warning method comprises the following steps: acquiring to-be-predicted data and a first historical training data set; determining a label value of each historical training data in the first historical training data set based on a historical time sequence multi-channel event alarm code; constructing a second historical training data set based on the first historical training data set and the label value of each historical training data; constructing an initial multi-level fault probability prediction model group; training the initial multi-level fault probability prediction model group based on the second historical training data set to obtain a trained multi-level fault probability prediction model group; and determining a fault early warning result of the to-be-predicted data based on the trained multi-level fault probability prediction model group, thereby improving the accuracy and reliability of fault early warning.

Description

Technical Field

[0001] This application relates to the technical field of fault early warning for wind turbine generators, and in particular to a fault early warning method, system, device and storage medium for wind turbine generators. Background Technology

[0002] As the core equipment of new energy power generation systems, the operating status of wind turbines directly affects the continuity and economy of power output. In actual wind farm operation, the units constantly face harsh environmental challenges such as strong winds, low temperatures, or high humidity, making mechanical components and electrical systems susceptible to progressive damage. Existing fault early warning methods generally use a single predictive model to process monitoring data, but they can only provide binary fault judgment results, failing to provide refined classification of potential risks. This makes it difficult for maintenance personnel to distinguish between different levels of potential faults, resulting in a lack of targeted maintenance strategies. This may lead to over-responding to low-risk warnings, wasting resources, or responding slowly to high-risk hazards, causing equipment damage. Summary of the Invention

[0003] This application aims to at least address the technical problems existing in the prior art. To this end, this application proposes a fault early warning method, system, device, and storage medium for wind turbine generators, capable of providing fault early warning at different alarm levels, thereby improving the reliability and accuracy of fault early warning.

[0004] A first aspect of this application provides a fault early warning method for wind turbine generators, comprising the following steps: Acquire the data to be predicted and the first historical training dataset, wherein the data to be predicted includes the sampled values ​​of several preset detection parameters of the wind turbine at the current moment, and the first historical training dataset includes the historical time-series multi-channel event alarm codes of the wind turbine within a preset historical period and the historical time-series sampled values ​​of the several preset detection parameters; Based on the historical time-series multi-channel event alarm codes, determine the label value of each historical training data in the first historical training dataset; and construct a second historical training dataset based on the first historical training dataset and the label value of each historical training data. An initial multi-level fault probability prediction model group is constructed. Based on the second historical training dataset, the initial multi-level fault probability prediction model group is trained to obtain a trained multi-level fault probability prediction model group. The trained multi-level fault probability prediction model group includes several trained fault probability prediction models, and each trained fault probability prediction model corresponds to a different alarm level. Based on the trained multi-level fault probability prediction model set, the fault warning result of the data to be predicted is determined.

[0005] The fault early warning method for wind turbines according to the embodiments of this application has at least the following beneficial effects: This application acquires the wind turbine's data to be predicted and the first historical training dataset, and constructs a labeled training dataset based on historical time-series multi-channel event alarm codes. It then trains a multi-level fault probability prediction model group capable of hierarchical prediction for different alarm levels. Thus, this application can overcome the limitations of existing single-model early warning, realize multi-level fault early warning for wind turbine fault risks, and enable operation and maintenance personnel to obtain early warning information of different levels in advance, thereby formulating targeted operation and maintenance strategies and significantly improving the accuracy and reliability of fault early warning.

[0006] A second aspect of this application provides a fault early warning system for wind turbine generators, the fault early warning system for wind turbine generators comprising: The data acquisition module is used to acquire the data to be predicted and the first historical training dataset. The data to be predicted includes the sampled values ​​of several preset detection parameters of the wind turbine at the current moment. The first historical training dataset includes the historical time-series multi-channel event alarm codes of the wind turbine within a preset historical period and the historical time-series sampled values ​​of the several preset detection parameters. The second historical training dataset construction module is used to determine the label value of each historical training data in the first historical training dataset based on the historical time series multi-channel event alarm code; and to construct the second historical training dataset based on the first historical training dataset and the label value of each historical training data. The model training module is used to construct an initial multi-level fault probability prediction model group. Based on the second historical training dataset, the initial multi-level fault probability prediction model group is trained to obtain a trained multi-level fault probability prediction model group. The trained multi-level fault probability prediction model group includes several trained fault probability prediction models, and each trained fault probability prediction model corresponds to a different alarm level. The fault warning result determination module is used to determine the fault warning result of the data to be predicted based on the trained multi-level fault probability prediction model group.

[0007] This system acquires the data to be predicted from wind turbines and the first historical training dataset, and constructs a labeled training dataset based on historical time-series multi-channel event alarm codes. It then trains a multi-level fault probability prediction model group capable of hierarchical prediction for different alarm levels. Thus, this application can overcome the limitations of existing single-model early warning, realize multi-level fault early warning for wind turbine fault risks, and enable operation and maintenance personnel to obtain early warning information of different levels in advance, thereby formulating targeted operation and maintenance strategies and significantly improving the accuracy and reliability of fault early warning.

[0008] A third aspect of this application provides an electronic device including at least one processor and a memory for communicatively connecting to the processor; the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to cause the at least one processor to perform a fault warning method for a wind turbine as described in the first aspect of this application.

[0009] A fourth aspect of this application provides a computer-readable storage medium storing computer-executable instructions for causing a computer to perform a fault early warning method for a wind turbine as described in the first aspect of this application.

[0010] It should be noted that the beneficial effects of the third and fourth aspects of this application with respect to the prior art are the same as the beneficial effects of the above-mentioned fault early warning method for wind turbine units with respect to the prior art, and will not be described in detail here.

