Wind generating set fault diagnosis system based on data analysis

Abstract

The invention relates to the technical field of wind power generation, and discloses a wind generating set fault diagnosis system based on data analysis, which comprises the following steps: acquiring wear reflection characteristic data and stress reflection characteristic data of each moving part in a wind generating set; inputting the wear reflection feature data of each moving part into a first recurrent neural network model to predict the future wear loss of each moving part; inputting the stress reflection characteristic data of each moving part into a second recurrent neural network model to predict a future stress value of each moving part; the future stress value and the future abrasion loss of each moving part are input into a pre-constructed nonlinear function, and the outage probability of the wind generating set under each moving part is obtained; judging whether the wind generating set has a fault moving part or not at the future moment according to all the outage probabilities; according to the invention, fault positioning and fault cause determination of the wind generating set are facilitated.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and more specifically, to a wind power generator fault diagnosis system based on data analysis. Background Art

[0002] As the global demand for renewable energy continues to increase, wind power generation, as a clean energy source, has become an important part of global electricity production; wind turbines are widely used in various environments and undertake important power supply tasks; however, due to the complex working environment of wind turbines and long-term exposure to natural conditions, various moving parts such as blades, gearboxes, bearings, etc. are prone to wear and damage under long-term loads, leading to equipment failures; these failures will not only affect the operating efficiency of the generator set, but may also cause long-term downtime and maintenance, thereby reducing the reliability and economic benefits of the entire wind power generation system.

[0003] At present, fault diagnosis of wind turbines mostly relies on manual inspections, regular checks and traditional diagnostic methods. Although these methods guarantee the normal operation of the equipment to a certain extent, they have limitations such as the inability to detect potential faults in a timely manner, high manual intervention costs, long equipment downtime, and inability to quickly identify the cause of the fault. Therefore, how to monitor and diagnose the operating status of wind turbines through scientific, accurate and real-time means has become the key to improving equipment reliability, reducing downtime and improving power generation efficiency. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a wind turbine generator fault diagnosis system based on data analysis.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A wind turbine generator fault diagnosis system based on data analysis, comprising:

[0007] The data acquisition module is used to collect the wear reflection characteristic data and stress reflection characteristic data of each moving part in the wind turbine generator set within a predetermined time range from TN to T, where T and N are integers greater than zero;

[0008] A first prediction module is used to input the wear reflection characteristic data of each moving part into a first recursive neural network model with a preset wear amount prediction, so as to predict the future wear amount of each moving part at a future time R, where R is an integer greater than zero;

[0009] A second prediction module is used to input the stress reflection characteristic data of each moving component into a second recursive neural network model with a preset predicted stress value, so as to predict the future stress value of each moving component at a future time R;

[0010] A probability analysis module, used for inputting the future stress value and the future wear amount of each moving component into a pre-built nonlinear function respectively, so as to obtain the shutdown probability of the wind turbine generator set under each moving component;

[0011] The fault location module is used to determine whether there is a faulty moving component in the wind turbine generator set at the future time R based on all outage probabilities. If not, T=T+v is set and the data acquisition module is triggered; if so, the unique identifier of the corresponding faulty moving component is fed back, and v is an integer greater than zero.

[0012] Furthermore, the wear reflection characteristic data includes operating status characteristic data and environmental characteristic data, and the stress reflection characteristic data includes operating condition characteristic data and environmental characteristic data, wherein the operating status characteristic data includes the cumulative operating time, operating speed, operating torque, wear amount, material thickness and lubricating oil amount at each moment; the operating condition characteristic data includes the cumulative operating time, stress value and material strength at each moment, and the environmental characteristic data includes ambient temperature, ambient humidity and ambient wind speed.

[0013] Furthermore, the training method of the first recursive neural network model is as follows:

[0014] Acquire historical wear amount training data, and divide the historical wear amount training data into a wear amount training set and a wear amount test set, wherein the historical wear amount training data includes a plurality of wear reflection feature data and their corresponding future wear amounts;

[0015] Constructing a first recursive neural network for performing a regression task, taking the wear reflection characteristic data in the wear amount training set as the input of the first recursive neural network, taking the future wear amount in the wear amount training set as the output of the first recursive neural network, training the first recursive neural network, and obtaining an initial first recursive neural network;

[0016] The initial first recursive neural network is model verified using the wear test set, and an initial first recursive neural network that is less than or equal to a preset first test error threshold is output as the first recursive neural network model for executing wear prediction.

