A wind turbine fault detection method, device and medium

Abstract

The embodiment of the application discloses a wind driven generator fault detection method. Including: obtaining multiple operation data of the wind driven generator; obtaining first probability values of different fault types corresponding to the multiple operation data, and constructing a first probability matrix according to the first probability values; adjusting the first probability values to obtain a second probability matrix; obtaining multiple different environment data in the operation process of the wind driven generator; based on the second probability matrix and the multiple different environment data, obtaining a first fault type set corresponding to the wind driven generator and a to-be-detected fault type set; adjusting second probability values in the second probability matrix to obtain a third probability matrix; based on a preset synthesis rule and the third probability matrix, obtaining third probability values corresponding to different fault types in the to-be-detected fault type set; based on the first fault type set and the third probability values, obtaining an output detection fault. Through the above method, the operation fault of the wind driven generator can be found in time.

Description

Technical Field

[0001] This application relates to the field of wind power generation technology, and in particular to a method, equipment and medium for detecting faults in wind turbine generators. Background Technology

[0002] Wind turbines are mostly installed in remote areas rich in wind energy resources, such as high mountains, wilderness, islands, and even offshore. Traditional maintenance strategies for wind turbine malfunctions heavily rely on regular routine inspections and reactive repairs. Therefore, in the event of a serious wind turbine failure, it is difficult to detect the fault in a timely manner, thus affecting the normal operation of the wind turbine. Summary of the Invention

[0003] This application provides a method, device, and medium for detecting faults in wind turbines, which addresses the following technical problem: in the event of a serious fault in a wind turbine, it is difficult to detect the fault in a timely manner, thus affecting the normal operation of the wind turbine.

[0004] The embodiments of this application adopt the following technical solutions:

[0005] This application provides a method for detecting faults in wind turbine generators. The method includes: acquiring various operational data of the wind turbine using a first set of sensors installed on the wind turbine; wherein the operational data includes at least one or more of the following: temperature, vibration frequency, and rotational speed of the wind turbine; inputting the acquired operational data into a deep belief network to obtain first probability values ​​for different fault types corresponding to the various operational data, and constructing a first probability matrix based on the first probability values; adjusting the first probability values ​​according to the weight of each sensor's operational data to obtain a second probability matrix; acquiring various environmental data during the wind turbine's operation using a second set of sensors installed on the wind turbine; wherein the environmental data includes at least one or more of the following: temperature, atmospheric pressure, and wind speed of the environment in which the wind turbine is located; obtaining a first fault type set and a set of fault types to be tested for the wind turbine based on the second probability matrix and the various environmental data; adjusting the second probability values ​​in the second probability matrix based on the various environmental data to obtain a third probability matrix; obtaining third probability values ​​corresponding to different fault types in the set of fault types to be tested based on a preset synthesis rule and the third probability matrix; and obtaining the output detection fault of the wind turbine based on the first fault type set and the third probability values.

[0006] This embodiment of the application, by installing a first set of sensors on the wind turbine, enables real-time monitoring of the wind turbine's operating data, allowing for the detection of the wind turbine's operating status based on the acquired data. Secondly, this embodiment establishes a first probability matrix based on the received operating data and adjusts the first probability matrix according to the weight value corresponding to each sensor, thereby improving the accuracy of its output of the wind turbine's fault type. Furthermore, since the wind turbine's operating status is also affected by various environmental factors, this embodiment installs a second set of sensors on the wind turbine. This second set of sensors acquires current environmental data in real time, and the environmental data is analyzed to adjust the second probability matrix, resulting in a third probability matrix. This further adjusts the probability matrix, making its output of fault types more accurate.

[0007] In one implementation of this application, before inputting the acquired multiple operational data into the deep belief network, the method further includes: acquiring operational data samples corresponding to a first sensor set; constructing a reference deep belief network to extract feature vectors corresponding to the operational data samples; superimposing a regression prediction layer on the reference deep belief network, predicting the fault state of the wind turbine based on the regression prediction layer and the feature vectors, and outputting a fault prediction value; comparing the fault prediction value with a preset fault reference value to obtain a fault prediction error value, and adjusting the parameters in the reference deep belief network based on the fault prediction error value to obtain the deep belief network.

[0008] This application's embodiments automatically extract features from engine component signal data based on neural networks, avoiding the need for manual analysis of data features and the significant effort required for feature extraction. Secondly, this application's embodiments construct a deep belief network (RDBM), combining multiple hidden layers, to effectively simulate the nonlinear relationships between engine component signal data. Furthermore, by adding a logistic regression layer at the top of the deep belief network, it can perform supervised health status prediction using labeled data, effectively utilizing the linear relationships between engine component signal data.

[0009] In one implementation of this application, multiple types of operational data are input into a deep belief network to obtain first probability values ​​for different fault types corresponding to the multiple types of operational data, and a first probability matrix is ​​constructed based on the first probability values. Specifically, this includes: inputting the operational data from the first sensor set into the deep belief network in sequence; outputting the first probability values ​​for different fault types corresponding to each sensor in the first sensor set through the deep belief network; and constructing a first probability matrix based on the first probability values ​​for different fault types corresponding to each sensor.

