Training method, processing method and system of wind driven generator fault early warning model

Abstract

The invention provides a training method, a processing method and a system for a fault early warning model of a wind driven generator. The training method comprises the following steps: constructing a data asset model of a target early warning component; acquiring training data of the wind driven generator; inputting the training set into a preset neural network, training to obtain an initial fault early warning model of an optimal hyper-parameter, and determining fault associated data of a first associated device in the training data; performing accumulative summation based on the fault associated data to obtain an HI index corresponding to the target early warning component; setting an initial HI index early warning threshold value of a target early warning component in the initial fault early warning model based on the verification set; setting a test HI index based on the test set, and obtaining an evaluation result of the initial HI index early warning threshold based on the test HI index and the initial HI index early warning threshold; and adjusting the initial fault early warning model based on the evaluation result to obtain a target fault early warning model and a corresponding target HI index early warning threshold. And the overall health condition of the wind driven generator is accurately evaluated.

Description

Technical Field

[0001] The present disclosure relates to the technical field of wind power generation, and in particular to a training method, a processing method and a system for a wind turbine fault warning model. Background Art

[0002] With the rapid development of artificial intelligence and data mining technology, the operation and maintenance management of wind turbines is gradually moving towards intelligence. At present, many digital innovation technologies have been widely used in the fault warning system of wind turbines, which has significantly improved the operation and maintenance efficiency. However, in the face of complex and changeable actual operation scenarios, simple fault warning can no longer meet the needs of the industry.

[0003] For on-site operation and maintenance personnel, it is crucial to accurately locate the cause of the fault and provide immediate and effective operational guidance. Unfortunately, most existing technologies separate fault warning from fault location and lack a systematic solution. More importantly, these technologies often ignore the operability and practicality of warning information, making it difficult for on-site personnel to take appropriate measures quickly after receiving the warning. Therefore, the development of a wind turbine fault warning technology that can integrate fault warning and precise positioning while ensuring that the warning information is highly operational has become a technical problem that needs to be solved urgently in the field of wind power generation. Summary of the invention

[0004] The technical problem to be solved by the present disclosure is to overcome the defect in the prior art of lacking a wind turbine fault warning technology that integrates fault warning and precise positioning, and to provide a training method, processing method, system, equipment, medium and program product for a wind turbine fault warning model.

[0005] The present invention solves the above technical problems through the following technical solutions:

[0006] In a first aspect, a training method for a wind turbine fault warning model is provided, wherein the wind turbine includes a target warning component, the target warning component includes a first associated device, and the training method includes:

[0007] Constructing a data asset model of the target early warning component;

[0008] Acquiring training data of the wind turbine;

[0009] Dividing the training data into a training set, a validation set, and a test set;

[0010] Inputting the training set into a preset neural network, outputting a reconstruction result corresponding to the training set, training to obtain an initial fault warning model with optimal hyperparameters, and determining fault-related data of the first associated component in the training data;

[0011] Based on the fault-related data of the first associated device, cumulative summation is performed to obtain a HI (Health Index) indicator corresponding to the target warning component;

[0012] Setting an initial HI indicator warning threshold corresponding to the target warning component in the initial fault warning model based on the verification set;

[0013] Setting a test HI indicator based on the test set, and obtaining an evaluation result of the initial HI indicator warning threshold based on the test HI indicator and the initial HI indicator warning threshold;

[0014] Adjust the initial fault warning model based on the evaluation result to obtain a target fault warning model and a corresponding target HI indicator warning threshold;

[0015] The target fault warning model is used to output the HI index and the target HI index warning threshold of the target warning component based on the input working data of the target warning component.

[0016] Preferably, the preset neural network is a sparse autoencoder model, and the step of training to obtain an initial fault warning model with optimal hyperparameters comprises:

[0017] The sparse autoencoder model is trained with L1 regularization (L1 norm) to obtain an initial fault warning model with optimal hyperparameters.

[0018] Preferably, the step of obtaining the training data of the wind turbine comprises:

[0019] Acquiring historical operation data of the target warning component;

[0020] The training data is obtained by filtering the historical operation data.

[0021] Preferably, the step of setting the test HI indicator based on the test set, and obtaining the evaluation result of the initial HI indicator warning threshold based on the test HI indicator and the initial HI indicator warning threshold comprises:

[0022] Obtaining a test warning result based on the test HI indicator and the corresponding HI indicator;

[0023] Obtaining actual fault results corresponding to the test set;

[0024] Constructing a confusion matrix based on the actual fault results and the corresponding test warning results;

[0025] The confusion matrix is ​​analyzed to obtain the evaluation result.

[0026] Preferably, the step of dividing the training data into a training set, a validation set and a test set comprises:

[0027] The training data is divided into the training set, the validation set and the test set in a ratio of 6:2:2.

[0028] Preferably, the target warning component includes at least one of a blade, a gearbox, a generator, a converter, a yaw system and a main bearing.

[0029] In a second aspect, a method for processing a wind turbine fault warning is provided, the method comprising:

[0030] Acquiring real-time operating data of the wind turbine;

[0031] Inputting the real-time operation data into a target fault warning model to obtain a real-time HI index and a corresponding HI index threshold of the warning component in the wind turbine;

[0032] Wherein, the target fault warning model is obtained based on the training method of the wind turbine fault warning model described in the first aspect;

[0033] Obtaining a warning value of the warning component based on the real-time HI index and the HI index threshold;

[0034] In response to the warning value of the warning component indicating that the warning component is faulty, an actual observation value of the warning component is acquired, and a fault analysis result of the warning component is obtained based on the warning value and the actual observation value.