[0011] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0012] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a flowchart illustrating an embodiment of the fault early warning method for wind turbines provided in this application; Figure 2 This is a schematic diagram comparing the model performance of different early warning time values ​​of an embodiment of the fault early warning method for wind turbines provided in this application; Figure 3 This is a schematic diagram of the structure of an embodiment of the fault early warning system for wind turbines provided in this application; Figure 4 This is a schematic diagram of the structure of an embodiment of the electronic device provided in this application. Detailed Implementation

[0013] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0014] In the description of this application, the use of terms such as "first," "second," etc., is for the purpose of distinguishing technical features only and should not be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated or the order of the technical features indicated.

[0015] In the description of this application, it should be understood that the orientation descriptions, such as up, down, etc., are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application.

[0016] In the description of this application, it should be noted that, unless otherwise explicitly defined, terms such as "setup," "installation," and "connection" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this application in conjunction with the specific content of the technical solution.

[0017] As the core equipment of new energy power generation systems, the operating status of wind turbines directly affects the continuity and economy of power output. In actual wind farm operation, the units constantly face harsh environmental challenges such as strong winds, low temperatures, or high humidity, making mechanical components and electrical systems susceptible to progressive damage. Existing fault early warning methods generally use a single predictive model to process monitoring data, but this only provides binary fault judgment results and cannot finely classify potential risks. This makes it difficult for maintenance personnel to distinguish between different levels of potential faults, resulting in a lack of targeted maintenance strategies. Over-responding to low-risk warnings may lead to resource waste, while delayed responses to high-risk hazards may cause equipment damage.

[0018] To address the aforementioned technical deficiencies, embodiments of this application provide a method, system, device, and storage medium for early warning of wind turbine generators.

[0019] Please see Figure 1 This is a flowchart illustrating a fault early warning method for wind turbines provided in an embodiment of this application. The method is applied to electronic devices, such as servers. Figure 1 As shown, the fault early warning method for this wind turbine includes: Step S101: Obtain the data to be predicted and the first historical training dataset. The data to be predicted includes the sampled values ​​of several preset detection parameters of the wind turbine at the current moment. The first historical training dataset includes the historical time-series multi-channel event alarm codes of the wind turbine within a preset historical period and the historical time-series sampled values ​​of several preset detection parameters. The aforementioned preset detection parameters can be at least one of temperature, acceleration, rotational speed, current, voltage, pressure, and wind speed.

[0020] The aforementioned preset historical period can be a value that is pre-set according to actual needs.

[0021] In step S101, the above-mentioned acquisition of the data to be predicted and the first historical training dataset can be used to acquire the data to be predicted and the first historical training dataset for SCADA (data acquisition and monitoring control system, a computer-based distributed control and power automation monitoring system).

[0022] Step S102: Based on the historical time series multi-channel event alarm codes, determine the label value of each historical training data in the first historical training dataset; and construct the second historical training dataset based on the first historical training dataset and the label value of each historical training data. The aforementioned label values ​​may include a first preset label value and a second preset label value. The first preset label value may be a value pre-set according to actual needs (which can be used to characterize historical training data as faulty samples); the second preset label value may be a value pre-set according to actual needs (which can be used to characterize historical training data as normal samples).

[0023] In step S102, the construction of the second historical training dataset based on the first historical training dataset and the label value of each historical training data can be achieved by combining the first historical training dataset and the label value of each historical training data to obtain the second historical training dataset.

[0024] Step S103: Construct an initial multi-level fault probability prediction model group. Based on the second historical training dataset, train the initial multi-level fault probability prediction model group to obtain a trained multi-level fault probability prediction model group. The trained multi-level fault probability prediction model group includes several trained fault probability prediction models, and each trained fault probability prediction model corresponds to a different alarm level. The aforementioned initial multi-level fault probability prediction model group may include an initial first alarm level fault probability prediction model, an initial second alarm level fault probability prediction model, and an initial third alarm level fault probability prediction model.

[0025] The initial first alarm level fault probability prediction model, the initial second alarm level fault probability prediction model, and the initial third alarm level fault probability prediction model in the above initial multi-level fault probability prediction model group can be random forest classification models.

[0026] The aforementioned set of trained multi-level fault probability prediction models may include a trained first alarm level fault probability prediction model, a trained second alarm level fault probability prediction model, and a trained third alarm level fault probability prediction model.

[0027] Step S104: Based on the trained multi-level fault probability prediction model group, determine the fault warning result of the data to be predicted.

[0028] The aforementioned fault warning results can include first-level alarms, second-level alarms, and third-level alarms. This application acquires the wind turbine's data to be predicted and a first historical training dataset, and constructs a labeled training dataset based on historical time-series multi-channel event alarm codes. This allows for the training of a multi-level fault probability prediction model group capable of hierarchical prediction for different alarm levels. Therefore, this application overcomes the limitations of existing single-model warnings, achieving multi-level fault warnings for wind turbine fault risks. This enables maintenance personnel to obtain different levels of warning information in advance, thereby developing targeted maintenance strategies and significantly improving the accuracy and reliability of fault warnings.

[0029] In some embodiments, steps S201 to S202 may be included before step S102: Step S201: After obtaining the historical time-series alarm code value of each event bit in the historical time-series multi-channel event alarm code, determine the comprehensive score value of each event bit based on the historical time-series alarm code value, wherein the comprehensive score value is a score scalar value used to characterize the event bit. Specifically, the aforementioned historical time-series multi-channel event alarm code includes historical time-series alarm code values ​​for several event bits. For example, a multi-channel event alarm code may include historical time-series alarm code values ​​for the "Event 1 to Event 16" fields, which are used to characterize the status of 16 independent events (Event 1 to Event 16).

[0030] In step S201, the historical time-series alarm code value of each event bit in the historical time-series multi-channel event alarm code can be obtained by extracting the historical time-series alarm code value of each event bit from the historical time-series multi-channel event alarm code.

