Wind turbine generator fault detection method and system based on data fusion and medium

Abstract

The invention discloses a wind turbine generator fault detection method and system based on data fusion and a medium, relates to the technical field of fault detection, and solves the problems that interaction between influence factors is not considered in the prior art, a large amount of data needs to be processed in real time in the fault detection process, and the fault detection efficiency is high. Therefore, the technical problems of inaccurate fault detection and low detection efficiency of the wind turbine generator are solved. According to the method, the artificial intelligence model is constructed through the cascade structure and is trained, and the training data comprises the influence of each influence factor on the unit fault and also comprises the influence of the influence factor on other influence factors, so that the artificial intelligence model can learn the interaction between the influence factors, and the training efficiency is improved. The prediction precision of the unit fault and the fault probability is improved; in the fault prediction process, only part of prediction data needs to be obtained, other data can be rapidly extracted through the fault correlation model, the data processing amount can be effectively reduced, and the fault detection efficiency is improved.

Description

Technical Field

[0001] This application belongs to the field of fault detection, and relates to the fault detection technology of wind turbines. Specifically, it is a method, system and medium for fault detection of wind turbines based on data fusion. Background Art

[0002] Due to the geographical condition limitations of wind farm resources, wind turbines are generally installed in extremely harsh environments and are constantly under severe impact loads, high and low temperature alternations, and high and low speed shears. Their related components are extremely vulnerable to damage. How to detect faults in wind turbines under harsh alternating working conditions is an urgent problem to be solved.

[0003] With the development of artificial intelligence technology, using machine learning or deep learning algorithms to achieve intelligent fault diagnosis can not only fuse and process a variety of associated data, but also has high accuracy and strong generalization ability, with broad application prospects. When analyzing the faults of wind turbines in existing solutions, an artificial intelligence model is generally trained through historical data, and various types of predicted data are input into the trained artificial intelligence model to achieve fault detection. However, some data can be predicted with high accuracy, some data are difficult to achieve high-precision prediction, and there are also mutual influences between various data, making it difficult to ensure the comprehensive accuracy of the predicted data, and thus unable to accurately predict wind turbine faults.

[0004] This application provides a method, system and medium for fault detection of wind turbines based on data fusion to solve the above technical problems. Summary of the Invention

[0005] This application aims to solve at least one of the technical problems existing in the prior art; for this purpose, this application proposes a method, system and medium for fault detection of wind turbines based on data fusion, which is used to solve the technical problems that the prior art does not consider the interaction between influencing factors and requires real-time processing of a large amount of data during the fault detection process, resulting in inaccurate wind turbine fault detection and low detection efficiency.

[0006] To achieve the above object, the first aspect of this application provides a method for fault detection of wind turbines based on data fusion, including:

[0007] Identifying unit faults and mining the influencing factors corresponding to the unit faults through the historical monitoring data of several wind turbines of the same model, and constructing a fault association model of the unit faults and the influencing factors in combination with a probabilistic graphical model; wherein, the probabilistic graphical model includes a Bayesian network;

[0008] Extract several data groups from the fault correlation model, and extract the first data sequence and the second data sequence from the several data groups; train an artificial intelligence model based on the first data sequence and the second data sequence to obtain a fault detection model; wherein, the first data sequence is a data sequence, and the second data sequence is a weight sequence;

[0009] Predict and obtain the prediction data of the target unit, input the prediction data into the fault detection model to obtain a fault type sequence; formulate emergency measures according to the fault type sequence; wherein, the fault type sequence includes unit faults and their occurrence probabilities, and the emergency measures are used to reduce the damage caused by unit faults.

[0010] Preferably, constructing a fault correlation model includes:

[0011] Call a network construction tool to construct a Bayesian network;

[0012] Extract a set of variables from the preprocessed historical monitoring data, and determine the nodes in the Bayesian network and the variable relationships between the nodes through the set of variables;

[0013] Use the historical monitoring data as training data to train the Bayesian network, and use the Bayesian network that passes the verification as the fault correlation model.