[0035] Preferably, the early warning component includes a second associated device, and the step of obtaining a fault analysis result of the early warning component based on the early warning value and the actual observation value includes:

[0036] Calculating a residual value of the warning component based on the warning value and the actual observation value;

[0037] Determining a residual contribution of the warning component based on the residual value;

[0038] The residual contribution represents the contribution of the second associated device in the warning component to the HI index;

[0039] A fault analysis result of the early warning component is obtained based on the residual contribution.

[0040] Preferably, the processing method further comprises:

[0041] Obtaining a residual offset time at which the residual value is greater than a residual threshold;

[0042] A fault analysis result of the early warning component is obtained based on the residual offset time and the residual contribution.

[0043] Preferably, the processing method further comprises:

[0044] Acquire corresponding fault operation data based on the fault analysis result;

[0045] A failure notification is generated based on the failure analysis result and the failure operation data.

[0046] In a third aspect, a training system for a fault warning model of a wind turbine is provided, wherein the wind turbine includes a target warning component, the target warning component includes a first associated device, and the training system includes a data asset construction module, a training data acquisition module, a data allocation module, an initial model training module, an HI index calculation module, an HI index warning threshold setting module, an HI index warning threshold evaluation module, and a target fault model construction module;

[0047] The data asset construction module constructs a data asset model of the target warning component;

[0048] The training data acquisition module is used to acquire the training data of the wind turbine;

[0049] The data allocation module is used to divide the training data into a training set, a validation set and a test set;

[0050] The initial model training module is used to input the training set into a preset neural network, output a reconstruction result corresponding to the training set, train to obtain an initial fault warning model with optimal hyperparameters, and determine the fault-related data of the first associated component in the training data;

[0051] The HI index calculation module is used to obtain the HI index corresponding to the target warning component by accumulating and summing the fault-related data of the first associated device;

[0052] The HI indicator warning threshold setting module is used to set the initial HI indicator warning threshold corresponding to the target warning component in the initial fault warning model based on the verification set;

[0053] The HI indicator warning threshold evaluation module is used to set a test HI indicator based on the test set, and obtain an evaluation result of the initial HI indicator warning threshold based on the test HI indicator and the initial HI indicator warning threshold;

[0054] The target fault model construction module is used to adjust the initial fault warning model based on the evaluation result to obtain a target fault warning model and a corresponding target HI indicator warning threshold;

[0055] The target fault warning model is used to output the HI index and the target HI index warning threshold of the target warning component based on the input working data of the target warning component.

[0056] In a fourth aspect, a wind turbine fault warning processing system is provided, the processing system comprising an operation data acquisition module, a warning data processing module, a warning judgment module and a fault analysis module;

[0057] The operation data acquisition module is used to acquire the real-time operation data of the wind turbine;

[0058] The early warning data processing module is used to input the real-time operation data into the target fault early warning model to obtain the real-time HI index of the early warning component in the wind turbine and the corresponding HI index threshold;

[0059] Wherein, the target fault warning model is obtained based on the training system of the wind turbine fault warning model described in the third aspect;

[0060] The early warning judgment module is used to obtain the early warning value of the early warning component based on the real-time HI index and the HI index threshold;

[0061] The fault analysis module is used to obtain the actual observation value of the warning component in response to the warning value of the warning component characterizing the fault of the warning component, and obtain the fault analysis result of the warning component based on the warning value and the actual observation value.

[0062] In a fifth aspect, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and used to run on the processor, wherein when the processor executes the computer program, the training method of the wind turbine fault warning model as described in the first aspect and the processing method of the wind turbine fault warning as described in the second aspect are implemented.

[0063] In a sixth aspect, a computer storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the training method of the wind turbine fault warning model as described in the first aspect and the processing method of the wind turbine fault warning as described in the second aspect are implemented.

[0064] In a seventh aspect, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the training method of the wind turbine fault warning model as described in the first aspect, and the processing method of the wind turbine fault warning as described in the second aspect.

[0065] On the basis of being in accordance with the common sense in the art, the above-mentioned preferred conditions can be arbitrarily combined to obtain the preferred embodiments of the present disclosure.

[0066] The positive progress of the present disclosure is that: according to the data asset model, a unique HI index is set for each target warning component, and a wind turbine fault warning model is obtained based on the HI index and the target HI index warning threshold training. The warning model can comprehensively consider the state accumulation of the component throughout the life cycle or working process to better reflect the progressive change process of the component such as fatigue and aging. The overall health status of the wind turbine can be more accurately evaluated. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 A flow chart of a method for training a wind turbine fault warning model provided by an exemplary embodiment of the present disclosure;

[0068] Figure 2 A flowchart of step S107 in a method for training a wind turbine fault warning model provided by an exemplary embodiment of the present disclosure;

[0069] Figure 3 A flow chart of a method for processing a wind turbine fault warning provided by an exemplary embodiment of the present disclosure;

[0070] Figure 4 A flowchart of step S204 in a method for processing a wind turbine fault warning provided by an exemplary embodiment of the present disclosure;

[0071] Figure 5 A schematic diagram of a module of a training system for a wind turbine fault warning model provided by an exemplary embodiment of the present disclosure;

[0072] Figure 6 A module schematic diagram of a wind turbine fault warning processing system provided by an exemplary embodiment of the present disclosure;

[0073] Figure 7 A schematic diagram of the hardware structure of an electronic device provided by an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0074] The present disclosure is further described below by way of examples, but the present disclosure is not limited to the scope of the examples.