[0031] In step S201, determining the comprehensive score value for each event bit based on historical time-series alarm code values ​​may include: The total number of alarms is calculated by counting the number of times each event bit has a historical time-series alarm code value of 1 in the historical time-series multi-channel event alarm codes within the preset historical period (the data sampling frequency of the wind turbine is 1 Hz). The total number of historical time-series alarm code values ​​(the total number of 1s and 0s) for each event bit in the historical time-series multi-channel event alarm codes within the preset historical period is taken as the total number of signals. Divide the total number of alarms by the total number of signals to obtain the alarm percentage for each event bit. Based on the alarm percentage, the normalized alarm percentage for each event bit is calculated using the following formula: ; in, The percentage of alarms corresponding to a specific event bit. To determine the optimal alarm percentage for a specific event bit pre-set according to actual needs. The normalized percentage of alarms corresponding to a specific event bit; Given all consecutive alarm segments (each consecutive alarm segment includes several (one or more) historical training data that are temporally consecutive and whose corresponding historical time-series alarm code value is 1), the normalized coefficient of variation for each event bit is calculated using the following formula: ; in, The coefficient of variation for a given event bit. The arithmetic mean of the durations of all pre-calculated consecutive alarm segments. The standard deviation of the duration of all consecutive alarm segments calculated in advance. The normalized coefficient of variation for a given event bit; When calculating the P-value of the historical time-series alarm code for each event bit within a preset historical period using the Mann-Kendall trend test method, the normalized P-value for each event bit is calculated using the following formula: ; in, The normalized P-value corresponding to a specific event bit. The P value corresponding to any event bit; In the case of reconstructing the phase space of the historical time-series alarm code value of each event bit within a preset historical period to obtain the reconstructed alarm code value of each event bit, the permutation entropy of each reconstructed alarm code value is calculated. The normalized permutation entropy corresponding to each event bit is calculated using the following formula: ; in, The normalized permutation entropy corresponding to a certain event bit. The permutation entropy corresponding to a certain event position; The Hurst exponent value of the historical timing alarm code value of each event bit within a preset historical period is calculated using the recalibrated range analysis method. Based on the normalized alarm percentage, normalized coefficient of variation, normalized p-value, normalized permutation entropy, and Hurst exponent, the comprehensive score for each event bit is calculated using the following formula: ; in, This is the overall score value corresponding to a specific event. The first weight value is preset according to actual needs. This is a second weight value pre-set according to actual needs. This is a third weight value pre-set according to actual needs. This is a fourth weight value pre-set according to actual needs. This is a fifth weight value pre-set according to actual needs. This is the Hurst exponent value corresponding to a specific event bit.

[0032] The sum of the first, second, third, fourth, and fifth weight values ​​mentioned above is one.

[0033] Step S202: Based on the comprehensive score value and the preset score threshold, determine the preset event bits, wherein the preset event bits are several event bits in the historical time series multi-channel event alarm code; The aforementioned preset scoring threshold can be a value pre-set according to actual needs.

[0034] In step S202, the above-mentioned determination of the preset event bit based on the comprehensive score value and the preset score threshold can be that the event bit whose comprehensive score value is greater than the preset score threshold is selected from all event bits in the historical time series multi-channel event alarm code, and the preset event bit can be one or more bits.

[0035] Step S102 may include step S203: Step S203: Based on the historical time series multi-channel event alarm code and preset event bits, determine the label value of each historical training data in the first historical training dataset.

[0036] This application first obtains the historical time-series alarm code value for each event bit and determines the comprehensive score value for each event bit based on the historical time-series alarm code value, thereby quantitatively characterizing the correlation between each event bit and the target fault. Subsequently, these comprehensive score values ​​are compared with preset score thresholds to filter out preset event bits that are highly correlated with the target fault. Finally, based on the historical time-series multi-channel event alarm codes and preset event bits, the label value of each historical training data in the first historical training dataset is determined, thereby effectively avoiding interference from irrelevant or low-correlation event bits on the labeling process, significantly improving the accuracy of the labels, and laying a solid foundation for the effective training of the subsequent multi-level fault probability prediction model, thus improving the accuracy and reliability of wind turbine fault early warning. At the same time, by reducing the number of event bits that need to be processed, the computational complexity of the label determination process is reduced, thereby improving the efficiency of wind turbine fault early warning.

[0037] In some embodiments, step S203 may include steps S301 to S303: Step S301: Obtain the event alarm code value corresponding to the preset event bit in the historical time series multi-channel event alarm code corresponding to each historical training data; In step S301, in the above-mentioned acquisition of the historical time series multi-channel event alarm code corresponding to each historical training data, the event alarm code value corresponding to the preset event bit can be obtained by right-shifting the corresponding preset event bit (any preset event bit being extracted above) when extracting the event alarm code value corresponding to any preset event bit, so that the corresponding preset event bit is at the lowest bit, and the event alarm code value corresponding to the lowest bit is used as the event alarm code value of the corresponding preset event bit.

[0038] Step S302: When the event alarm code value is the first preset alarm code value, the first preset label value is used as the label value of the corresponding historical training data. The first preset alarm code value mentioned above can be a value preset according to actual needs, and can be 1.

[0039] Step S303: When the event alarm code value is the second preset alarm code value, the second preset label value is used as the label value of the corresponding historical training data.

[0040] The aforementioned second preset alarm code value can be a value preset according to actual needs, and can be 0.

[0041] This application provides a clear foundation for distinguishing different states in historical training data by pre-setting a first preset label value and a second preset label value. At the same time, after obtaining the event alarm code value corresponding to the preset event bit in the historical time series multi-channel event alarm code corresponding to each historical training data, the first preset label value or the second preset label value is assigned to the corresponding historical training data according to the matching of the alarm code value with the first preset alarm code value or the second preset alarm code value. This ensures the accuracy and consistency of label generation and provides a more accurate data basis for the training of the subsequent multi-level fault probability prediction model, thereby improving the accuracy of wind turbine fault early warning.