[0017] Furthermore, the training method of the second recursive neural network model is as follows:

[0018] Acquire historical stress value training data, and divide the historical stress value training data into a stress value training set and a stress value test set, wherein the historical stress value training data includes a plurality of stress reflection characteristic data and corresponding future stress values;

[0019] A second recursive neural network for performing a regression task is constructed, the stress reflection characteristic data in the stress value training set is used as the input of the second recursive neural network, and the future stress value in the stress value training set is used as the output of the second recursive neural network, and the second recursive neural network is trained to obtain an initial second recursive neural network;

[0020] The initial second recursive neural network is model verified using the stress value test set, and an initial second recursive neural network that is less than or equal to a preset second test error threshold is output as the second recursive neural network model for executing stress value prediction.

[0021] Furthermore, the expression of the pre-constructed nonlinear function is as follows:

[0022]

[0023] Where: P 停运 (R) represents the outage probability of the wind turbine, σ′R represents the normalized future stress value, M′R represents the normalized future wear value, ω1 and ω2 are weight coefficients greater than zero, and b is the bias.

[0024] Furthermore, the values ​​of the weight coefficient and the bias in the nonlinear function are obtained by optimizing by the least square method;

[0025] Among them, the optimization method using the least squares method is as follows:

[0026] a1: defines the loss function used to optimize the nonlinear function, and randomly initializes the weight coefficients ω1, ω2 and bias b;

[0027] The specific expression of the loss function is as follows:

[0028]

[0029] Where: J(ω1,ω2,b) is the loss function, which represents the sum of the prediction errors of all sample points, P 停运，j represents the actual outage probability of the jth sample, and N is the number of samples;

[0030] a2: Calculate the gradient of the loss function with respect to the weight coefficients ω1, ω2 and bias b;

[0031] Among them, the calculation formula of the gradient of the weight coefficient ω1 is as follows:

[0032]

[0033] Among them, the calculation formula of the gradient of the weight coefficient ω2 is as follows:

[0034]

[0035] Among them, the calculation formula of the gradient of bias b is as follows:

[0036]

[0037] Where: represents the gradient of the weight coefficient ω1, represents the gradient of the weight coefficient ω2, represents the gradient of bias b;

[0038] a3: Use stochastic gradient descent method to update weight coefficients ω1, ω2 and bias b. The update formula is as follows:

[0039]

[0040] Where: t represents the number of current iterations, η is the learning rate;

[0041] a4: Repeat steps a2 and a3 above until the value of the loss function J(ω1,ω2,b) is less than or equal to the preset error threshold, or the predetermined number of iterations is reached, then the loop ends and the values ​​of the weight coefficients and biases in the optimized nonlinear function are obtained.

[0042] Furthermore, the step of judging whether there is a faulty moving component in the wind turbine generator set at the future time R includes:

[0043] b1: extracting the i-th outage probability from all outage probabilities, where the number of outage probabilities is K, and i and K are integers greater than zero;

[0044] b2: Compare the ith outage probability with the preset outage probability threshold. If the ith outage probability is less than the outage probability threshold, it is determined that there is no faulty moving component in the wind turbine generator set at the future time R, and i=i+1 is set, and the process returns to step b1; if the ith outage probability is greater than or equal to the outage probability threshold, it is determined that there is a faulty moving component in the wind turbine generator set at the future time R, and the moving component corresponding to the ith outage probability is marked as a faulty moving component.

[0045] b3: Repeat the above steps b1 to b2 until i=K, then end the loop and obtain a judgment result, which includes one of the following situations: there is a faulty moving part in the wind turbine generator set and there is no faulty moving part in the wind turbine generator set.

[0046] Furthermore, the system further comprises:

[0047] An information acquisition module, for acquiring a vibration signal set of the faulty moving component within a predetermined time period when a faulty moving component is found, and generating an actual vibration waveform diagram of the faulty moving component according to the vibration signal set, wherein the vibration signal set includes P vibration values, where P is an integer greater than zero;

[0048] An information retrieval module, used to obtain corresponding fault analysis data of the faulty moving component according to a mapping relationship between the unique identifier and the fault analysis data, wherein the fault analysis data includes multiple fault causes and a fault vibration waveform corresponding to each fault cause;

[0049] The fault diagnosis module is used to compare and analyze the actual vibration waveform diagram and the fault vibration waveform diagram to determine the fault cause of the faulty moving part.