[0014] Preferably, extracting the first data sequence and the second data sequence from several data groups includes:

[0015] Extract numerical features from several data groups, and generate the first data sequence according to the numerical features; wherein, the numerical features are the numerical values of unit faults and their influencing factors;

[0016] Extract probability features from several data groups, and generate the second data sequence according to the probability features; wherein, the probability features represent the influence degrees of the influencing factors on unit faults and between the influencing factors.

[0017] Preferably, obtaining a fault detection model includes:

[0018] Perform standardization processing on the first data sequence and the second data sequence respectively to obtain the first feature sequence and the second feature sequence; wherein, the standardization processing is to ensure the training effect of the artificial intelligence model, including normalization processing or data augmentation;

[0019] Train the constructed artificial intelligence model through the first feature sequence and the second feature sequence to obtain a fault detection model.

[0020] Preferably, the artificial intelligence model is a cascade structure, including: a feature sharing layer and two recognition sub-networks, and the output data of one recognition sub-network is used to construct the input of the other recognition sub-network.

[0021] Preferably, training the artificial intelligence model of the cascade mechanism includes:

[0022] Constructing an artificial intelligence model with a cascade structure; extracting Data One and Data Two from the first data sequence; wherein, the value of Data One affects Data Two.

[0023] Training one recognition sub-network in the artificial intelligence model to recognize Data Two based on Data One, and training another recognition sub-network to recognize the faults of the unit based on Data One and Data Two.

[0024] Preferably, the division rules of Data One and Data Two include:

[0025] Dividing the environmental data in the working area of the wind turbine into Data One, and dividing the operation status data of the wind turbine into Data Two; wherein, the change of the environmental data will cause the change of the operation status data.

[0026] Preferably, inputting the prediction data into the fault detection model to obtain the fault type sequence includes:

[0027] Extracting the prediction data corresponding to the target unit, and combining the prediction data with the second data sequence corresponding to the fault correlation model to generate fault prediction data.

[0028] Inputting the fault prediction data into the fault detection model to obtain the fault type sequence.

[0029] The second aspect of the present application provides a wind turbine fault detection system based on data fusion, including a central decision-making module, and a data acquisition module and an emergency processing module connected thereto;

[0030] Data acquisition module: used to collect the historical monitoring data of the wind turbine and the prediction data of the target unit; wherein, the target unit is the wind turbine that needs to be subjected to fault detection.

[0031] Central decision-making module: used to train the artificial intelligence model according to the historical monitoring data to obtain the fault detection module; and, identifying and obtaining the fault type sequence of the target unit through the fault detection module.

[0032] Emergency processing module: used to formulate emergency measures according to the fault type sequence, and timely remind the staff to carry out preventive treatment according to the emergency measures.

[0033] The third aspect of the present application provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program, and when the computer program is executed, the above-mentioned wind turbine fault detection method based on data fusion is implemented.

[0034] Compared with the prior art, the beneficial effects of the present application are:

[0035] This application constructs and trains an artificial intelligence model through a cascaded structure. The training data includes both the impacts of various influencing factors on the unit faults and the impacts of influencing factors on other influencing factors, enabling the artificial intelligence model to learn the interactions between influencing factors and improve the prediction accuracy of unit faults and fault probabilities. Moreover, in the process of fault prediction, this application only needs to obtain partial prediction data, and other data can be quickly extracted through the fault correlation model, which can effectively reduce the data processing volume and improve the fault detection efficiency.

[0036] This application constructs a fault correlation model by learning the interactions between various influencing factors and the comprehensive impacts of various influencing factors on unit faults from wind turbines of the same model. In this fault correlation model, there are not only the impacts and degrees of influence of a certain influencing factor on other influencing factors, but also the comprehensive impacts and degrees of influence of various influencing factors on a certain unit fault. Through the fault correlation model, the interactions between various influencing factors can be mined, providing a data basis for training the fault detection model and fault detection, and improving the accuracy of fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0038] Figure 1 It is a schematic flowchart of the wind turbine fault detection method in Embodiment 1 of the present application;

[0039] Figure 2 It is a schematic structural diagram of the artificial intelligence model in Embodiment 1 of the present application;

[0040] Figure 3 It is a schematic system diagram of the wind turbine fault detection system of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] The following will clearly and completely describe the technical solutions of the present application in combination with the embodiments. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present application.