[0010] In one implementation of this application, before adjusting the first probability value according to the weight of each sensor's operating data to obtain the second probability matrix, the method further includes: determining the covariance between any two sensor operating data and determining the product of the standard deviations between any two sensor operating data; determining the correlation coefficient between any two sensor operating data based on the product of the covariance and the standard deviation, and constructing a correlation matrix based on the correlation coefficient; determining the credibility of each sensor operating data based on the correlation matrix; comparing the credibility with a preset weight assignment template to determine the weight of each sensor operating data; and adjusting the first probability value according to the weight of each sensor operating data to obtain the second probability matrix, specifically including: determining multiple first probability values ​​corresponding to each sensor operating data; and performing a weighted calculation on the multiple first probability values ​​corresponding to the current sensor based on the weight of the current sensor operating data to obtain the second probability matrix.

[0011] In one implementation of this application, a first set of fault types and a set of fault types to be tested corresponding to a wind turbine are obtained based on a second probability matrix and various environmental data. Specifically, this includes: performing synthesis calculations on the second probability values ​​in the second probability matrix based on preset synthesis rules to obtain first synthesized probability values ​​corresponding to different fault types; inputting various environmental data into a preset environmental analysis neural network model to obtain reference fault types and reference fault probabilities corresponding to various environmental data; sorting the obtained first synthesized probability values ​​to obtain a first fault type that meets preset fault conditions; sorting the obtained reference fault probabilities to obtain a second fault type that meets preset fault conditions; establishing a first set of fault types based on fault types that belong to both the first and second fault types; and establishing a set of fault types to be tested based on fault types that belong to either the first or the second fault type.

[0012] This embodiment of the application synthesizes the second probability values ​​in the second probability matrix to obtain a first synthesized probability value, and inputs the acquired environmental data into a pre-set environmental analysis neural network model to obtain a reference fault type and a reference fault probability. This allows for the comprehensive detection of wind turbine operational faults by combining wind turbine operating data and environmental data, thereby improving the accuracy of fault detection. Furthermore, this embodiment sorts the detected fault types according to their probability of occurrence, thereby filtering out fault types with higher probabilities to narrow down the fault range and facilitate inspection and maintenance of the wind turbine by personnel.

[0013] In one implementation of this application, a third probability matrix is ​​obtained by adjusting the probability values ​​in a second probability matrix based on multiple different environmental data. Specifically, this includes: determining the average reference fault probability corresponding to each reference fault type based on the reference fault type and reference fault probability corresponding to the multiple different environmental data; determining the second probability value corresponding to the reference fault type in the second probability matrix, and multiplying the average reference probability value and the second probability value to adjust the probability values ​​in the second probability matrix; and obtaining the third probability matrix based on the adjusted probability values.

[0014] In one implementation of this application, after obtaining the third probability values ​​corresponding to different fault types in the set of fault types to be tested, the method further includes: assigning the minimum value to the data with zero probability in the third probability matrix; determining the product of multiple third probability values ​​corresponding to the same fault type in the third probability matrix based on the set of fault types to be tested; summing the products of the third probability values ​​corresponding to different fault types in the third probability matrix to obtain the total probability; and obtaining the probability values ​​corresponding to different fault types in the set of fault types to be tested based on the product of multiple third probability values ​​corresponding to the same fault type and the total probability.

[0015] In one implementation of this application, the output detection fault of the wind turbine is obtained based on the first fault type set and the third probability value. Specifically, this includes: comparing the probability values ​​corresponding to different fault types in the fault type set to be tested with a preset fault probability threshold to remove fault types that are less than the preset fault probability threshold; sorting the remaining fault types in order of their fault probabilities to select a preset number of fault types as the second fault type set; and using the fault types in the first fault type set and the second fault type set as the detection faults of the wind turbine.

[0016] This application provides a wind turbine fault detection device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: acquire various operating data of the wind turbine through a first set of sensors installed on the wind turbine; wherein the operating data includes at least one or more of the following: temperature, vibration frequency, and rotational speed of the wind turbine; input the acquired various operating data into a deep belief network to obtain first probability values ​​for different fault types corresponding to the various operating data, and construct a first probability matrix based on the first probability values; and, based on the weight of each sensor's operating data, assign a first probability value to the first probability matrix. The probability values ​​are adjusted to obtain a second probability matrix. A second set of sensors installed on the wind turbine acquires various environmental data during the wind turbine's operation. These environmental data include at least one or more of the following: temperature, atmospheric pressure, and wind speed. Based on the second probability matrix and the various environmental data, a first set of fault types and a set of fault types to be tested for the wind turbine are obtained. The second probability value in the second probability matrix is ​​adjusted based on the various environmental data to obtain a third probability matrix. Based on a preset synthesis rule and the third probability matrix, third probability values ​​corresponding to different fault types in the set of fault types to be tested are obtained. Based on the first set of fault types and the third probability values, the output fault detection of the wind turbine is obtained.