[0050] Further, the similarity between the actual vibration waveform and each fault vibration waveform is calculated using a perceptual hash algorithm to obtain multiple similarities;

[0051] The multiple similarities are sorted from large to small according to the numerical values, and the fault cause corresponding to the first fault vibration waveform diagram in the similarity sorting is taken as the fault cause of the faulty moving component.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] The present application discloses a wind turbine generator fault diagnosis system based on data analysis, comprising: collecting wear reflection characteristic data and stress reflection characteristic data of each moving part in the wind turbine generator set; inputting the wear reflection characteristic data of each moving part into a first recursive neural network model respectively to predict the future wear amount of each moving part; inputting the stress reflection characteristic data of each moving part into a second recursive neural network model respectively to predict the future stress value of each moving part; inputting the future stress value and future wear amount of each moving part into a pre-constructed nonlinear function respectively to obtain the shutdown probability of the wind turbine generator set under each moving part; judging whether there is a faulty moving part in the wind turbine generator set at a future moment according to all the shutdown probabilities; based on the above technical features, the present invention is conducive to dynamically predicting the shutdown risk caused by each moving part of the wind turbine generator set, providing early warning before the fault occurs, reducing unplanned shutdown caused by the fault, and improving the availability and operation efficiency of the unit; in addition, by further analyzing the vibration data of the faulty running part on the basis of determining the faulty running part, the present invention is conducive to quickly identifying the specific fault cause of the faulty moving part, avoiding the inefficiency of relying on manual intervention and experience judgment in the traditional method, and greatly improving the diagnosis efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1A schematic diagram of the module structure of a wind turbine generator set fault diagnosis system based on data analysis provided in Example 1 of the present invention;

[0055] Figure 2 A schematic diagram of the module structure of a wind turbine generator set fault diagnosis system based on data analysis provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] Example 1

[0058] See also Figure 1 As shown, this embodiment discloses a wind turbine generator fault diagnosis system based on data analysis, the system comprising:

[0059] The data acquisition module 110 is used to collect wear reflection characteristic data and stress reflection characteristic data of each moving part in the wind turbine generator set within a predetermined time range TN to T, where T and N are integers greater than zero;

[0060] It should be understood that the fault types of wind turbine generator sets mainly include mechanical faults and electrical faults. Compared with electrical fault types, since the various moving parts of wind turbine generator sets are under long-term load movement and exposed to the natural environment, they are more likely to fail or be damaged. Therefore, this solution is an application solution provided to solve the problem of mechanical fault diagnosis, aiming to greatly improve the operating reliability of wind turbine generator sets, reduce the length and number of downtime, and improve power generation efficiency; wherein, the moving parts include but are not limited to the blades, gearboxes and bearings of wind turbine generator sets;

[0061] The wear reflection characteristic data includes running state characteristic data and environmental characteristic data, and the stress reflection characteristic data includes running condition characteristic data and environmental characteristic data, wherein the running state characteristic data includes but is not limited to the accumulated running time, running speed, running torque, wear amount, material thickness and lubricating oil amount at each moment; the running condition characteristic data includes but is not limited to the accumulated running time, stress value and material strength at each moment, and the environmental characteristic data includes but is not limited to the ambient temperature, ambient humidity and ambient wind speed;

[0062] It should be noted that the wear reflection characteristic data and stress reflection characteristic data are collected through various sensors, including but not limited to speed sensors, torque sensors, visual sensors, temperature and humidity sensors, etc.