[0042] In some embodiments, steps S401 to S403 may be included before step S103: Step S401: Determine the sliding window value based on the historical time-series sampled values ​​in the second historical training dataset and the label value of each historical training data; Step S402: Based on the sliding window value, set a first warning advance time value, a second warning advance time value, and a third warning advance time value, wherein the first warning advance time value is greater than the sliding window value, the second warning advance time value is greater than the first warning advance time value, and the third warning advance time value is greater than the second warning advance time value; In step S402, the above-mentioned setting of the first warning advance time value, the second warning advance time value, and the third warning advance time value based on the sliding window value can be manually set according to actual needs. The first warning advance time value is greater than the sliding window value, the second warning advance time value is greater than the first warning advance time value, and the third warning advance time value is greater than the second warning advance time value.

[0043] Step S403: Based on the sliding window value and historical time-series sampling values, determine the first time-series statistical features of each historical training data using the sliding window method. The first time-series statistical features include the mean of each preset detection parameter within the sliding window, the standard deviation of each preset detection parameter within the sliding window, the minimum value of each preset detection parameter within the sliding window, and the maximum value of each preset detection parameter within the sliding window. In step S403, the first temporal statistical feature of each historical training data determined by the sliding window method based on the sliding window value and historical time-series sampling value can be calculated by using the sliding window method to calculate the mean, standard deviation, minimum and maximum values ​​of each preset detection parameter within the sliding window, as the first temporal statistical feature of the latest corresponding second historical training data within the sliding window (e.g., if the generation time of all second historical training data within the sliding window is from 10:30:30 to 10:50:30, then the latest corresponding second historical training data is the second historical training data generated at 10:50:30).

[0044] Step S103 may include step S404: Step S404: Based on the second historical training dataset and the first time-series statistical features, train the initial multi-level fault probability prediction model group to obtain the trained multi-level fault probability prediction model group.

[0045] This application determines a sliding window value based on historical time-series sampled values ​​and the label values ​​of each historical training data point. This allows the selected window size to adaptively match the fault occurrence patterns of the wind turbine itself, providing a more accurate time scale basis for subsequent feature extraction. Based on this sliding window value, multiple incremental early warning time values ​​are set, providing a clear time-series benchmark for model training at different alarm levels. Then, the first time-series statistical features of each historical training data point are extracted using the sliding window method, particularly the mean, standard deviation, minimum, and maximum values. These features can capture the changing trends and fluctuation ranges of the preset detection parameters over a period of time. Compared to isolated raw sampled values, they can more effectively reflect key information in the fault development process, significantly improving the discriminative power of the model input features. Finally, the initial multi-level fault probability prediction model group is trained using a second historical training dataset containing these rich time-series statistical features, enabling the model to learn more representative evolution patterns, thereby improving the accuracy and reliability of fault early warning.

[0046] In some embodiments, step S401 may include steps S501 to S503: Step S501: Based on the historical time series sampling values ​​and the label values ​​of each historical training data, determine the fault interval time set, wherein each fault interval time in the fault interval time set is used to characterize the time difference between the end of the previous fault and the start of the current fault. In step S501, the fault interval time set determined based on the historical time-series sampling values ​​and the label values ​​of each historical training data can be the set of time differences between two adjacent fault events in the second historical training dataset, calculated when all the second historical training data with the label values ​​of the first preset label values ​​in the second historical training dataset are selected as alarm historical training data and all the second historical training data with the label values ​​of the second preset label values ​​in the second historical training dataset are selected as normal historical training data. Each fault interval time in the fault interval time set is used to characterize the time difference between the end of the previous fault (indicating that the next adjacent historical training data of an alarm historical training data is normal historical training data and the generation time of the next adjacent historical training data is the end time of the previous fault) and the start of the current fault (indicating that the next adjacent historical training data of a normal historical training data is alarm historical training data and the generation time of the next adjacent historical training data is the start time of the current fault).

[0047] Step S502: Sort all fault interval times in the fault interval time set in descending order to obtain a sorted fault interval time set; and select the Nth sorted fault interval time in the sorted fault interval time set as the preset interval time, where N is a constant value preset according to actual needs. Step S503: Determine the sliding window value based on the preset interval time, historical time series sampling values, and the label value of each historical training data.

[0048] In step S503, the sliding window value determined based on the preset interval time, historical time-series sampled values, and the label value of each historical training data can be determined as follows: When obtaining each current fault start time (indicating that the next adjacent historical training data of a normal historical training data is an alarm historical training data, and the generation time of the next adjacent historical training data is the current fault start time), the sliding window value is backtracked by a preset backtracking time value to obtain the backtracking time point corresponding to each current fault start time. The historical time-series sampled values ​​corresponding to each backtracking time point are extracted to obtain a set of steady-state sampled values. All steady-state sampled values ​​corresponding to each preset detection parameter in the set of steady-state sampled values ​​are sorted in descending order to obtain a set of sorted steady-state sampled values ​​corresponding to each preset detection parameter. The Lth sorted steady-state sampled value in the set of sorted steady-state sampled values ​​is selected as the steady-state reference value for each preset detection parameter. All sampled values ​​corresponding to each preset detection parameter of all alarm historical training data are sorted in descending order to obtain a set of sorted sampled values ​​corresponding to each preset detection parameter. The Mth sorted sampled value in the set of sorted sampled values ​​is selected as the corresponding... The abnormal threshold of the preset detection parameters is calculated. The first interval time set of each preset detection parameter is calculated. The first interval times of all the first interval time sets (the first interval times of all preset detection parameters) are sorted in descending order to obtain the sorted total interval time set. The Kth sorted total interval time in the sorted total interval time set is selected as the sliding window value. The preset backtracking time value is 0.5 times the preset interval time. L is a constant value preset according to actual needs. M is a constant value preset according to actual needs. K is a constant value preset according to actual needs. Each first interval time in the first interval time set is used to represent the time difference between the previous steady-state reference value generation time (meaning that the sampled value of the corresponding preset detection parameter in a certain historical training data is the steady-state reference value, and the generation time of a certain historical training data is used as the previous steady-state reference value generation time) and the next abnormal threshold generation time (meaning that the sampled value of the preset detection parameter corresponding to the closest historical training data is greater than or equal to the corresponding abnormal threshold, and the closest historical training data is used as the next abnormal threshold generation time).