[0042] Due to geographical restrictions, wind farms and wind turbines are generally installed in remote and harsh environments. Wind turbines operating in harsh environments will be affected by various factors, such as load impact, temperature changes, wind speed changes, etc. These factors will not only directly lead to wind turbine failures, but also affect the operating status of various components and indirectly lead to wind turbine failures.

[0043] Existing solutions have begun to use artificial intelligence algorithms to solve complex data processing problems, such as predicting the various factors that affect the state of wind turbines, and inputting these factors into the trained artificial intelligence model to obtain fault identification results. However, as mentioned above, there are many factors that affect the state of wind turbines, some of which have direct effects, and some of which indirectly affect the state of wind turbines by affecting other factors on the basis of direct effects. The above problems lead to the fact that predicting the possible faults of wind turbines by predicting various influencing factors is not accurate enough, and it is impossible to comprehensively consider the influence of various factors. Even if the relationship between the various influencing factors is analyzed through algorithms, it will lead to a large amount of data processing, which will affect the efficiency of fault identification.

[0044] Embodiment 1: This embodiment constructs a fault detection model by analyzing the relationship between various influencing factors in advance, and then combines a small amount of prediction data to complete high-precision fault detection of wind turbines.

[0045] See also Figure 1 - Figure 2 The first embodiment of the present application provides a wind turbine fault detection method based on data fusion, comprising:

[0046] Through the historical monitoring data of several wind turbines of the same model, the unit failure is identified and the influencing factors corresponding to the unit failure are mined, and the failure association model of the unit failure and the influencing factors is constructed by combining the probabilistic graphical model; wherein the probabilistic graphical model includes the Bayesian network;

[0047] Extracting a number of data groups from the fault association model, extracting a first data sequence and a second data sequence from the number of data groups; training an artificial intelligence model based on the first data sequence and the second data sequence to obtain a fault detection model; wherein the first data sequence is a data sequence, and the second data sequence is a weight sequence;

[0048] Prediction: Obtain prediction data of the target unit, input the prediction data into the fault detection model to obtain a fault type sequence; formulate emergency measures based on the fault type sequence; wherein the fault type sequence includes unit failures and the probability of occurrence, and the emergency measures are used to reduce the damage caused by the unit failure.

[0049] In order to avoid a large amount of data processing during subsequent fault detection, in this embodiment, by analyzing and mining the historical monitoring data of the wind turbine generator set, the possible unit faults of the wind turbine generator set and their influencing factors are extracted, and then the influence relationship of the influencing factors on the unit faults is established to obtain a fault correlation model.

[0050] The specific construction process of the fault correlation model includes:

[0051] Call a network construction tool to construct a Bayesian network, which is the original network structure; then extract the unit faults and corresponding influencing factors from the historical monitoring data to generate a variable set; use each variable in the variable set as a network node in the Bayesian network, and finally train and obtain a conditional probability table according to the historical monitoring data to get a complete Bayesian network, that is, the fault correlation model.

[0052] There are many kinds of tools for constructing Bayesian networks, such as software like Netica, AgenaRisk, Analytica, BayesiaLab, etc. The construction of Bayesian networks is a relatively mature technology and will not be elaborated here. For wind turbine generator sets, the network nodes in the Bayesian network are unit faults and influencing factors; and considering the mutual influence among the influencing factors, it can be known that not only do unit faults have a parent node combination, but some influencing factors also have a parent node combination. If influencing factor A is affected by several other influencing factors, then the set of several other influencing factors can be used as the parent node set of this influencing factor A.

[0053] It should be noted that in order to improve the generalization ability of the fault correlation model, in this embodiment, the historical monitoring data of several wind turbine generator sets of the same model are used to construct the fault correlation model. The working condition ranges corresponding to the historical monitoring data of several wind turbine generator sets of the same model are wider, and the influencing factors are more complex, which can expand the application range of the fault correlation model and thus improve its generalization ability.