[0017] This application provides a non-volatile computer storage medium storing computer-executable instructions. The computer-executable instructions are configured to: acquire various operating data of a wind turbine using a first set of sensors installed on the wind turbine; wherein the operating data includes at least one or more of the following: temperature, vibration frequency, and rotational speed of the wind turbine; input the acquired operating data into a deep belief network to obtain first probability values ​​for different fault types corresponding to the various operating data, and construct a first probability matrix based on the first probability values; adjust the first probability values ​​according to the weight of each sensor's operating data to obtain a second probability matrix; and then, using the data installed on the wind turbine... The second set of sensors acquires various environmental data during the operation of the wind turbine. These environmental data include at least one or more of the following: temperature, atmospheric pressure, and wind speed. Based on the second probability matrix and the various environmental data, a first set of fault types and a set of fault types to be tested for the wind turbine are obtained. Based on the various environmental data, the second probability value in the second probability matrix is ​​adjusted to obtain a third probability matrix. Based on a preset synthesis rule and the third probability matrix, third probability values ​​corresponding to different fault types in the set of fault types to be tested are obtained. Based on the first set of fault types and the third probability values, the output fault detection of the wind turbine is obtained.

[0018] The above-mentioned at least one technical solution adopted in this application embodiment can achieve the following beneficial effects: Firstly, by installing a first sensor set on the wind turbine, the operating data of the wind turbine can be monitored in real time to detect its operating status based on the acquired operating data. Secondly, this application embodiment establishes a first probability matrix based on the received operating data and adjusts the first probability matrix according to the weight value corresponding to each sensor, thereby improving the accuracy of the output fault type of the wind turbine. Furthermore, since the operating status of the wind turbine is also affected by different environmental factors, this application embodiment installs a second sensor set on the wind turbine. This second sensor set acquires current environmental data in real time, and the environmental data is analyzed to adjust the second probability matrix, obtaining a third probability matrix. This further adjusts the probability matrix, making the output fault type more accurate. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0020] Figure 1 A flowchart of a wind turbine fault detection method provided in this application embodiment;

[0021] Figure 2 This is a schematic diagram of the structure of a wind turbine fault detection device provided in an embodiment of this application. Detailed Implementation

[0022] This application provides a method, equipment, and medium for detecting faults in wind turbine generators.

[0023] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0024] Wind turbines are mostly installed in remote areas rich in wind energy resources, such as high mountains, wilderness, islands, and even offshore. Traditional maintenance strategies for wind turbine malfunctions heavily rely on regular routine inspections and reactive repairs. Therefore, in the event of a serious wind turbine failure, it is difficult to detect the fault in a timely manner, thus affecting the normal operation of the wind turbine.

[0025] To address the aforementioned issues, this application provides a method, device, and medium for detecting wind turbine faults. By installing a first set of sensors on the wind turbine, its operational data can be monitored in real time, allowing for the detection of the wind turbine's operating status based on the acquired data. Secondly, this application establishes a first probability matrix based on the received operational data and adjusts the first probability matrix according to the weight value corresponding to each sensor, thereby improving the accuracy of the output fault type for the wind turbine. Furthermore, since the operating status of the wind turbine is also affected by various environmental factors, this application installs a second set of sensors on the wind turbine. This second set of sensors acquires current environmental data in real time, and the environmental data is analyzed to adjust the second probability matrix, resulting in a third probability matrix. This further adjusts the probability matrix, making the output fault type more accurate.

[0026] The technical solutions proposed in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0027] Figure 1 This is a flowchart illustrating a wind turbine fault detection method provided in an embodiment of this application. Figure 1 As shown, the wind turbine fault detection method includes the following steps:

[0028] S101. Various operating data of the wind turbine are acquired through a first set of sensors installed on the wind turbine.

[0029] In one embodiment of this application, the first sensor set includes multiple sensors, such as temperature sensors and vibration sensors. Different sensors are positioned at different locations on the wind turbine to acquire operational data of the wind turbine. The operational data includes at least one or more of the following: temperature, vibration frequency, and rotational speed of the wind turbine.

[0030] S102 The wind turbine generator fault detection equipment inputs the acquired various operating data into a deep belief network to obtain the first probability value of different fault types corresponding to the various operating data, and constructs a first probability matrix based on the first probability value.

[0031] In one embodiment of this application, operational data samples corresponding to a first sensor set are acquired. A reference deep belief network is constructed to extract feature vectors from the operational data samples. A regression prediction layer is superimposed on the reference deep belief network. Based on the regression prediction layer and the feature vectors, the fault state of the wind turbine is predicted, and a fault prediction value is output. The fault prediction value is compared with a preset fault reference value to obtain a fault prediction error value. The parameters in the reference deep belief network are adjusted based on the fault prediction error value to obtain a deep belief network.

[0032] Specifically, the reference deep belief network in this embodiment is a Restricted Boltzmann Machine (RBM) at its bottom layer, consisting of a visible layer (data input layer) and a hidden layer (feature extraction layer). The visible layer has 6 neurons (adjustable according to actual needs), the hidden layer has 3 neurons, and the hidden layer has 3 layers. The output of the RBM formed by the first visible layer and the first hidden layer is used as the input to the next hidden layer, thus obtaining the output of the next hidden layer. This process is repeated to form a complete deep belief network.