[0075] Example 1

[0076] A training method for a wind turbine fault warning model is provided, wherein the wind turbine comprises a target warning component, wherein the target warning component comprises a first associated device, such as Figure 1 As shown, the training method comprises:

[0077] S101, constructing a data asset model of the target early warning component;

[0078] S102, obtaining training data of the wind turbine;

[0079] S103, dividing the training data into a training set, a validation set and a test set;

[0080] S104, inputting the training set into a preset neural network, outputting a reconstruction result corresponding to the training set, training to obtain an initial fault warning model with optimal hyperparameters, and determining fault-related data of the first associated component in the training data;

[0081] S105, obtaining a HI index corresponding to the target warning component by cumulatively summing the fault-related data of the first associated device;

[0082] S106, setting an initial HI indicator warning threshold corresponding to the target warning component in the initial fault warning model based on the verification set;

[0083] S107, setting a test HI indicator based on the test set, and obtaining an evaluation result of the initial HI indicator warning threshold based on the test HI indicator and the initial HI indicator warning threshold;

[0084] S108, adjusting the initial fault warning model based on the evaluation result to obtain a target fault warning model and a corresponding target HI indicator warning threshold;

[0085] The target fault warning model is used to output the HI index and the target HI index warning threshold of the target warning component based on the input working data of the target warning component.

[0086] In step S101, a corresponding data asset model is built according to the target warning components of the wind turbine. The large components in the wind turbine are usually used as the target warning components to decouple the data signals of the large components. The data asset model can include various parameters, performance indicators, structural relationships and other information of the large components for subsequent data analysis and evaluation. Taking the generator as an example, the generator includes fault-related components such as generator bearings.

[0087] In step S102, the training data may be SCADA (Supervisory Control And Data Acquisition) data of the wind turbine generator to improve the accuracy of the model obtained by training.

[0088] In step S103, the training set is used to train the model by inputting a large number of known data samples to allow the model to learn the rules and patterns in the data; the validation set is used to adjust parameters and evaluate the performance of the model during the model training process to find the best model configuration. In this embodiment, it is mainly used to set the initial HI indicator warning threshold; the test set is used to evaluate the generalization ability of the model after the model training is completed, and to adjust the parameters accordingly. In this embodiment, it is mainly used to adjust the initial HI indicator warning threshold.

[0089] In step S104, during the training process, the neural network can be used to adopt some optimization algorithms, such as stochastic gradient descent, to find the optimal parameter combination. In the case of optimal hyperparameters, the reconstruction error between the reconstruction result and the training set is less than the preset error threshold; by adjusting the hyperparameters of the model, such as the number of hidden layer nodes, the sparsity penalty coefficient, etc., the performance of the model is improved to obtain the optimal hyperparameters. At the optimal hyperparameter, the initial fault warning model learns the effective feature representation of the data, and determines the corresponding fault-related data of the first associated device from the data asset model of the target warning component through the learned effective feature representation, which provides effective data for the HI indicator calculation of the target warning component in the subsequent steps.

[0090] In step S105, based on the data asset model constructed in step S101, the fault-related data of each first associated device of the target warning component that needs to be processed is cumulatively summed (Cumsum) to obtain the cumulative value of the first associated device at different time points or states. The HI index of the target warning component is calculated using the cumulative summation result of the first associated device. The HI index is used to evaluate the overall state or performance of the target warning component.

[0091] Specifically, the fault-associated data corresponding to each first associated device in the target warning component in the training data within the preset time period are obtained, and the first cumulative value corresponding to each first associated device is obtained, and then the first cumulative values ​​of all first associated devices are summed to obtain the cumulative sum of the target warning component. Taking the generator as the target warning component as an example, the generator includes first associated devices such as generator bearings, windings, air inlets, and air outlets. During the preset monitoring period, if the contribution of the generator bearing is 3 on the first day, 2 on the second day, 3 on the third day... and so on, and the degradation contribution of the winding is 0 on the first day, 0 on the second day, 0 on the third day..., the cumulative sum of the generator is the sum of the contribution of the generator bearing and the contribution of the winding. The calculated cumulative sum is the degradation index, and the HI index is inversely proportional to the cumulative sum. The corresponding HI index is calculated based on the cumulative sum.

[0092] In step S106, the verification HI index corresponding to the verification set is obtained in the same manner as in step S105, and the verification HI index of the target warning component in the verification set is statistically analyzed to obtain the distribution characteristics thereof. For example, the mean, standard deviation, median and other statistics of the verification HI index are calculated. According to the distribution of the verification HI index, combined with the actual application scenarios and requirements, a reasonable HI index warning threshold is set. For example, considering the cost-benefit analysis, a HI index warning threshold that strikes a balance between the false alarm rate and the missed alarm rate is determined.

[0093] In step S107, the test HI indicator corresponding to the test set is obtained in the same manner as in step S105, and the test HI indicator of the target warning component in the test set is compared with the HI indicator warning threshold set in the verification set. When the test HI indicator exceeds the HI indicator warning threshold, it is regarded as issuing a test warning signal. The actual fault situation corresponding to the test set data and the warning situation obtained by the test are analyzed to obtain an evaluation result of the HI indicator warning threshold.