[0054] After the basic structure of the Bayesian network is constructed, the network nodes are defined through the variable set, and then the network parameters are calculated through the historical monitoring data of several wind turbine generator sets to determine the dependency relationship between each network node. This dependency relationship can be understood as a conditional probability table. When calculating the conditional probability of a network node, first determine the parent node set of the target network node, count the number of times the target network node takes different values under different value combinations, and calculate the conditional probability value.

[0055] In this embodiment, the conditional probability between each network node in the fault correlation model is used as the weight to evaluate the influence of the numerical change of a network node on other network nodes. In some other preferred embodiments, the weight can also be calculated according to the conditional probability table.

[0056] It should be noted that since the relationships between influencing factors and between each influencing factor and the unit failure are relatively complex, different values of influencing factors correspond to different weights in the fault correlation model, and different combinations of each influencing factor and weight correspond to different unit failures. However, due to the limited historical monitoring data, it is difficult to cover all working conditions. Therefore, an artificial intelligence model is introduced to learn the influence relationship of influencing factors on unit failures to improve the generalization ability of the artificial intelligence model.

[0057] Compared with traditional artificial intelligence models that only consider the influence of influencing factors on wind turbine unit failures, this embodiment introduces the mutual influence between influencing factors, that is, first considers the influence of each influencing factor on other influencing factors, and then comprehensively considers the influence of each influencing factor on wind turbine unit failures. For this purpose, this embodiment divides several data groups. Specifically, a first data sequence and a second data sequence are extracted from several data groups, including: extracting numerical features from several data groups and generating a first data sequence according to the numerical features; where the numerical features are the numerical values of unit failures and their influencing factors.

[0058] Extract probability features from several data groups and generate a second data sequence according to the probability features; where the probability features represent the influence degree of influencing factors on unit failures and the influence between influencing factors.

[0059] Each of the above data groups can be understood as the combination of unit failures and their corresponding influencing factors (unit failures and influencing factors are in one-to-one correspondence with the network nodes in the Bayesian network), that is, a data group includes unit failures and corresponding influencing factors, as well as the influence degree of each influencing factor on the unit failure and the influence degree of the influencing factor on other influencing factors.

[0060] Extract the numerical values of unit failures (which can be numerically processed, such as setting digital labels) and their influencing factors in the data group as the first data sequence, and extract the mutual influence between influencing factors and the influence of influencing factors on unit failures as the second data sequence.

[0061] Compared with the prior art, the prior art generally uses the first data sequence as the input of the prediction model, that is, only considers the influence degree of influencing factors on unit failures, while ignoring the influence degree of influencing factors on other influencing factors. This embodiment introduces the mutual influence between influencing factors, that is, the second data sequence, to improve the accuracy of unit failure prediction. The advantage of this embodiment is that the reliability of unit failure detection can be improved by introducing the mutual influence between influencing factors.

[0062] The fault detection model of this example is trained through the first data sequence and the second data sequence. The specific training process can be referred to:

[0063] After standardizing the first data sequence corresponding to the data group, the first feature sequence is obtained. Similarly, after standardizing the second data sequence, the second feature sequence is obtained. Taking the first feature sequence and the second feature sequence as the feature data sequences corresponding to the data group, there are several data groups corresponding to several feature data sequences. Training the artificial intelligence model through several feature data sequences can obtain a fault detection model. It should be noted that the input of the fault detection model is the influencing factors corresponding to the unit fault, the weights, and the weights between the influencing factors, and the output is the unit fault and the probability corresponding to each unit fault.

[0064] The above data standardization includes data division and extraction, data matching, etc. This operation is to clarify the input and output of the artificial intelligence model and perform necessary processing on the data to suit the training of the artificial intelligence model. Data division and extraction, such as extracting the input data and output data of the model from the first data sequence and the second data sequence; data matching, such as matching the influencing factors with their influencing weights on the unit fault, which is essentially to match the data in the first data sequence and the second data sequence.