[0033] Furthermore, in this embodiment, a regression prediction layer is superimposed on a reference deep belief network. The data features extracted by the deep belief network, along with selected label data, are used to perform supervised prediction of the engine health status. The data features extracted by the deep belief network are essentially feature vectors converted from the original input data through a neural network. Secondly, to avoid the "gradient diffusion" problem during training, this embodiment employs a greedy layer-by-layer algorithm for neural network training. The obtained fault prediction value is compared with a preset fault reference value to obtain the error between the fault prediction value output by the current regression prediction layer and the preset fault reference value. Based on this error, the parameters in the reference deep belief network are readjusted until the error between the fault prediction value output by the regression prediction layer and the preset fault reference value meets the current prediction conditions. This completes the model adjustment, resulting in the deep belief network.

[0034] In one embodiment of this application, the operational data acquired from the first sensor set are sequentially input into a deep belief network. The deep belief network outputs a first probability value for each sensor in the first sensor set corresponding to different fault types. Based on the first probability values ​​for each sensor corresponding to different fault types, a first probability matrix is ​​constructed.

[0035] Specifically, after acquiring the operational data sent by each sensor in the first sensor set, the operational data is input into a deep belief network. This deep belief network analyzes each piece of operational data to obtain multiple fault types corresponding to each data point, as well as a first probability value for each fault type. Based on these different probability values, a first probability matrix can be constructed.

[0036] S103. Adjust the first probability value according to the weight of the operating data of each sensor to obtain the second probability matrix.

[0037] In one embodiment of this application, the covariance and the standard deviation product between any two sensor operating data are determined. Based on the product of covariance and standard deviation, a correlation coefficient between any two sensor operating data is determined, and a correlation matrix is ​​constructed based on the correlation coefficient. The reliability of each sensor operating data is determined based on the correlation matrix. The reliability is compared with a preset weight assignment template to determine the weight of each sensor operating data. According to the weight of each sensor operating data, the first probability value is adjusted to obtain a second probability matrix, specifically including determining multiple first probability values ​​corresponding to each sensor operating data. Based on the weight of the current sensor operating data, the multiple first probability values ​​corresponding to the current sensor are weighted and calculated to obtain the second probability matrix.

[0038] Specifically, assume m i (T n ) is the probability value that the diagnostic result of the i-th sensor's operating data is fault type n, and m is the probability value of the fault type n. j (T n Let be the probability value of the diagnostic result of the j-th sensor's operating data being fault type n. Based on the probability values ​​of these two sensors, calculate their covariance and standard deviation product. Calculate the ratio between this covariance and the standard deviation product; this ratio is the correlation coefficient between the two sensor operating data. Calculate the correlation coefficient between any two sensor data points and construct a matrix based on the obtained correlation coefficients to obtain the correlation matrix.

[0039] Furthermore, according to the formula

[0040]

[0041] Obtain sensor data m i The credibility of. Among them, S ij This is the correlation coefficient.

[0042] Furthermore, by comparing the obtained credibility with the preset weight assignment template, the weight value corresponding to the credibility can be determined, and the first probability value can be adjusted based on the determined weight value.

[0043] Specifically, in the first probability matrix, the first probability values ​​corresponding to multiple different fault types for the same sensor data are determined. By multiplying the weight value corresponding to the sensor data with the corresponding first probability value, the first probability value can be adjusted to obtain the second probability matrix.

[0044] S104. By using a second set of sensors installed on the wind turbine, various environmental data during the operation of the wind turbine are acquired.

[0045] In one embodiment of this application, different environmental conditions can also have a certain impact on wind turbines. For example, environmental conditions such as excessively low temperature, low air pressure, or excessively high wind speed can all have a certain impact on wind turbines.

[0046] Specifically, in order to acquire environmental data from the wind turbine, this embodiment of the application also installs a second set of sensors on the wind turbine. For example, a wind speed sensor can be installed at the center intersection of multiple blades of the wind turbine to monitor the wind speed in real time; similarly, a temperature sensor can be installed on the back of the wind turbine to monitor the current ambient temperature in real time. It should be noted that the environmental data in this embodiment of the application includes at least one or more of the following: air temperature, atmospheric pressure, and wind speed of the environment where the wind turbine is located.

[0047] S105. Based on the second probability matrix and various environmental data, obtain the first set of fault types and the set of fault types to be tested corresponding to the wind turbine.

[0048] In one embodiment of this application, based on preset synthesis rules, the second probability values ​​in the second probability matrix are synthesized to obtain first synthesized probability values ​​corresponding to different fault types. Multiple environmental data are input into a preset environmental analysis neural network model to obtain reference fault types and reference fault probabilities corresponding to the different environmental data. The obtained first synthesized probability values ​​are sorted to obtain first fault types that meet preset fault conditions. The obtained reference fault probabilities are sorted to obtain second fault types that meet preset fault conditions. A first fault type set is established based on fault types belonging to both the first and second fault types, and a set of fault types to be tested is established based on fault types belonging to either the first or second fault type.

[0049] Specifically, in the embodiments of this application, the synthesis rules of DS evidence theory are used to calculate the probability value of each fault type.