[0094] In step S108, the evaluation results obtained in the test set are verified, the initial HI indicator warning threshold is adjusted to obtain the target HI indicator warning threshold, and the parameters of the initial warning model are adjusted according to the evaluation results to obtain a target fault warning model with better performance, thereby improving the accuracy, recall and precision of the target fault warning model.

[0095] In this solution, a unique HI indicator is set for each target warning component according to the data asset model, and a wind turbine fault warning model is obtained based on the HI indicator and the target HI indicator warning threshold training. The warning model can fully consider the state accumulation of the component throughout the life cycle or working process to better reflect the progressive changes of the component such as fatigue and aging. It can more accurately evaluate the overall health of the wind turbine instead of just focusing on the state at a certain moment.

[0096] As an achievable manner, the preset neural network is a sparse autoencoder model. In step S104, the step of training to obtain an initial fault warning model with optimal hyperparameters includes:

[0097] The sparse autoencoder model is trained with L1 regularization (L1 norm) to obtain an initial fault warning model with optimal hyperparameters.

[0098] In this solution, the constructed sparse autoencoder model, through learning the training set data, L1 regularization will cause some of the model's weights to approach zero. In the sparse autoencoder, this means that the connection weights of some neurons will be suppressed during the training process. From the perspective of feature representation, it is equivalent to automatic feature selection, which can automatically mine the potential features in the data. The data-driven approach does not rely on manual feature extraction, avoiding the problem that artificial feature engineering may miss important features or introduce subjective bias. For example, in an asset model, the data of a component may have a complex high-dimensional structure, and the autoencoder can convert it into a more meaningful low-dimensional representation, thereby providing more effective input features for subsequent warning models. At the same time, the sparse representation obtained by L1 regularization sparsification helps to improve the interpretability of the initial fault warning model. For a target warning component containing multiple device parameters, the sparse representation can highlight the parameters of several key components that are closely related to the failure of the target warning component, so as to intuitively understand the factors that play a key role in the model decision.

[0099] On the other hand, in the subsequent process of building the early warning model, the high-quality, sparse features extracted by the L1 regularized sparse autoencoder can provide more effective input for the early warning model. These features can more accurately reflect the state changes and potential failure modes of the components, which helps to improve the performance of the early warning model. When calculating the HI index of the target early warning component, the fault-related data of the first associated device extracted by the sparse autoencoder can be more accurately accumulated and summed, so that the HI index can better reflect the actual working status of the target early warning component.

[0100] As an implementable manner, step S102 includes:

[0101] Acquiring historical operation data of the target warning component;

[0102] The training data is obtained by filtering the historical operation data.

[0103] In this scheme, the historical operation data of the target warning component is obtained based on the SCADA data of the wind turbine, and the abnormal data in the historical operation data is cleaned to obtain the corresponding training data. The data can be filtered and cleaned through the recognition of thresholds or abnormal patterns.

[0104] As a feasible approach, Figure 2 As shown, step S107 includes:

[0105] S1071, obtaining a test warning result based on the test HI indicator and the corresponding HI indicator;

[0106] S1072, obtaining actual fault results corresponding to the test set;

[0107] S1073, constructing a confusion matrix based on the actual fault result and the corresponding test warning result;

[0108] S1074. Analyze the confusion matrix to obtain the evaluation result.

[0109] In this solution, by constructing a confusion matrix with the test warning results and the actual fault results, we can obtain true positive examples (True Positive, TP): the actual fault occurred and the model issued a warning, false positive examples (False Positive, FP): the actual fault did not occur but the model issued a warning, true negative examples (True Negative, TN): the actual fault did not occur and the model did not issue a warning, and false negative examples (False Negative, FN): the actual fault occurred but the model did not issue a warning. By analyzing the various elements in the confusion matrix, some evaluation indicators can be calculated to measure the performance of the warning model, such as the evaluation results of accuracy, recall, precision, etc. If the initial HI indicator warning threshold set according to the validation set performs poorly on the test set (such as low accuracy, too many false positive examples, etc.), the initial HI indicator warning threshold will be adjusted to obtain the target HI indicator warning threshold.

[0110] As an implementable manner, step S103 includes:

[0111] The training data is divided into the training set, the validation set and the test set in a ratio of 6:2:2.

[0112] In this scheme, most of the data is assigned to the training set, which provides the fault prediction model with enough data to learn the patterns and regularities in the data. Sufficient training data allows the parameters of the model to be fully adjusted to better fit the data. The existence of the validation set and the test set enables the model to be effectively evaluated during the training process; among them, the distribution of important variables in the training set, validation set and test set is the same. Such as the variables that can affect the HI index, such as the power generation, generator bearing temperature, generator outlet temperature, generator inlet temperature, etc. This division method ensures that the performance evaluation of the model is comprehensive, taking into account both the performance of the model during the training process and the generalization ability of the model.

[0113] As an achievable manner, the target warning component includes at least one of a blade, a gear box, a generator, a converter, a yaw system and a main bearing.