[0063] The first prediction module 120 is used to input the wear reflection characteristic data of each moving component into a first recursive neural network model for predicting the wear amount, so as to predict the future wear amount of each moving component at a future time R, where R is an integer greater than zero;

[0064] In implementation, the training method of the first recursive neural network model is as follows:

[0065] Acquire historical wear amount training data, and divide the historical wear amount training data into a wear amount training set and a wear amount test set, wherein the historical wear amount training data includes a plurality of wear reflection feature data and their corresponding future wear amounts;

[0066] It should be noted that the wear reflection characteristic data in the historical wear amount training data is actually collected by the technicians based on the historical data and experimental data, and the historical wear amount training data in the historical wear amount training data is actually recorded by the technicians based on the historical data and experimental data;

[0067] Constructing a first recursive neural network for performing a regression task, taking the wear reflection characteristic data in the wear amount training set as the input of the first recursive neural network, taking the future wear amount in the wear amount training set as the output of the first recursive neural network, training the first recursive neural network, and obtaining an initial first recursive neural network;

[0068] The initial first recursive neural network is model verified using the wear test set, and an initial first recursive neural network that is less than or equal to a preset first test error threshold is output as the first recursive neural network model for executing wear prediction.

[0069] The second prediction module 130 is used to input the stress reflection characteristic data of each moving component into the second recursive neural network model with preset predicted stress values, so as to predict the future stress value of each moving component at the future time R;

[0070] In implementation, the training method of the second recursive neural network model is as follows:

[0071] Acquire historical stress value training data, and divide the historical stress value training data into a stress value training set and a stress value test set, wherein the historical stress value training data includes a plurality of stress reflection characteristic data and corresponding future stress values;

[0072] It should be noted that the stress reflection characteristic data in the historical stress value training data is actually collected by the technicians based on the historical data and experimental data, and the historical stress value training data in the historical stress value training data is actually recorded by the technicians based on the historical data and experimental data;

[0073] A second recursive neural network for performing a regression task is constructed, the stress reflection characteristic data in the stress value training set is used as the input of the second recursive neural network, and the future stress value in the stress value training set is used as the output of the second recursive neural network, and the second recursive neural network is trained to obtain an initial second recursive neural network;

[0074] The initial second recursive neural network is model verified using the stress value test set, and an initial second recursive neural network that is less than or equal to a preset second test error threshold is output as the second recursive neural network model for executing stress value prediction.

[0075] The probability analysis module 140 is used to input the future stress value and the future wear amount of each moving component into the pre-built nonlinear function to obtain the shutdown probability of the wind turbine generator set under each moving component;

[0076] The expression of the pre-constructed nonlinear function is as follows:

[0077]

[0078] Where: P 停运 (R) represents the outage probability of the wind turbine generator set, σ′R represents the normalized future stress value, M′R represents the normalized future wear amount, ω1 and ω2 are weight coefficients greater than zero, indicating the contribution of stress value and wear amount to the outage probability, and b is the bias, which is used to adjust the offset of the output;

[0079] It should be understood that before the future stress value and the future wear amount are input into the pre-constructed nonlinear function, the future stress value and the future wear amount need to be normalized. The normalization is achieved by calculating the ratio of the future stress value and the future wear amount. The ratio calculation formula is: Wherein: σR or MR represents the future stress value and the future wear amount, SR represents the reference amount of the stress value or the wear amount, and the reference amount is set manually by the technician;

[0080] The values ​​of the weight coefficient and the bias in the nonlinear function are obtained by optimizing by the least square method;

[0081] Specifically, the optimization method using the least squares method is as follows:

[0082] a1: defines the loss function used to optimize the nonlinear function, and randomly initializes the weight coefficients ω1, ω2 and bias b;

[0083] The specific expression of the loss function is as follows:

[0084]

[0085] Where: J(ω1,ω2,b) is the loss function, which represents the sum of the prediction errors of all sample points, P 停运，j represents the actual outage probability of the jth sample, and N is the number of samples;

[0086] a2: Calculate the gradient of the loss function with respect to the weight coefficients ω1, ω2 and bias b;

[0087] Among them, the calculation formula of the gradient of the weight coefficient ω1 is as follows:

[0088]

[0089] Among them, the calculation formula of the gradient of the weight coefficient ω2 is as follows:

[0090]

[0091] Among them, the calculation formula of the gradient of bias b is as follows:

[0092]

[0093] Where: represents the gradient of the weight coefficient ω1, represents the gradient of the weight coefficient ω2, represents the gradient of bias b;

[0094] a3: Use stochastic gradient descent method to update weight coefficients ω1, ω2 and bias b. The update formula is as follows:

[0095]