[0050] First, calculate the normalization coefficient k:

[0051]

[0052] The formula for calculating the first composite probability value is:

[0053]

[0054] Using the above formula, the first composite probability value corresponding to different fault types can be obtained. All the obtained first composite probability values ​​are sorted from largest to smallest to determine the first fault type that meets the preset fault conditions. For example, assuming that a probability value of 70% or higher is a qualified first fault type, then fault types with a probability value greater than 70% are considered qualified first fault types.

[0055] Furthermore, this application embodiment pre-sets a pre-configured environmental analysis neural network model. The training process of this model is as follows: multiple environmental data samples are acquired as input values, the fault type and fault probability corresponding to each environmental sample are used as output values, and the input and output values ​​are input into the neural network model for training to obtain the pre-configured environmental analysis neural network model.

[0056] Furthermore, the environmental data corresponding to the wind turbine is input into the pre-set environmental analysis neural network model to obtain reference fault types and reference fault probabilities corresponding to various environmental data. The obtained reference fault probabilities are sorted to obtain a second fault type that meets the preset fault conditions. For example, the preset fault condition can be set to a fault probability greater than 70%, and the fault type corresponding to the probability value greater than 70% can be used as the second fault type.

[0057] Furthermore, the first fault type is compared with the second fault type to identify the fault types that share the same fault type, and these fault types are designated as the first fault type set. Next, fault types that belong only to the first fault type or only to the second fault type are identified as the fault types to be tested for further analysis.

[0058] S106. Based on various environmental data, the second probability value in the second probability matrix is ​​adjusted to obtain the third probability matrix.

[0059] In one embodiment of this application, an average reference fault probability is determined for each reference fault type based on reference fault types and reference fault probabilities corresponding to various environmental data. A second probability value corresponding to each reference fault type is determined in a second probability matrix, and the average reference probability value is multiplied by the second probability value to adjust the probability values ​​in the second probability matrix. A third probability matrix is ​​obtained based on the adjusted probability values.

[0060] Specifically, in the embodiments of this application, each piece of environmental data corresponds to multiple different reference fault types, and each reference fault type corresponds to a reference fault probability. The fault probability corresponding to the same fault type in different environmental data is not necessarily the same. Therefore, by averaging the probability values ​​of the same reference fault type corresponding to different environmental data, the average reference fault probability corresponding to each reference fault type can be obtained.

[0061] Furthermore, in the second probability matrix, the second probability value corresponding to the reference fault type is determined, and the average reference fault probability corresponding to the reference fault type is multiplied by the corresponding second probability value. The second probability matrix can then be adjusted based on the environmental data of the wind turbine to obtain the third probability matrix.

[0062] S107. Based on the preset synthesis rules and the third probability matrix, obtain the third probability values ​​corresponding to different fault types in the set of fault types to be tested.

[0063] In one embodiment of this application, the data with a probability of zero in the third probability matrix are assigned a minimum value. Based on the set of fault types to be tested, the product of multiple third probability values ​​corresponding to the same fault type is determined in the third probability matrix. The products of the third probability values ​​corresponding to different fault types in the third probability matrix are summed to obtain the total probability. Based on the product of multiple third probability values ​​corresponding to the same fault type and the total probability, the probability values ​​corresponding to different fault types in the set of fault types to be tested are obtained.

[0064] Specifically, to prevent the product of probability values ​​from being zero, this embodiment assigns a minimum value to the data that are zero in the third probability matrix; for example, 0 can be assigned the value 0.01. Next, data with the same fault type in the third probability matrix are divided, and the probability values ​​of the same fault type are multiplied. This yields the product of the third probability values ​​corresponding to different fault types. Then, the summation of the products of multiple third probability values ​​corresponding to the third matrix yields the total probability. Finally, the ratio of the product of the third probability values ​​corresponding to each fault type to the total probability is calculated to obtain the probability value corresponding to each fault type.

[0065] S108. Based on the first fault type set and the third probability value, the output detection fault of the wind turbine is obtained.

[0066] In one embodiment of this application, the probability values ​​corresponding to different fault types in the set of fault types to be tested are compared with a preset fault probability threshold to eliminate fault types with probabilities lower than the preset threshold. The remaining fault types are then sorted according to their fault probabilities to select a preset number of fault types as a second fault type set. The fault types in the first fault type set and the second fault type set are used as the detected faults of the wind turbine.

[0067] Specifically, based on the set of fault types to be tested and the probability values ​​corresponding to different fault types in the third probability matrix, the probability value corresponding to each fault type in the set of fault types to be tested can be obtained. This probability value is then compared with a preset fault probability threshold, and fault types with a probability value less than the preset fault probability threshold are removed.

[0068] Furthermore, the remaining fault types are sorted from highest to lowest probability value, and a corresponding number of fault types are selected from highest to lowest probability value to form a second fault type set. For example, the top 3 fault types by probability value can be used as the second fault type set. The fault types in the first and second fault type sets are used as the detected faults of the wind turbine, and these detected faults are displayed to remind staff to perform maintenance and inspection on the wind turbine.