[0049] This application determines the fault interval time set based on historical time-series sampled values ​​and the label values ​​of each historical training data. Relying on the historical training data with completed labeling, it can accurately calculate the interval time between two adjacent faults in the historical operation of the target wind turbine. After sorting all fault interval times in descending order, the interval value of the corresponding position is selected as the preset interval time. This can effectively filter out occasional excessively long interval anomalies, avoid extreme abnormal data interfering with the determination of the sliding window size, and ensure that the preset interval can conform to the interval pattern of most fault occurrences. Finally, based on the preset interval time, historical time-series sampled values, and the label values ​​of each historical training data, the sliding window value is determined. This allows the sliding window size to adapt to the actual fault interval pattern of the target wind turbine. It can ensure that the sliding window can cover the complete development trend information before the fault occurs, and will not introduce irrelevant historical data to cause interference due to the excessively large sliding window size. This effectively ensures the accuracy of the subsequently extracted time-series statistical features, provides a more accurate data basis for the training of subsequent multi-level fault probability prediction models, and thus improves the accuracy of fault early warning.

[0050] In some embodiments, step S404 may include steps S601 to S607: Step S601: Select all historical training data in the second historical training dataset whose label values ​​are the first preset label values, and use them as alarm historical training data; Step S602: For the first A second historical training data corresponding to each alarm historical training data in the second historical training dataset, add a label with the value of first-level alarm to obtain the third historical training dataset (the third historical training dataset includes second historical training data with the label of first-level alarm added and second historical training data without the label of first-level alarm added), where A is the total number of second historical training data within the first warning advance time value; Step S603: For the first B second historical training data corresponding to each alarm historical training data in the second historical training dataset, add a label with the value of second-level alarm to obtain the fourth historical training dataset (the fourth historical training dataset includes second historical training data with the label of second-level alarm added and second historical training data without the label of second-level alarm added), where B is the total number of second historical training data within the second warning advance time value; Step S604: For the first C second historical training data corresponding to each alarm historical training data in the second historical training dataset, add a label with the value of third-level alarm to obtain the fifth historical training dataset (the fifth historical training dataset includes the second historical training data with the label of third-level alarm added and the second historical training data without the label of third-level alarm added), where C is the total number of second historical training data within the third warning advance time value; Step S605: Based on the third historical training dataset and the first time-series statistical features, train the initial first alarm level fault probability prediction model to obtain the trained first alarm level fault probability prediction model. In step S605, the initial first alarm level fault probability prediction model is trained based on the third historical training dataset and the first time-series statistical features to obtain a trained first alarm level fault probability prediction model. This can be achieved by inputting the third historical training dataset and the first time-series statistical features into the initial first alarm level fault probability prediction model to iteratively train the initial first alarm level fault probability prediction model until the maximum number of iterations is preset according to actual needs, resulting in a set of iteratively trained first alarm level fault probability prediction models (each iteratively trained first alarm level fault probability prediction model in the set of iteratively trained first alarm level fault probability prediction models is the first alarm level fault probability prediction model obtained in each training session). From the set of iteratively trained first alarm level fault probability prediction models, all iteratively trained first alarm level fault probability prediction models with a recall rate greater than the recall rate threshold preset according to actual needs are selected as the set of first alarm level fault probability prediction models to be selected. From the set of first alarm level fault probability prediction models to be selected, the first alarm level fault probability prediction model with the highest precision is selected as the trained first alarm level fault probability prediction model.

[0051] Step S606: Based on the fourth historical training dataset and the first time-series statistical features, train the initial second alarm level fault probability prediction model to obtain the trained second alarm level fault probability prediction model. In step S606, the calculation process of training the initial second alarm level fault probability prediction model based on the fourth historical training dataset and the first time-series statistical features to obtain the trained second alarm level fault probability prediction model is similar to the calculation process of training the initial first alarm level fault probability prediction model based on the third historical training dataset and the first time-series statistical features in step S605 to obtain the trained first alarm level fault probability prediction model, and will not be repeated here.

[0052] Step S607: Based on the fifth historical training dataset and the first time-series statistical features, train the initial third alarm level fault probability prediction model to obtain the trained third alarm level fault probability prediction model.

[0053] In step S607, the calculation process of training the initial third alarm level fault probability prediction model based on the fifth historical training dataset and the first time-series statistical features to obtain the trained third alarm level fault probability prediction model is similar to the calculation process of training the initial first alarm level fault probability prediction model based on the third historical training dataset and the first time-series statistical features in step S605 to obtain the trained first alarm level fault probability prediction model, and will not be repeated here.

[0054] Specifically, refer to Figure 2 To determine the optimal advance warning amount, warning labels were constructed using 10 advance warning time values ​​of 10 minutes, 15 minutes, 30 minutes, 45 minutes, 60 minutes, 90 minutes, 120 minutes, 180 minutes, 240 minutes, and 360 minutes, and corresponding warning models were trained.

[0055] Depend on Figure 2 It can be seen that as the lead time for warnings increases, the AUC generally rises first and then falls, with accuracy remaining at a high level (greater than 90%) within 30 minutes, and decreasing significantly after 60 minutes. Based on Figure 2 Based on performance analysis, a three-level early warning system was constructed: Level 1 (first alarm level, with an early warning time of 10 minutes), Level 2 (second alarm level, with an early warning time of 30 minutes), and Level 3 (third alarm level, with an early warning time of 60 minutes). Initial fault probability prediction models for the first alarm level, the second alarm level, and the third alarm level were trained respectively.