[0065] This embodiment takes the training of the artificial intelligence model with a cascaded structure as an example, specifically as follows:

[0066] Construct an artificial intelligence model: The artificial intelligence model includes an input layer, a feature sharing layer, two recognition sub-networks, and an output layer; the feature sharing layer recognizes the features of the input data corresponding to the input layer for use by the recognition sub-networks; the inputs and outputs of the two recognition sub-networks are different, and the output of one of the recognition sub-networks is a part of the input of the other recognition sub-network. The recognition sub-network can be a model that meets the training requirements, such as a BP neural network, an LBF neural network, etc.

[0067] Obtain a training data set: Extract the first data sequence and the second data sequence corresponding to several data groups, and divide them into data type one and data type two according to the characteristics of the influencing factors themselves; data type one refers to the influencing factors that are not affected by other influencing factors, such as temperature, humidity, wind force, etc., and data type two refers to the influencing factors that are affected by other influencing factors, such as the working temperature and service life of the wind turbine components;

[0068] Extract the data corresponding to data type one from the first data sequence as data one; extract the data corresponding to data type two from the first data sequence as data two. Simply put, data one can not only directly affect the operating state of the wind turbine, but also indirectly affect the operating state of the wind turbine by affecting data two.

[0069] Data 1 can be used as an independent variable, and the corresponding dependent variable is Data 2. There is no linear relationship between the two because the interaction between them is relatively complex. Therefore, a mapping relationship between the two is established by constructing an identification sub-network. The input data for training the identification sub-network should include Data 1, Data 2 that is not affected by Data 1, and the weights between Data 1 and Data 2 extracted from the second data sequence; the output data is Data 2 affected by Data 1. In this way, the training of the identification sub-network can be completed through several groups of corresponding first data sequences and second data sequences. It should be noted that Data 2 that is not affected by Data 1 here can be obtained through simulation tests. For example, the life of a component at a certain future time can be obtained by simulating its operation in a standard environment.

[0070] Next, use Data 1, the Data 2 identified by the above identification sub-network, and the weights between Data 1, Data 2 and the unit failure as input data, and use the unit failure and the failure probability as output data to train another identification sub-network. After the training is completed, a fault detection model can be obtained. It should be noted that Data 2 is the output of one of the identification sub-networks, and the weights between Data 1, Data 2 and the unit failure are extracted from the second data sequence. The data can be normalized and processed according to the model input requirements.

[0071] The specific training process can be referred to as follows:

[0072] Multiple data groups can be extracted from the fault correlation diagram. Each data group corresponds to several influencing factors and several unit failures and failure probabilities caused by the combination of these influencing factors. Therefore, the data group should at least include influencing factors, unit failures, and the weight relationship between the two, influencing factors, other influencing factors, and the weight relationship between the two.

[0073] The influencing factors, unit failures, and the weight relationship between the two are simply represented as {[(A1, α1), (A2, α2), …, (Ai, αi)], [C, γ]}; where i is the number of influencing factors, Ai is the actual value of influencing factor i, αi is the weight of influencing factor i on unit failure C, and γ is the occurrence probability of unit failure C under the combination of influencing factors Ai.

[0074] The influencing factors, other influencing factors, and the weight relationship between the two are simply represented as {[(A1, β1), (A2, β2), …, (Ai-1, βi-1)], [Ai]}; where βi-1 is the influencing weight of influencing factor i-1 on influencing factor Ai. Of course, the influence of each influencing factor on other influencing factors is based on the base value of other influencing factors. The base value refers to the predicted value or estimated value when not affected by other factors.

[0075] During the training process, first, a recognition sub-network in the artificial intelligence model is trained according to {[(A1, β1), (A2, β2), …, (Ai-1, βi-1)], Ax, [Ai]}, where Ax is the base value of influencing factor i. Then, another recognition sub-network is trained through {[(A1, α1), (A2, α2), …, (Ai, αi)], [C, γ]}. It should be noted that {[(A1, α1), (A2, α2), …, (Ai, αi)], [C, γ]} represents the probabilities of multiple influencing factors leading to a unit failure under their respective weights. Of course, it can also be adjusted to the probabilities of multiple influencing factors leading to multiple unit failures under multiple weights, but the amount of data is relatively large.