[0069] Figure 2 This is a schematic diagram of the structure of a wind turbine fault detection device provided in an embodiment of this application. Figure 2 As shown, a wind turbine fault detection device includes:

[0070] At least one processor; and,

[0071] A memory communicatively connected to the at least one processor; wherein,

[0072] The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to:

[0073] A first set of sensors installed on the wind turbine acquires various operating data of the wind turbine; wherein the operating data includes at least one or more of the following: temperature, vibration frequency, and rotational speed of the wind turbine.

[0074] The acquired multiple operational data are input into a deep belief network to obtain first probability values ​​for different fault types corresponding to the multiple operational data, and a first probability matrix is ​​constructed based on the first probability values.

[0075] The first probability value is adjusted according to the weight of the data from each sensor to obtain the second probability matrix;

[0076] A second set of sensors installed on the wind turbine acquires various environmental data during the operation of the wind turbine; wherein the environmental data includes at least one or more of the following: temperature, atmospheric pressure, and wind speed of the environment in which the wind turbine is located.

[0077] Based on the second probability matrix and the various environmental data, the first set of fault types and the set of fault types to be tested corresponding to the wind turbine are obtained.

[0078] Based on the various environmental data, the second probability value in the second probability matrix is ​​adjusted to obtain the third probability matrix;

[0079] Based on the preset synthesis rules and the third probability matrix, the third probability values ​​corresponding to different fault types in the set of fault types to be tested are obtained respectively;

[0080] Based on the first set of fault types and the third probability value, the output detection fault of the wind turbine is obtained.

[0081] This application embodiment also includes a non-volatile computer storage medium storing a computer-executable...

[0082] The computer-executable instructions are set as follows:

[0083] A first set of sensors installed on the wind turbine acquires various operating data of the wind turbine; wherein the operating data includes at least one or more of the following: temperature, vibration frequency, and rotational speed of the wind turbine.

[0084] The acquired multiple operational data are input into a deep belief network to obtain first probability values ​​for different fault types corresponding to the multiple operational data, and a first probability matrix is ​​constructed based on the first probability values.

[0085] The first probability value is adjusted according to the weight of the data from each sensor to obtain the second probability matrix;

[0086] A second set of sensors installed on the wind turbine acquires various environmental data during the operation of the wind turbine; wherein the environmental data includes at least one or more of the following: temperature, atmospheric pressure, and wind speed of the environment in which the wind turbine is located.

[0087] Based on the second probability matrix and the various environmental data, the first set of fault types and the set of fault types to be tested corresponding to the wind turbine are obtained.

[0088] Based on the various environmental data, the second probability value in the second probability matrix is ​​adjusted to obtain the third probability matrix;

[0089] Based on the preset synthesis rules and the third probability matrix, the third probability values ​​corresponding to different fault types in the set of fault types to be tested are obtained respectively;

[0090] Based on the first set of fault types and the third probability value, the output detection fault of the wind turbine is obtained.

[0091] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0092] The foregoing has described specific embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0093] The above description is merely an embodiment of this application and is not intended to limit this application. For those skilled in the art, various modifications and variations can be made to the embodiments of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the embodiments of this application should be included within the scope of the claims of this application.

Claims

1. A method for detecting faults in a wind turbine generator, characterized in that, The method includes: A first set of sensors installed on the wind turbine acquires various operating data of the wind turbine; wherein, the operating data includes one or more of the following: temperature, vibration frequency, and rotational speed of the wind turbine. The acquired operational data are input into a deep belief network to obtain first probability values ​​for different fault types corresponding to the acquired operational data, and a first probability matrix is ​​constructed based on the first probability values. Specifically, this includes: inputting the acquired operational data from a first sensor set into the deep belief network in sequence; outputting the first probability values ​​for different fault types corresponding to each sensor in the first sensor set through the deep belief network; and constructing the first probability matrix based on the first probability values ​​for different fault types corresponding to each sensor. Determine the covariance between any two sensor operating data, and determine the product of the standard deviations between the any two sensor operating data; Based on the product of the covariance and the standard deviation, the correlation coefficient between any two sensor operating data is determined, and a correlation matrix is ​​constructed based on the correlation coefficient. The reliability of the operational data of each sensor is determined based on the correlation matrix. The confidence level is compared with a preset weight assignment template to determine the weight of each sensor's operating data. The first probability value is adjusted according to the weight of each sensor's operating data to obtain a second probability matrix; specifically, this includes: determining multiple first probability values ​​corresponding to each sensor's operating data; and performing a weighted calculation on the multiple first probability values ​​corresponding to the current sensor based on the weight of the current sensor's operating data to obtain the second probability matrix. A second set of sensors installed on the wind turbine acquires various environmental data during the operation of the wind turbine; wherein, the environmental data includes one or more of the following: temperature, atmospheric pressure, and wind speed of the environment in which the wind turbine is located. Based on the second probability matrix and various environmental data, a first set of fault types and a set of fault types to be tested corresponding to the wind turbine are obtained; specifically, based on a preset synthesis rule, the second probability values ​​in the second probability matrix are synthesized to obtain the first synthesized probability values ​​corresponding to different fault types. The various environmental data are input into a pre-set environmental analysis neural network model to obtain the reference fault type and reference fault probability corresponding to the various environmental data respectively; The obtained first synthetic probability values ​​are sorted to obtain a first fault type that meets the preset fault conditions; The obtained reference fault probabilities are sorted to obtain a second fault type that meets the preset fault conditions; Based on fault types that belong to both the first fault type and the second fault type, a first fault type set is established; and based on fault types that belong to either the first fault type or the second fault type, the set of fault types to be tested is established. Based on the various environmental data, the second probability value in the second probability matrix is ​​adjusted to obtain a third probability matrix; specifically, this includes: determining the average reference fault probability corresponding to each reference fault type based on the reference fault type and reference fault probability corresponding to the various environmental data; determining the second probability value corresponding to the reference fault type in the second probability matrix, and multiplying the average reference fault probability value with the second probability value to adjust the probability values ​​in the second probability matrix; and obtaining the third probability matrix based on the adjusted probability values. Based on the preset synthesis rules and the third probability matrix, the third probability values ​​corresponding to different fault types in the set of fault types to be tested are obtained, and the data with a probability of zero in the third probability matrix are assigned the minimum value. Based on the set of fault types to be tested, the product of multiple third probability values ​​corresponding to the same fault type is determined in the third probability matrix; The total probability is obtained by summing the products of the third probability values ​​corresponding to different fault types in the third probability matrix. Based on the product of multiple third probability values ​​corresponding to the same fault type and the total probability, the probability values ​​corresponding to different fault types in the set of fault types to be tested are obtained respectively. Based on the first set of fault types and the third probability value, the output fault of the wind turbine is obtained; The probability values ​​corresponding to different fault types in the set of fault types to be tested are compared with a preset fault probability threshold, so as to remove fault types that are less than the preset fault probability threshold. The remaining fault types are sorted in order of their fault probability, and a preset number of fault types are selected as the second fault type set. The fault types in the first fault type set and the second fault type set are used as the detected faults of the wind turbine.