[0056] This application uses historical training data of alarms as a benchmark, and for the first, second, and third early warning time values, it backtracks and marks the corresponding quantities (A, B, and C) of historical data to construct the third, fourth, and fifth historical training datasets. Subsequently, using these customized datasets and combining them with the first time-series statistical features, the initial first-alarm-level fault probability prediction model, the initial second-alarm-level fault probability prediction model, and the initial third-alarm-level fault probability prediction model are specifically trained. This ensures that each trained fault probability prediction model can accurately learn the fault characteristics of its corresponding risk advance, thereby improving the accuracy and reliability of fault warnings.

[0057] In some embodiments, step S104 may include steps S701 to S705: Step S701: Based on the sliding window value and the data to be predicted, determine the second time-series statistical characteristics of the data to be predicted using the sliding window method; In step S701, the calculation process of determining the second temporal statistical feature of the data to be predicted based on the sliding window value and the data to be predicted using the sliding window method is similar to the calculation process of determining the first temporal statistical feature of each historical training data based on the sliding window value and historical time-series sampling value using the sliding window method in step S403, and will not be repeated here.

[0058] Step S702: Input the data to be predicted and the second time-series statistical features into the trained first alarm level fault probability prediction model to obtain the first alarm level predicted value output by the trained first alarm level fault probability prediction model; input the data to be predicted and the second time-series statistical features into the trained second alarm level fault probability prediction model to obtain the second alarm level predicted value output by the trained second alarm level fault probability prediction model; input the data to be predicted and the second time-series statistical features into the trained third alarm level fault probability prediction model to obtain the third alarm level predicted value output by the trained third alarm level fault probability prediction model. Step S703: If the first alarm level prediction value is the first preset alarm prediction value, the first level alarm is taken as the fault warning result of the data to be predicted. Step S704: When the predicted value of the first alarm level is the second preset alarm prediction value and the predicted value of the second alarm level is the first preset alarm prediction value, the second level alarm is taken as the fault warning result of the data to be predicted. Step S705: When the predicted value of the first alarm level is the second preset alarm prediction value, the predicted value of the second alarm level is the second preset alarm prediction value, and the predicted value of the third alarm level is the first preset alarm prediction value, the third level alarm is taken as the fault warning result of the data to be predicted.

[0059] This application determines the second time-series statistical features of the data to be predicted using a sliding window method based on a sliding window value. This provides accurate and effective feature input for subsequent prediction models at different levels. Subsequently, the data to be predicted and the extracted second time-series statistical features are input into the trained first-alarm-level fault probability prediction model, the trained second-alarm-level fault probability prediction model, and the trained third-alarm-level fault probability prediction model, respectively. This simultaneously obtains prediction results for different early warning times, providing a complete basis for subsequent tiered judgment. Based on this, this application sequentially judges and outputs data in order from high urgency to low urgency. When the predicted value of the first alarm level meets the first preset alarm prediction value requirement, a first-level alarm is directly output. The first alarm level corresponds to a shorter early warning time, representing a closer proximity to the fault occurrence and lower risk. For higher urgency levels, directly outputting a Level 1 alarm allows maintenance personnel to be aware of urgent fault risks immediately and prioritize handling urgent issues. If the predicted value of the Level 1 alarm is the predicted value of the second preset alarm, then the prediction result of the trained Level 2 alarm probability prediction model is evaluated. If the conditions are met, a Level 2 alarm is output. Level 2 alarms correspond to a longer warning lead time, and the urgency level of the risk is lower than that of the Level 1 alarm. This progressive judgment method neither misses fault risks with lower urgency nor disrupts the warning priority of high-urgency faults. It achieves hierarchical fault warning output with different warning lead times, allowing maintenance personnel to clearly and intuitively understand the urgency level of fault risks, facilitating reasonable scheduling of maintenance work. This fully leverages the hierarchical warning advantages of the multi-level fault probability prediction model group, improving the practicality and reliability of fault warnings.

[0060] Additionally, refer to Figure 3 One embodiment of this application provides a fault early warning system for wind turbine generators, including a data acquisition module 1100, a second historical training dataset construction module 1200, a model training module 1300, and a fault early warning result determination module 1400, wherein: The data acquisition module 1100 is used to acquire the data to be predicted and the first historical training dataset. The data to be predicted includes the sampled values ​​of several preset detection parameters of the wind turbine at the current moment. The first historical training dataset includes the historical time series multi-channel event alarm codes of the wind turbine within a preset historical period and the historical time series sampled values ​​of several preset detection parameters. The second historical training dataset construction module 1200 is used to determine the label value of each historical training data in the first historical training dataset based on the historical time series multi-channel event alarm code; and to construct the second historical training dataset based on the first historical training dataset and the label value of each historical training data. The model training module 1300 is used to construct an initial multi-level fault probability prediction model group. Based on the second historical training dataset, the initial multi-level fault probability prediction model group is trained to obtain a trained multi-level fault probability prediction model group. The trained multi-level fault probability prediction model group includes several trained fault probability prediction models, and each trained fault probability prediction model corresponds to a different alarm level. The fault warning result determination module 1400 is used to determine the fault warning result of the data to be predicted based on the trained multi-level fault probability prediction model group.

[0061] This system acquires the data to be predicted from wind turbines and the first historical training dataset, and constructs a labeled training dataset based on historical time-series multi-channel event alarm codes. It then trains a multi-level fault probability prediction model group capable of hierarchical prediction for different alarm levels. Thus, this application can overcome the limitations of existing single-model early warning, realize multi-level fault early warning for wind turbine fault risks, and enable operation and maintenance personnel to obtain early warning information of different levels in advance, thereby formulating targeted operation and maintenance strategies and significantly improving the accuracy and reliability of fault early warning.