[0096] Where: t represents the number of current iterations, η is the learning rate, which controls the step size of each parameter update; the learning rate is determined based on experimental data;

[0097] a4: Repeat steps a2 and a3 above until the value of the loss function J(ω1,ω2,b) is less than or equal to the preset error threshold, or the predetermined number of iterations is reached, then the loop ends and the weight coefficient and bias value of the optimized nonlinear function are obtained;

[0098] The fault location module 150 is used to determine whether there is a faulty moving component of the wind turbine generator set at the future time R according to all outage probabilities. If not, T=T+v is set and the data acquisition module 110 is triggered; if yes, the unique identifier of the corresponding faulty moving component is fed back, where v is an integer greater than zero;

[0099] In implementation, the step of determining whether there is a faulty moving component in the wind turbine generator set at the future time R includes:

[0100] b1: extracting the i-th outage probability from all outage probabilities, where the number of outage probabilities is K, and i and K are integers greater than zero;

[0101] b2: Compare the ith outage probability with the preset outage probability threshold. If the ith outage probability is less than the outage probability threshold, it is determined that there is no faulty moving component in the wind turbine generator set at the future time R, and i=i+1 is set, and the process returns to step b1; if the ith outage probability is greater than or equal to the outage probability threshold, it is determined that there is a faulty moving component in the wind turbine generator set at the future time R, and the moving component corresponding to the ith outage probability is marked as a faulty moving component.

[0102] b3: repeating the above steps b1 to b2 until i=K, ending the loop and obtaining a judgment result, wherein the judgment result includes one of the following situations: there is a faulty moving part in the wind turbine generator set and there is no faulty moving part in the wind turbine generator set;

[0103] It should be noted that: multiple unique identifiers are pre-stored in the system database, each of which corresponds to a moving part, and the unique identifier includes but is not limited to a UUID device serial number or a MAC address, etc.;

[0104] By collecting and analyzing the wear and stress reflection characteristic data of each moving part of the wind turbine, it can not only accurately identify mechanical faults and provide intelligent fault diagnosis, but also optimize maintenance scheduling, avoid unnecessary shutdown and repairs, and improve the availability and economic benefits of the unit; this dynamic prediction helps to detect potential faults in advance, significantly reduce the risk of wind turbine shutdown, reduce maintenance costs, and ensure the operating efficiency and power generation benefits of the wind turbine.

[0105] Example 2

[0106] See also Figure 2 As shown, based on the above-mentioned embodiment 1, this embodiment is a further design improvement made around the fault diagnosis of a wind turbine generator set. This embodiment discloses a wind turbine generator set fault diagnosis system based on data analysis, and the system further includes:

[0107] The information acquisition module 210 is used to collect a vibration signal set of the faulty moving component within a predetermined time period when a faulty moving component is found, and generate an actual vibration waveform diagram of the faulty moving component according to the vibration signal set, wherein the vibration signal set includes P vibration values, where P is an integer greater than zero;

[0108] It should be noted that: the predetermined time duration is manually set by the technicians in advance, such as 1 minute, 3 minutes or 5 minutes, etc., and the actual vibration waveform diagram is formed with time as the horizontal axis and vibration value as the vertical axis;

[0109] An information retrieval module 220, for acquiring corresponding fault analysis data of a faulty moving component according to a mapping relationship between a unique identifier and the fault analysis data, wherein the fault analysis data includes a plurality of fault causes and a fault vibration waveform diagram corresponding to each fault cause;

[0110] It should be noted that: in the system database, there are pre-stored mapping relationships between unique identifiers and fault analysis data. In each mapping relationship, there is a unique identifier and the corresponding fault analysis data. Each unique identifier and the corresponding fault analysis data are manually associated and bound in advance by the technicians to form an actual mapping relationship.