[0114] In this solution, blades, gearboxes, generators, converters, yaw systems, and main bearings are key components of wind turbines. Including these components in the category of target warning components can comprehensively represent the operating status of the entire wind turbine system. For example, blades directly affect the efficiency of wind energy capture, gearboxes are responsible for converting the low-speed rotation of blades into the high-speed rotation required by generators, generators convert mechanical energy into electrical energy, converters process electrical energy, yaw systems ensure that wind turbines are aligned with the wind direction, and main bearings support the entire rotating system. By comprehensively considering these key components, the overall health of the equipment can be more accurately assessed.

[0115] The training method of the wind turbine fault warning model provided in this embodiment sets a unique HI indicator for each target warning component according to the data asset model, and obtains the wind turbine fault warning model based on the HI indicator and the target HI indicator warning threshold training. The warning model can comprehensively consider the status accumulation of the components throughout the life cycle or working process to better reflect the progressive changes such as fatigue and aging of the components, and can more accurately evaluate the overall health status of the motor.

[0116] Example 2

[0117] This embodiment provides a method for processing wind turbine fault warning, such as Figure 3 As shown, the processing method includes:

[0118] S201, obtaining real-time operation data of the wind turbine;

[0119] S202, inputting the real-time operation data into a target fault warning model to obtain a real-time HI index and a corresponding HI index threshold of a warning component in the wind turbine;

[0120] Wherein, the target fault warning model is obtained based on the training method of the wind turbine fault warning model described in Example 1;

[0121] S203, obtaining a warning value of the warning component based on the real-time HI index and the HI index threshold;

[0122] S204: In response to the warning value of the warning component indicating that the warning component is faulty, an actual observation value of the warning component is acquired, and a fault analysis result of the warning component is obtained based on the warning value and the actual observation value.

[0123] In this solution, the target fault warning model obtained by the training method of the wind turbine fault warning model in Example 1 is used to obtain the HI index of the warning component in the operation of the wind turbine and the corresponding HI index threshold. In the case of a fault in the warning component, the warning value calculated according to the HI index and the HI index threshold and the actual observation value are combined to obtain the fault analysis result of the warning component. The accumulation of the state of the component throughout the life cycle or working process can be fully considered to better reflect the progressive changes such as fatigue and aging of the component, and the overall health status of the motor can be more accurately evaluated.

[0124] As a feasible approach, Figure 4 As shown, in step S204, the step of obtaining the fault analysis result of the warning component based on the warning value and the actual observation value includes:

[0125] S2041. Calculate a residual value of the warning component based on the warning value and the actual observation value;

[0126] S2042, determining the residual contribution of the warning component based on the residual value;

[0127] The residual contribution represents the contribution of the second associated device in the warning component to the HI index;

[0128] S2043. Obtain a fault analysis result of the warning component based on the residual contribution.

[0129] In this solution, in the context of model prediction, the residual value refers to the difference between the actual observation value and the model prediction value. The target warning component includes multiple second-related components, and each second-related component has a certain contribution to the total residual of the target warning component. The total residual is decomposed into each second-related component according to certain preset rules by calculating the residual contribution to quantify the relative importance of each second-related component in the process of generating residuals. The preset rule can be to use the ratio of the residual of each second-related component to the sum of the residuals of the warning component as the residual contribution corresponding to the second-related component. By calculating the residual contribution, the relative contribution of each second-related component to the fault can be clarified when a fault warning occurs. It helps to accurately locate the root cause of the fault, so that maintenance personnel can directly focus on the components that contribute more to the fault and improve maintenance efficiency.

[0130] As an achievable manner, the processing method further includes:

[0131] Obtaining a residual offset time at which the residual value is greater than a residual threshold;

[0132] A fault analysis result of the early warning component is obtained based on the residual offset time and the residual contribution.

[0133] In this solution, the residual offset time refers to the time point when the residual begins to deviate significantly from the normal range. It is also possible to continuously monitor the residual value sequence. When the residual value exceeds the residual threshold for multiple times in a row, it is considered that the residual begins to offset. Or use statistical process control methods, such as the Shewhart control chart. When the residual value exceeds the control limit, the control limit is determined based on the mean and standard deviation of the residual, and it is determined as residual offset. At the same time, it is also possible to consider using moving average, exponential smoothing, etc. to smooth the residual sequence in order to more accurately identify the residual offset time. Combining the residual contribution and the residual offset time, the time point of the fault can be determined more accurately. It is helpful to trace the starting process of the fault and understand the time context of the fault development. By observing the residual offset time and the subsequent residual change trend, the development speed and mode of the fault can be analyzed. If the residual value offsets rapidly in a short period of time and continues to increase, it may indicate that the fault is sudden, such as damage caused by foreign matter intrusion into the bearing; if the residual value is slowly and gradually offset, it may mean that the fault is caused by progressive wear or performance degradation of the component. This analysis of the fault development process can provide a basis for predicting the severity of the fault and the remaining service life.

[0134] As an achievable manner, the processing method further includes:

[0135] Acquire corresponding fault operation data based on the fault analysis result;

[0136] A failure notification is generated based on the failure analysis result and the failure operation data.

[0137] In this solution, for the fault analysis results of the target warning components, such as sudden device damage or gradual device wear, the corresponding fault operation data in the wind turbine operation manual is queried, and a notification is generated together with the fault analysis results and the fault operation data, which is sent to the operator responsible for the target warning component, thereby improving the operator's operability and maintenance efficiency of the warning results.