[0062] It should be noted that the system embodiments described above are based on the same inventive concept as the method embodiments described above. Therefore, the relevant content of the method embodiments described above is also applicable to the system embodiments described above, and will not be repeated here.

[0063] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with relevant regulations. The acquisition, storage, use and processing of data in the technical solution of this application all comply with the relevant provisions of national laws and regulations.

[0064] like Figure 4 One embodiment of this application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned fault early warning method for wind turbine generators. The electronic device includes: At least one memory; At least one processor; At least one program; The program is stored in memory, and the processor executes at least one program to implement a fault early warning method for wind turbines according to the above embodiments of this disclosure.

[0065] Electronic devices can be any smart terminal, including mobile phones, tablets, personal digital assistants (PDAs), and in-vehicle computers.

[0066] The electronic devices according to embodiments of this application will now be described in detail.

[0067] The processor 1600 can be implemented using a general-purpose central processing unit (CPU), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this disclosure. The memory 1700 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 1700 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 1700 and is called and executed by the processor 1600 to execute a fault early warning method for a wind turbine according to an embodiment of this disclosure.

[0068] The input / output interface 1800 is used to implement information input and output. The communication interface 1900 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 2000 transmits information between various components of the device (e.g., processor 1600, memory 1700, input / output interface 1800, and communication interface 1900); The processor 1600, memory 1700, input / output interface 1800 and communication interface 1900 are connected to each other within the device via bus 2000.

[0069] This disclosure also provides a storage medium, which is a computer-readable storage medium storing computer-executable instructions for causing a computer to execute the above-described wind turbine fault early warning method.

[0070] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0071] The above is a detailed description of the preferred embodiments of this application. However, the embodiments of this application are not limited to the above-described implementation methods. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the embodiments of this application. All such equivalent modifications or substitutions are included within the scope defined by the claims of the embodiments of this application.

Claims

1. A fault early warning method for wind turbine generators, characterized in that, The fault early warning method for the wind turbine includes: Acquire the data to be predicted and the first historical training dataset, wherein the data to be predicted includes the sampled values ​​of several preset detection parameters of the wind turbine at the current moment, and the first historical training dataset includes the historical time-series multi-channel event alarm codes of the wind turbine within a preset historical period and the historical time-series sampled values ​​of the several preset detection parameters; Having obtained the historical time-series alarm code value for each event bit in the historical time-series multi-channel event alarm code, the normalized alarm percentage, normalized coefficient of variation, normalized P-value, normalized permutation entropy, and Hurst exponent value are determined based on the historical time-series alarm code value; and based on the normalized alarm percentage, normalized coefficient of variation, normalized P-value, normalized permutation entropy, and Hurst exponent value, a comprehensive score value for each event bit is determined, wherein the comprehensive score value is a scalar value used to characterize the event bit; Based on the comprehensive score and the preset score threshold, a preset event bit is determined, wherein the preset event bit is a number of event bits in the historical time series multi-channel event alarm code; Based on the historical time-series multi-channel event alarm codes and the preset event bits, the label value of each historical training data in the first historical training dataset is determined, wherein the label value includes a first preset label value and a second preset label value, specifically: Obtain the event alarm code value corresponding to the preset event bit in the historical time series multi-channel event alarm code corresponding to each of the historical training data; When the event alarm code value is a first preset alarm code value, the first preset label value is used as the label value of the corresponding historical training data. When the event alarm code value is the second preset alarm code value, the second preset label value is used as the label value of the corresponding historical training data; and a second historical training dataset is constructed based on the first historical training dataset and the label value of each historical training data. An initial multi-level fault probability prediction model group is constructed. Based on the second historical training dataset, the initial multi-level fault probability prediction model group is trained to obtain a trained multi-level fault probability prediction model group. The trained multi-level fault probability prediction model group includes several trained fault probability prediction models, and each trained fault probability prediction model corresponds to a different alarm level. Based on the trained multi-level fault probability prediction model set, the fault warning result of the data to be predicted is determined.

2. The fault early warning method for wind turbine units according to claim 1, characterized in that, Before training the initial multi-level fault probability prediction model group based on the second historical training dataset to obtain the trained multi-level fault probability prediction model group, the method further includes: Based on the historical time-series sampled values ​​in the second historical training dataset and the label values ​​of each historical training data, the sliding window value is determined; Based on the sliding window value, a first warning advance time value, a second warning advance time value, and a third warning advance time value are set, wherein the first warning advance time value is greater than the sliding window value, the second warning advance time value is greater than the first warning advance time value, and the third warning advance time value is greater than the second warning advance time value; Based on the sliding window value and the historical time-series sampling value, a first time-series statistical feature of each of the historical training data is determined by the sliding window method. The first time-series statistical feature includes the mean of each preset detection parameter within the sliding window, the standard deviation of each preset detection parameter within the sliding window, the minimum value of each preset detection parameter within the sliding window, and the maximum value of each preset detection parameter within the sliding window. The step of training the initial multi-level fault probability prediction model group based on the second historical training dataset to obtain a trained multi-level fault probability prediction model group includes: Based on the second historical training dataset and the first time-series statistical features, the initial multi-level fault probability prediction model group is trained to obtain a trained multi-level fault probability prediction model group.

3. The fault early warning method for wind turbine units according to claim 2, characterized in that, The step of determining the sliding window value based on the historical time-series sampled values ​​in the second historical training dataset and the label value of each historical training data point includes: Based on the historical time-series sampled values ​​and the label values ​​of each of the historical training data, a set of fault interval times is determined, wherein each fault interval time in the set of fault interval times is used to characterize the time difference between the end of the previous fault and the start of the current fault. Sort all the fault interval times in the fault interval time set in descending order to obtain a sorted fault interval time set; and select the Nth sorted fault interval time in the sorted fault interval time set as the preset interval time, where N is a preset constant value. The sliding window value is determined based on the preset interval time, the historical time series sampling value, and the label value of each historical training data.