2. The method for detecting faults in a wind turbine generator according to claim 1, characterized in that, Before inputting the acquired various operational data into the deep belief network, the method further includes: Obtain the operational data samples corresponding to the first sensor set; A reference deep belief network is constructed to extract the feature vectors corresponding to the running data samples through the reference deep belief network; A regression prediction layer is superimposed on the reference deep belief network. Based on the regression prediction layer and the feature vector, the fault state of the wind turbine is predicted, and the fault prediction value is output. The fault prediction value is compared with the preset fault reference value to obtain the fault prediction error value. The parameters in the reference depth belief network are adjusted based on the fault prediction error value to obtain the depth belief network.

3. A wind turbine generator fault detection device, comprising: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to: A first set of sensors installed on the wind turbine acquires various operating data of the wind turbine; wherein, the operating data includes one or more of the following: temperature, vibration frequency, and rotational speed of the wind turbine. The acquired operational data are input into a deep belief network to obtain first probability values ​​for different fault types corresponding to the acquired operational data, and a first probability matrix is ​​constructed based on the first probability values. Specifically, this includes: inputting the acquired operational data from a first sensor set into the deep belief network in sequence; outputting the first probability values ​​for different fault types corresponding to each sensor in the first sensor set through the deep belief network; and constructing the first probability matrix based on the first probability values ​​for different fault types corresponding to each sensor. Determine the covariance between any two sensor operating data, and determine the product of the standard deviations between the any two sensor operating data; Based on the product of the covariance and the standard deviation, the correlation coefficient between any two sensor operating data is determined, and a correlation matrix is ​​constructed based on the correlation coefficient. The reliability of the operational data of each sensor is determined based on the correlation matrix. The confidence level is compared with a preset weight assignment template to determine the weight of each sensor's operating data. The first probability value is adjusted according to the weight of each sensor's operating data to obtain a second probability matrix; specifically, this includes: determining multiple first probability values ​​corresponding to each sensor's operating data; and performing a weighted calculation on the multiple first probability values ​​corresponding to the current sensor based on the weight of the current sensor's operating data to obtain the second probability matrix. A second set of sensors installed on the wind turbine acquires various environmental data during the operation of the wind turbine; wherein, the environmental data includes one or more of the following: temperature, atmospheric pressure, and wind speed of the environment in which the wind turbine is located. Based on the second probability matrix and various environmental data, a first set of fault types and a set of fault types to be tested corresponding to the wind turbine are obtained; specifically, based on a preset synthesis rule, the second probability values ​​in the second probability matrix are synthesized to obtain the first synthesized probability values ​​corresponding to different fault types. The various environmental data are input into a pre-set environmental analysis neural network model to obtain the reference fault type and reference fault probability corresponding to the various environmental data respectively; The obtained first synthetic probability values ​​are sorted to obtain a first fault type that meets the preset fault conditions; The obtained reference fault probabilities are sorted to obtain a second fault type that meets the preset fault conditions; Based on fault types that belong to both the first fault type and the second fault type, a first fault type set is established; and based on fault types that belong to either the first fault type or the second fault type, the set of fault types to be tested is established. Based on the various environmental data, the second probability value in the second probability matrix is ​​adjusted to obtain a third probability matrix; specifically, this includes: determining the average reference fault probability corresponding to each reference fault type based on the reference fault type and reference fault probability corresponding to the various environmental data; determining the second probability value corresponding to the reference fault type in the second probability matrix, and multiplying the average reference fault probability value with the second probability value to adjust the probability values ​​in the second probability matrix; and obtaining the third probability matrix based on the adjusted probability values. Based on the preset synthesis rules and the third probability matrix, the third probability values ​​corresponding to different fault types in the set of fault types to be tested are obtained, and the data with a probability of zero in the third probability matrix are assigned the minimum value. Based on the set of fault types to be tested, the product of multiple third probability values ​​corresponding to the same fault type is determined in the third probability matrix; The total probability is obtained by summing the products of the third probability values ​​corresponding to different fault types in the third probability matrix. Based on the product of multiple third probability values ​​corresponding to the same fault type and the total probability, the probability values ​​corresponding to different fault types in the set of fault types to be tested are obtained respectively. Based on the first set of fault types and the third probability value, the output fault of the wind turbine is obtained; The probability values ​​corresponding to different fault types in the set of fault types to be tested are compared with a preset fault probability threshold, so as to remove fault types that are less than the preset fault probability threshold. The remaining fault types are sorted in order of their fault probability, and a preset number of fault types are selected as the second fault type set. The fault types in the first fault type set and the second fault type set are used as the detected faults of the wind turbine.