[0076] When it is necessary to detect faults in the target unit, data one (not affected by other influencing factors) corresponding data is obtained through a third-party platform, such as ambient temperature, ambient humidity, wind force, wind direction, etc.; then data two is obtained, that is, the data value at the corresponding moment is predicted according to the data change curve under the ideal state corresponding to the wind turbine unit, such as operating temperature, operating current, and operating voltage.

[0077] Finally, the weights corresponding to data one and data two are extracted from the fault correlation model, and after integrating each data, it is input into the fault detection model to obtain the corresponding fault type sequence. According to this fault type sequence, the staff can timely formulate emergency measures to prevent the possible faults of the wind turbine unit in time and minimize the losses.

[0078] In this embodiment, by learning the interaction between various influencing factors and the comprehensive influence of each influencing factor on the unit failure from the same type of wind turbine units, a fault correlation model is constructed. In this fault correlation model, there is not only the influence of a certain influencing factor on other influencing factors and the degree of influence, but also the comprehensive influence of each influencing factor on a certain unit failure and the degree of influence. And by constructing and training an artificial intelligence model through a cascade structure, the artificial intelligence model can learn the interaction between influencing factors and improve the prediction accuracy of unit failures and failure probabilities.

[0079] When predicting and obtaining the corresponding fault type sequence by combining prediction data, only the above-mentioned data one needs to be obtained through a third-party platform, the above-mentioned data two needs to be predicted based on the working state of the wind turbine under the built-in standard environment, and then the weight relationship between data one and data two, as well as the weight relationship between data one and the corrected data two with respect to each unit fault, are extracted from the fault correlation model. After integrating these data and inputting them into the fault detection model, the probability of each unit fault occurring can be obtained. Compared with the full fault prediction of this embodiment (after sufficient data preparation, predicting the fault probability of possible unit faults in the wind turbine), in some other preferred embodiments, it is also possible to predict a single unit fault. When training the recognition sub-network, only the weight between the corresponding unit fault and each influencing factor needs to be extracted from the fault correlation model. This method is more efficient for single unit faults.

[0080] Please refer to Figure 3 , an embodiment of the second aspect of the present application provides a wind turbine fault detection system based on data fusion, including: a central decision-making module, and a data acquisition module and an emergency processing module connected thereto;

[0081] Data acquisition module: used to collect historical monitoring data of the wind turbine and prediction data of the target unit; wherein, the target unit is the wind turbine that needs to be fault-detected;

[0082] Central decision-making module: used to train an artificial intelligence model based on historical monitoring data to obtain a fault detection module; and, identify and obtain the fault type sequence of the target unit through the fault detection module;

[0083] Emergency processing module: used to formulate emergency measures according to the fault type sequence, and timely remind the staff to carry out preventive treatment according to the emergency measures.

[0084] Generally, the central decision-making module of this embodiment is not set in the wind turbine, but remotely detects the fault of the wind turbine, that is, collects historical monitoring data through various channels, stores the change curve of the working state of the wind turbine under the ideal environment, and is responsible for the training and application of the fault correlation model and the fault detection model. The data acquisition module is mainly responsible for data acquisition, collects various data that may cause faults through sensors installed on the wind turbine, and obtains predicted weather data from a third-party data platform. The emergency processing module is mainly used to remind the staff so that the staff can take preventive measures in time.

[0085] An embodiment of the third aspect of the present application provides a computer-readable storage medium, including: the computer-readable storage medium stores a computer program, and when the computer program is executed, it implements the above-mentioned method for detecting faults in a wind turbine based on data fusion.

[0086] The above embodiments are only used to illustrate the technical solutions of the present application and not to limit them. Although the present application has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present application can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A fault detection method for wind turbines based on data fusion, characterized in that, Including: Identifying the faults of the wind turbines through the historical monitoring data of several wind turbines of the same type, mining the influencing factors corresponding to the faults of the wind turbines, and constructing a fault correlation model of the wind turbine faults and the influencing factors in combination with the probabilistic graphical model; wherein, the probabilistic graphical model includes a Bayesian network; Extracting several data groups from the fault correlation model, and extracting a first data sequence and a second data sequence from the several data groups; training an artificial intelligence model based on the first data sequence and the second data sequence to obtain a fault detection model; wherein, the first data sequence is a data sequence, and the second data sequence is a weight sequence; Predicting and obtaining the prediction data of the target wind turbine, inputting the prediction data into the fault detection model to obtain a fault type sequence; formulating emergency measures according to the fault type sequence; wherein, the fault type sequence includes the wind turbine faults and the occurrence probabilities, and the emergency measures are used to reduce the damage caused by the wind turbine faults.