4. The fault early warning method for wind turbine units according to claim 2, characterized in that, The initial multi-level fault probability prediction model set includes an initial first alarm level fault probability prediction model, an initial second alarm level fault probability prediction model, and an initial third alarm level fault probability prediction model. The trained multi-level fault probability prediction model set includes a trained first alarm level fault probability prediction model, a trained second alarm level fault probability prediction model, and a trained third alarm level fault probability prediction model. The initial multi-level fault probability prediction model set is trained based on the second historical training dataset and the first time-series statistical features to obtain a trained multi-level fault probability prediction model set, including: Select all historical training data in the second historical training dataset whose label values ​​are the first preset label values, and use them as alarm historical training data. For each alarm historical training data in the second historical training dataset, the first A second historical training data are labeled with a first-level alarm value to obtain the third historical training dataset, where A is the total number of second historical training data within the first warning advance time value. For each alarm historical training data in the second historical training dataset, add a label with the label value of the second level alarm to the first B second historical training data, to obtain the fourth historical training dataset, where B is the total number of second historical training data within the second warning advance time value. For each of the alarm historical training data in the second historical training dataset, add a label with the label value of the third level alarm to the first C second historical training data to obtain the fifth historical training dataset, where C is the total number of second historical training data within the third warning advance time value; Based on the third historical training dataset and the first time-series statistical features, the initial first alarm level fault probability prediction model is trained to obtain the trained first alarm level fault probability prediction model. Based on the fourth historical training dataset and the first time-series statistical features, the initial second alarm level fault probability prediction model is trained to obtain the trained second alarm level fault probability prediction model. Based on the fifth historical training dataset and the first time-series statistical features, the initial third alarm level fault probability prediction model is trained to obtain the trained third alarm level fault probability prediction model.

5. A fault early warning method for wind turbine units according to claim 4, characterized in that, The fault warning results include first-level alarms, second-level alarms, and third-level alarms. The step of determining the fault warning results for the data to be predicted based on the trained multi-level fault probability prediction model set includes: Based on the sliding window value and the data to be predicted, the second time-series statistical characteristics of the data to be predicted are determined using the sliding window method. The data to be predicted and the second time-series statistical features are input into the trained first alarm level fault probability prediction model to obtain the first alarm level predicted value output by the trained first alarm level fault probability prediction model; the data to be predicted and the second time-series statistical features are input into the trained second alarm level fault probability prediction model to obtain the second alarm level predicted value output by the trained second alarm level fault probability prediction model; the data to be predicted and the second time-series statistical features are input into the trained third alarm level fault probability prediction model to obtain the third alarm level predicted value output by the trained third alarm level fault probability prediction model. When the predicted value of the first alarm level is the first preset alarm prediction value, the first alarm level is taken as the fault warning result of the data to be predicted; When the predicted value of the first alarm level is the second preset alarm prediction value, and the predicted value of the second alarm level is the first preset alarm prediction value, the second level alarm is taken as the fault warning result of the data to be predicted. When the predicted value of the first alarm level is the second preset alarm prediction value, the predicted value of the second alarm level is the second preset alarm prediction value, and the predicted value of the third alarm level is the first preset alarm prediction value, the third level alarm is taken as the fault warning result of the data to be predicted.

6. A fault early warning system for wind turbine generators, characterized in that, The fault early warning system for the wind turbine includes: The data acquisition module is used to acquire the data to be predicted and the first historical training dataset. The data to be predicted includes the sampled values ​​of several preset detection parameters of the wind turbine at the current moment. The first historical training dataset includes the historical time-series multi-channel event alarm codes of the wind turbine within a preset historical period and the historical time-series sampled values ​​of the several preset detection parameters. The second historical training dataset construction module is used to, upon obtaining the historical time-series alarm code value of each event bit in the historical time-series multi-channel event alarm code, determine the normalized alarm proportion, normalized coefficient of variation, normalized P-value, normalized permutation entropy, and Hurst exponent value based on the historical time-series alarm code value; and determine the comprehensive score value of each event bit based on the normalized alarm proportion, the normalized coefficient of variation, the normalized P-value, the normalized permutation entropy, and the Hurst exponent value, wherein the comprehensive score value is a scalar value used to characterize the event bit; Based on the comprehensive score and the preset score threshold, a preset event bit is determined, wherein the preset event bit is a number of event bits in the historical time series multi-channel event alarm code; Based on the historical time-series multi-channel event alarm codes and the preset event bits, the label value of each historical training data in the first historical training dataset is determined, wherein the label value includes a first preset label value and a second preset label value, specifically: Obtain the event alarm code value corresponding to the preset event bit in the historical time series multi-channel event alarm code corresponding to each of the historical training data; When the event alarm code value is a first preset alarm code value, the first preset label value is used as the label value of the corresponding historical training data. When the event alarm code value is the second preset alarm code value, the second preset label value is used as the label value of the corresponding historical training data; and a second historical training dataset is constructed based on the first historical training dataset and the label value of each historical training data; and a second historical training dataset is constructed based on the first historical training dataset and the label value of each historical training data. The model training module is used to construct an initial multi-level fault probability prediction model group. Based on the second historical training dataset, the initial multi-level fault probability prediction model group is trained to obtain a trained multi-level fault probability prediction model group. The trained multi-level fault probability prediction model group includes several trained fault probability prediction models, and each trained fault probability prediction model corresponds to a different alarm level. The fault warning result determination module is used to determine the fault warning result of the data to be predicted based on the trained multi-level fault probability prediction model group.

7. An electronic device, characterized in that, It includes at least one processor and a memory for communicatively connecting to the processor; the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to cause the at least one processor to perform a fault early warning method for a wind turbine as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions for causing a computer to perform a fault early warning method for a wind turbine as described in any one of claims 1 to 5.

Images