[0138] The processing method for wind turbine fault warning provided in this embodiment uses the target fault warning model obtained by the training method of the wind turbine fault warning model in Example 1, and obtains the HI index and the corresponding HI index threshold of the warning component in the operation of the wind turbine through the target fault warning model. In the case of a fault in the warning component, the warning value calculated according to the HI index and the HI index threshold and the actual observation value are combined to obtain the fault analysis result of the warning component. For the fault analysis result of the target warning component, a notification is generated together with the fault analysis result and the fault operation data, and sent to the operator responsible for the target warning component, so as to improve the operator's operability and maintenance efficiency of the warning result.

[0139] Example 3

[0140] This embodiment provides a training system 100 for a wind turbine fault warning model. Figure 5 As shown, the training system 100 includes a data asset construction module 101, a training data acquisition module 102, a data allocation module 103, an initial model training module 104, an HI index calculation module 105, an HI index warning threshold setting module 106, an HI index warning threshold evaluation module 107 and a target fault model construction module 108;

[0141] The data asset construction module 101 is used to construct a data asset model of the target warning component;

[0142] The training data acquisition module 102 is used to acquire the training data of the wind turbine;

[0143] The data allocation module 103 is used to divide the training data into a training set, a validation set and a test set;

[0144] The initial model training module 104 is used to input the training set into a preset neural network, output the reconstruction result corresponding to the training set, train to obtain the initial fault warning model with the optimal hyperparameters, and determine the fault-related data of the first associated device in the training data;

[0145] The HI index calculation module 105 is used to obtain the HI index corresponding to the target warning component by accumulating and summing the fault-related data of the first associated device;

[0146] The HI indicator warning threshold setting module 106 is used to set the initial HI indicator warning threshold corresponding to the target warning component in the initial fault warning model based on the verification set;

[0147] The HI indicator warning threshold evaluation module 107 is used to set a test HI indicator based on the test set, and obtain an evaluation result of the initial HI indicator warning threshold based on the test HI indicator and the initial HI indicator warning threshold;

[0148] The target fault model construction module 108 is used to adjust the initial fault warning model based on the evaluation result to obtain a target fault warning model and a corresponding target HI indicator warning threshold;

[0149] The target fault warning model is used to output the HI index and the target HI index warning threshold of the target warning component based on the input working data of the target warning component.

[0150] As an implementable manner, the preset neural network is a sparse autoencoder model, and the initial model training module 104 is further used to train the sparse autoencoder model with L1 regularization sparsification to obtain an initial fault warning model with optimal hyperparameters.

[0151] As an implementable manner, the training data acquisition module 102 includes a data acquisition unit and a data filtering unit;

[0152] The data acquisition unit is used to acquire the historical operation data of the target warning component;

[0153] The data filtering unit is used to filter the historical operation data to obtain the training data.

[0154] As an implementable manner, the HI indicator warning threshold evaluation module 107 includes a test warning unit, an actual fault result unit, a confusion matrix unit and an evaluation unit:

[0155] The test warning unit is used to obtain a test warning result based on the test HI indicator and the corresponding HI indicator;

[0156] The actual fault result unit is used to obtain the actual fault result corresponding to the test set;

[0157] The confusion matrix unit constructs a confusion matrix based on the actual fault result and the corresponding test warning result;

[0158] The evaluation unit is used to analyze the confusion matrix to obtain the evaluation result.

[0159] As an implementable manner, the data allocation module 103 is further configured to divide the training data into the training set, the validation set and the test set in a ratio of 6:2:2.

[0160] As an achievable manner, the target warning component includes at least one of a blade, a gear box, a generator, a converter, a yaw system and a main bearing.

[0161] As for the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The system embodiment described above is only illustrative, wherein the units described as separate components may or may not be physically separated, and the components as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the disclosed solution.

[0162] The training system for the wind turbine fault warning model provided in this embodiment sets a unique HI indicator for each target warning component according to the data asset model, and obtains the wind turbine fault warning model based on the HI indicator and the target HI indicator warning threshold training. The warning model can comprehensively consider the status accumulation of the components throughout the life cycle or working process to better reflect the progressive changes such as fatigue and aging of the components, and can more accurately evaluate the overall health status of the motor.

[0163] Example 4

[0164] This embodiment provides a wind turbine fault warning processing system 200, such as Figure 6 As shown, the processing system 200 includes an operation data acquisition module 201, the warning data processing module 202, the warning judgment module 203 and a fault analysis module 204;

[0165] The operation data acquisition module 201 is used to acquire the real-time operation data of the wind turbine;

[0166] The warning data processing module 202 is used to input the real-time operation data into the target fault warning model to obtain the real-time HI index of the warning component in the wind turbine and the corresponding HI index threshold;

[0167] Wherein, the target fault warning model is obtained based on the training system of the wind turbine fault warning model described in the third aspect;

[0168] The warning judgment module 203 is used to obtain the warning value of the warning component based on the real-time HI index and the HI index threshold;

[0169] The fault analysis module 204 is configured to obtain an actual observation value of the warning component in response to the warning value of the warning component indicating the fault of the warning component, and obtain a fault analysis result of the warning component based on the warning value and the actual observation value.