2. The method for detecting faults of a wind turbine based on data fusion according to claim 1, characterized in that Constructing the fault correlation model includes: Invoking a network construction tool to construct a Bayesian network; Extracting a variable set from the preprocessed historical monitoring data, and determining the nodes and the variable relationships between the nodes in the Bayesian network through the variable set; Training the Bayesian network with the historical monitoring data as training data, and using the Bayesian network that passes the verification as the fault correlation model.

3. A fault detection method for a wind turbine based on data fusion according to claim 1, characterized in that Extracting the first data sequence and the second data sequence from several data groups includes: Extracting numerical features from several of the data groups, and generating a first data sequence according to the numerical features; wherein, the numerical features are the numerical values of the wind turbine faults and their influencing factors; Extracting probability features from several of the data groups, and generating a second data sequence according to the probability features; wherein, the probability features represent the influence degrees of the influencing factors on the wind turbine faults and between the influencing factors.

4. A fault detection method for a wind turbine based on data fusion according to claim 1, characterized in that, Obtaining the fault detection model includes: Performing standardization processing on the first data sequence and the second data sequence respectively to obtain a first feature sequence and a second feature sequence; wherein, the standardization processing is to ensure the training effect of the artificial intelligence model, including normalization processing or data augmentation; Training the constructed artificial intelligence model with the first feature sequence and the second feature sequence to obtain the fault detection model.

5. A fault detection method for a wind turbine based on data fusion according to claim 1, characterized in that, The artificial intelligence model is a cascade structure, including: a feature sharing layer and two recognition sub-networks, and the output data of one recognition sub-network is used to construct the input of the other recognition sub-network.

6. The method for detecting faults of a wind turbine based on data fusion according to claim 5, wherein, Training the artificial intelligence model with a cascade structure includes: Constructing an artificial intelligence model with a cascade structure; extracting data one and data two from the first data sequence; wherein, the value of data one affects data two; Training one recognition sub-network in the artificial intelligence model to recognize data two according to data one, and training the other recognition sub-network to recognize the wind turbine faults according to data one and data two.

7. A fault detection method for a wind turbine based on data fusion according to claim 6, characterized in that The division rules of the data one and the data two include: Dividing the environmental data of the working area of the wind turbine into data one, and dividing the operation state data of the wind turbine into data two; wherein, the change of the environmental data will cause the change of the operation state data.

8. A fault detection method for a wind turbine based on data fusion according to claim 7, characterized in that, Input the prediction data into a fault detection model to obtain a fault type sequence, including: Extract the prediction data corresponding to the target unit, and combine the prediction data with the second data sequence corresponding to the fault correlation model to generate fault prediction data; Input the fault prediction data into the fault detection model to obtain a fault type sequence.

9. A wind turbine fault detection system based on data fusion, which is used to execute a wind turbine fault detection method based on data fusion according to any one of claims 1 to 8, characterized in that, It includes a central decision-making module, and a data acquisition module and an emergency handling module connected thereto; Data acquisition module: used to collect historical monitoring data of the wind turbine and prediction data of the target unit; wherein, the target unit is the wind turbine that needs to be fault-detected; Central decision-making module: used to train an artificial intelligence model based on historical monitoring data to obtain a fault detection module; and, identify and obtain the fault type sequence of the target unit through the fault detection module; Emergency handling module: used to formulate emergency measures according to the fault type sequence, and timely remind the staff to carry out preventive treatment according to the emergency measures.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and when the computer program is executed, it implements a wind turbine fault detection method based on data fusion according to any one of claims 1 to 8.

Images