4. A non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured as follows: A first set of sensors installed on the wind turbine acquires various operational data of the wind turbine; among which, The operating data includes one or more of the temperature, vibration frequency, and rotational speed of the wind turbine. The acquired operational data are input into a deep belief network to obtain first probability values ​​for different fault types corresponding to the acquired operational data, and a first probability matrix is ​​constructed based on the first probability values. Specifically, this includes: inputting the acquired operational data from a first sensor set into the deep belief network in sequence; outputting the first probability values ​​for different fault types corresponding to each sensor in the first sensor set through the deep belief network; and constructing the first probability matrix based on the first probability values ​​for different fault types corresponding to each sensor. Determine the covariance between any two sensor operating data, and determine the product of the standard deviations between the any two sensor operating data; Based on the product of the covariance and the standard deviation, the correlation coefficient between any two sensor operating data is determined, and a correlation matrix is ​​constructed based on the correlation coefficient. The reliability of the operational data of each sensor is determined based on the correlation matrix. The confidence level is compared with a preset weight assignment template to determine the weight of each sensor's operating data. The first probability value is adjusted according to the weight of each sensor's operating data to obtain a second probability matrix; specifically, this includes: determining multiple first probability values ​​corresponding to each sensor's operating data; and performing a weighted calculation on the multiple first probability values ​​corresponding to the current sensor based on the weight of the current sensor's operating data to obtain the second probability matrix. A second set of sensors installed on the wind turbine acquires various environmental data during the operation of the wind turbine; wherein, the environmental data includes one or more of the following: temperature, atmospheric pressure, and wind speed of the environment in which the wind turbine is located. Based on the second probability matrix and various environmental data, a first set of fault types and a set of fault types to be tested corresponding to the wind turbine are obtained; specifically, based on a preset synthesis rule, the second probability values ​​in the second probability matrix are synthesized to obtain the first synthesized probability values ​​corresponding to different fault types. The various environmental data are input into a pre-set environmental analysis neural network model to obtain the reference fault type and reference fault probability corresponding to the various environmental data respectively; The obtained first synthetic probability values ​​are sorted to obtain a first fault type that meets the preset fault conditions; The obtained reference fault probabilities are sorted to obtain a second fault type that meets the preset fault conditions; Based on fault types that belong to both the first fault type and the second fault type, a first fault type set is established; and based on fault types that belong to either the first fault type or the second fault type, the set of fault types to be tested is established. Based on the various environmental data, the second probability value in the second probability matrix is ​​adjusted to obtain a third probability matrix; specifically, this includes: determining the average reference fault probability corresponding to each reference fault type based on the reference fault type and reference fault probability corresponding to the various environmental data; determining the second probability value corresponding to the reference fault type in the second probability matrix, and multiplying the average reference fault probability value with the second probability value to adjust the probability values ​​in the second probability matrix; and obtaining the third probability matrix based on the adjusted probability values. Based on the preset synthesis rules and the third probability matrix, the third probability values ​​corresponding to different fault types in the set of fault types to be tested are obtained, and the data with a probability of zero in the third probability matrix are assigned the minimum value. Based on the set of fault types to be tested, the product of multiple third probability values ​​corresponding to the same fault type is determined in the third probability matrix; The total probability is obtained by summing the products of the third probability values ​​corresponding to different fault types in the third probability matrix. Based on the product of multiple third probability values ​​corresponding to the same fault type and the total probability, the probability values ​​corresponding to different fault types in the set of fault types to be tested are obtained respectively. Based on the first set of fault types and the third probability value, the output fault of the wind turbine is obtained; The probability values ​​corresponding to different fault types in the set of fault types to be tested are compared with a preset fault probability threshold, so as to remove fault types that are less than the preset fault probability threshold. The remaining fault types are sorted in order of their fault probability, and a preset number of fault types are selected as the second fault type set. The fault types in the first fault type set and the second fault type set are used as the detected faults of the wind turbine.

Images