[0170] As an implementable manner, the early warning component includes a second associated device, and the fault analysis module 204 includes a residual calculation unit, a residual contribution determination unit and a result analysis unit:

[0171] The residual calculation unit is used to calculate the residual value of the warning component based on the warning value and the actual observation value;

[0172] The residual contribution determination unit is used to determine the residual contribution of the warning component based on the residual value;

[0173] The residual contribution represents the contribution of the second associated device in the warning component to the HI index;

[0174] The result analysis unit is used to obtain the fault analysis result of the warning component based on the residual contribution.

[0175] As an implementable manner, the fault analysis module 204 further includes a residual offset acquisition unit;

[0176] The residual offset acquisition unit is used to acquire the residual offset time when the residual value is greater than the residual threshold;

[0177] The result analysis unit is further configured to obtain a fault analysis result of the warning component based on the residual deviation time and the residual contribution.

[0178] As an implementable manner, the processing system further includes a fault operation module and a notification module;

[0179] The fault operation module is used to obtain corresponding fault operation data based on the fault analysis result;

[0180] The notification module is used to generate a fault notification based on the fault analysis result and the fault operation data.

[0181] Example 5

[0182] Figure 7 This is a structural schematic diagram of an electronic device shown in an example embodiment of the present disclosure, the electronic device includes a memory, a processor, and a computer program stored in the memory and used to run on the processor, when the processor executes the computer program, it implements the training method of the wind turbine fault warning model or the processing method of the wind turbine fault warning described in any of the above embodiments. Figure 7 The electronic device 90 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0183] like Figure 7 As shown, the electronic device 90 may be in the form of a general-purpose computing device, for example, it may be a server device. The components of the electronic device 90 may include, but are not limited to: at least one processor 91, at least one memory 92, and a bus 93 connecting different system components (including the memory 92 and the processor 91).

[0184] The bus 93 includes a data bus, an address bus, and a control bus.

[0185] The memory 92 may include a volatile memory, such as a random access memory (RAM) 921 and / or a cache memory 922 , and may further include a read-only memory (ROM) 923 .

[0186] The memory 92 may also include a program tool 925 (or utility) having a set (at least one) of program modules 924, such program modules 924 including but not limited to: an operating system, one or more application programs, other program modules and program data, each of which or some combination may include an implementation of a network environment.

[0187] The processor 91 executes various functional applications and data processing by running the computer program stored in the memory 92, such as the training method of the wind turbine fault warning model or the processing method of the wind turbine fault warning provided in any of the above embodiments.

[0188] The electronic device 90 may also communicate with one or more external devices 94 (e.g., keyboards, pointing devices, etc.). Such communication may be performed via an input / output (I / O) interface 95. Furthermore, the electronic device 90 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 96. As shown, the network adapter 96 communicates with other modules of the electronic device 90 via a bus 93. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 90, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems, etc.

[0189] It should be noted that although several units / modules or sub-units / modules of the electronic device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided into multiple units / modules to be embodied.

[0190] Example 6

[0191] The embodiments of the present disclosure also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for training a wind turbine fault warning model or a method for processing a wind turbine fault warning provided in any of the above embodiments.

[0192] The readable storage medium may include but is not limited to: a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device or any suitable combination of the above.

[0193] Example 7

[0194] The embodiment of the present disclosure also provides a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned methods for training a wind turbine fault warning model or a method for processing a wind turbine fault warning.

[0195] Among them, the program code for executing the computer program product of the present disclosure can be written in any combination of one or more programming languages, and the program code can be executed completely on the user device, partially on the user device, as an independent software package, partially on the user device and partially on a remote device, or completely on the remote device.

[0196] Although the specific embodiments of the present disclosure are described above, those skilled in the art should understand that this is only an example, and the protection scope of the present disclosure is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present disclosure, but these changes and modifications all fall within the protection scope of the present disclosure.

Claims

1. A training method for a wind turbine fault warning model, characterized in that: The wind turbine generator includes a target warning component, the target warning component includes a first associated device, and the training method includes: Constructing a data asset model of the target early warning component; Acquiring training data of the wind turbine; Dividing the training data into a training set, a validation set, and a test set; Inputting the training set into a preset neural network, outputting a reconstruction result corresponding to the training set, training to obtain an initial fault warning model with optimal hyperparameters, and determining fault-related data of the first associated component in the training data; Obtaining a HI index corresponding to the target warning component by cumulatively summing the fault-related data of the first associated device; Setting an initial HI indicator warning threshold corresponding to the target warning component in the initial fault warning model based on the verification set; Setting a test HI indicator based on the test set, and obtaining an evaluation result of the initial HI indicator warning threshold based on the test HI indicator and the initial HI indicator warning threshold; Adjust the initial fault warning model based on the evaluation result to obtain a target fault warning model and a corresponding target HI indicator warning threshold; The target fault warning model is used to output the HI index and the target HI index warning threshold of the target warning component based on the input working data of the target warning component.

2. The training method of the wind turbine fault warning model according to claim 1 is characterized in that: The preset neural network is a sparse autoencoder model, and the steps of obtaining an initial fault warning model with optimal hyperparameters through training and determining the fault-related data of the first associated component in the training data include: The sparse autoencoder model is trained with L1 regularization sparsification to obtain an initial fault warning model with optimal hyperparameters.

3. The training method of the wind turbine fault warning model according to claim 1 is characterized in that: The step of obtaining the training data of the wind turbine comprises: Acquiring historical operation data of the target warning component; The training data is obtained by filtering the historical operation data.

4. The training method of the wind turbine fault warning model according to claim 1 is characterized in that: The step of setting a test HI indicator based on the test set, and obtaining an evaluation result of the initial HI indicator warning threshold based on the test HI indicator and the initial HI indicator warning threshold comprises: Obtaining a test warning result based on the test HI indicator and the corresponding HI indicator; Obtaining actual fault results corresponding to the test set; Constructing a confusion matrix based on the actual fault results and the corresponding test warning results; The confusion matrix is ​​analyzed to obtain the evaluation result.

5. The training method of the wind turbine fault warning model according to claim 3 is characterized in that: The step of dividing the training data into a training set, a validation set and a test set comprises: The training data is divided into the training set, the validation set and the test set in a ratio of 6:2:

2.

6. The training method of the wind turbine fault warning model according to claim 1 is characterized in that: The target warning component includes at least one of a blade, a gear box, a generator, a converter, a yaw system and a main bearing.

7. A method for processing wind turbine fault warning, characterized in that: The processing method comprises: Acquiring real-time operating data of the wind turbine; Inputting the real-time operation data into a target fault warning model to obtain a real-time HI index and a corresponding HI index threshold of the warning component in the wind turbine; Wherein, the target fault warning model is obtained based on the training method of the wind turbine fault warning model according to any one of claims 1 to 6; Obtaining a warning value of the warning component based on the real-time HI index and the HI index threshold; In response to the warning value of the warning component indicating that the warning component is faulty, an actual observation value of the warning component is acquired, and a fault analysis result of the warning component is obtained based on the warning value and the actual observation value.

8. The method for processing wind turbine fault warning according to claim 7, characterized in that: The early warning component includes a second associated device, and the step of obtaining a fault analysis result of the early warning component based on the early warning value and the actual observation value includes: Calculating a residual value of the warning component based on the warning value and the actual observation value; Determining a residual contribution of the warning component based on the residual value; The residual contribution represents the contribution of the second associated device in the warning component to the HI index; A fault analysis result of the early warning component is obtained based on the residual contribution.

9. The method for processing wind turbine fault warning according to claim 8, characterized in that: The processing method also includes: Obtaining a residual offset time at which the residual value is greater than a residual threshold; A fault analysis result of the early warning component is obtained based on the residual offset time and the residual contribution.

10. The method for processing wind turbine fault warning according to claim 8, characterized in that: The processing method also includes: Acquire corresponding fault operation data based on the fault analysis result; A failure notification is generated based on the failure analysis result and the failure operation data.

11. A training system for a wind turbine fault warning model, characterized in that: The wind turbine includes a target warning component, the target warning component includes a first associated device, and the training system includes a data asset construction module, a training data acquisition module, a data allocation module, an initial model training module, a HI index calculation module, a HI index warning threshold setting module, a HI index warning threshold evaluation module and a target fault model construction module; The data asset construction module is used to construct a data asset model of the target early warning component; The training data acquisition module is used to acquire the training data of the wind turbine; The data allocation module is used to divide the training data into a training set, a validation set and a test set; The initial model training module is used to input the training set into a preset neural network, output a reconstruction result corresponding to the training set, train to obtain an initial fault warning model with optimal hyperparameters, and determine the fault-related data of the first associated component in the training data; The HI index calculation module is used to obtain the HI index corresponding to the target warning component by accumulating and summing the fault-related data of the first associated device; The HI indicator warning threshold setting module is used to set the initial HI indicator warning threshold corresponding to the target warning component in the initial fault warning model based on the verification set; The HI indicator warning threshold evaluation module is used to set a test HI indicator based on the test set, and obtain an evaluation result of the initial HI indicator warning threshold based on the test HI indicator and the initial HI indicator warning threshold; The target fault model construction module is used to adjust the initial fault warning model based on the evaluation result to obtain a target fault warning model and a corresponding target HI indicator warning threshold; The target fault warning model is used to output the HI index and the target HI index warning threshold of the target warning component based on the input working data of the target warning component.

12. A wind turbine fault warning processing system, characterized in that: The processing system includes an operation data acquisition module, an early warning data processing module, an early warning judgment module and a fault analysis module; The operation data acquisition module is used to acquire the real-time operation data of the wind turbine; The early warning data processing module is used to input the real-time operation data into the target fault early warning model to obtain the real-time HI index of the early warning component in the wind turbine and the corresponding HI index threshold; Wherein, the target fault warning model is obtained based on the training system of the wind turbine fault warning model according to claim 11; The early warning judgment module is used to obtain the early warning value of the early warning component based on the real-time HI index and the HI index threshold; The fault analysis module is used to obtain the actual observation value of the warning component in response to the warning value of the warning component characterizing the fault of the warning component, and obtain the fault analysis result of the warning component based on the warning value and the actual observation value.

13. An electronic device comprising a memory, a processor, and a computer program stored in the memory and used to run on the processor, characterized in that: When the processor executes the computer program, it implements the training method of the wind turbine fault warning model as described in any one of claims 1 to 6, and the processing method of the wind turbine fault warning as described in any one of claims 7 to 10.

14. A computer storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the training method of the wind turbine fault warning model as described in any one of claims 1 to 6 and the processing method of the wind turbine fault warning as described in any one of claims 7 to 10 are implemented.

15. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, it implements the training method of the wind turbine fault warning model as described in any one of claims 1 to 6, and the processing method of the wind turbine fault warning as described in any one of claims 7 to 10.

Images