[0111] A fault diagnosis module 230 is used to compare and analyze the actual vibration waveform diagram and the fault vibration waveform diagram to determine the fault cause of the faulty moving component;

[0112] In implementation, the comparison and analysis of the actual vibration waveform diagram and the fault vibration waveform diagram includes:

[0113] The similarity between the actual vibration waveform and each fault vibration waveform is calculated using a perceptual hash algorithm to obtain multiple similarities;

[0114] It should be understood that: Perceptual Hashing (pHash for short) is an algorithm used to compare the similarity of image, audio or video content; Unlike traditional hash algorithms (such as MD5, SHA-256, etc.), the goal of the perceptual hash algorithm is not to generate a unique identifier for data, but to create a "fingerprint" or "summary" to represent the perceptual characteristics of an image or other content; this fingerprint can effectively capture the main features of an image or signal, and can still maintain a high degree of similarity when the image undergoes small changes (such as scaling, rotation, color adjustment, slight noise, etc.);

[0115] Sort the multiple similarities in descending order, and use the fault cause corresponding to the first fault vibration waveform diagram in the similarity sorting as the fault cause of the faulty moving component;

[0116] By collecting the vibration signals of each moving part of the wind turbine, generating the actual vibration waveform diagram, and comparing and analyzing it with the fault vibration waveform diagram, and combining the perceptual hash algorithm to calculate the similarity, the cause of the fault can be identified more accurately, avoiding the misjudgment and omission of the traditional fault diagnosis method. At the same time, with the help of the system's automated diagnostic process and data analysis capabilities, the need for manual intervention is reduced, and through data-driven methods, the accuracy of fault diagnosis is improved, and the errors caused by human factors are reduced.

[0117] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters, weights and thresholds in the formula are set by technicians in this field according to actual conditions.

[0118] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center through a wired network or a wireless network. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD) or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0119] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0120] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0121] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only one, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0122] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0123] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0124] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0125] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A wind turbine fault diagnosis system based on data analysis, characterized in that: include: The data acquisition module is used to collect the wear reflection characteristic data and stress reflection characteristic data of each moving part in the wind turbine generator set within a predetermined time range from TN to T, where T and N are integers greater than zero; A first prediction module is used to input the wear reflection characteristic data of each moving component into a first recursive neural network model with a preset wear amount prediction, so as to predict the future wear amount of each moving component at a future time R, where R is an integer greater than zero; A second prediction module is used to input the stress reflection characteristic data of each moving component into a second recursive neural network model with a preset predicted stress value, so as to predict the future stress value of each moving component at a future time R; A probability analysis module, used for inputting the future stress value and the future wear amount of each moving component into a pre-built nonlinear function respectively, so as to obtain the shutdown probability of the wind turbine generator set under each moving component; The fault location module is used to determine whether there is a faulty moving component in the wind turbine generator set at the future time R based on all outage probabilities. If not, T=T+v is set and the data acquisition module is triggered; if so, the unique identifier of the corresponding faulty moving component is fed back, and v is an integer greater than zero.

2. The wind turbine generator fault diagnosis system based on data analysis according to claim 1, characterized in that: The wear reflection characteristic data includes operating state characteristic data and environmental characteristic data, and the stress reflection characteristic data includes operating condition characteristic data and environmental characteristic data, wherein the operating state characteristic data includes the cumulative operating time, operating speed, operating torque, wear amount, material thickness and lubricating oil amount at each moment; the operating condition characteristic data includes the cumulative operating time, stress value and material strength at each moment, and the environmental characteristic data includes ambient temperature, ambient humidity and ambient wind speed.

3. The wind turbine generator fault diagnosis system based on data analysis according to claim 2 is characterized in that: The training method of the first recursive neural network model is as follows: Acquire historical wear amount training data, and divide the historical wear amount training data into a wear amount training set and a wear amount test set, wherein the historical wear amount training data includes a plurality of wear reflection feature data and their corresponding future wear amounts; Constructing a first recursive neural network for performing a regression task, taking the wear reflection characteristic data in the wear amount training set as the input of the first recursive neural network, taking the future wear amount in the wear amount training set as the output of the first recursive neural network, training the first recursive neural network, and obtaining an initial first recursive neural network; The initial first recursive neural network is model verified using the wear test set, and an initial first recursive neural network that is less than or equal to a preset first test error threshold is output as the first recursive neural network model for executing wear prediction.

4. The wind turbine generator fault diagnosis system based on data analysis according to claim 3 is characterized in that: The training method of the second recursive neural network model is as follows: Acquire historical stress value training data, and divide the historical stress value training data into a stress value training set and a stress value test set, wherein the historical stress value training data includes a plurality of stress reflection characteristic data and corresponding future stress values; A second recursive neural network for performing a regression task is constructed, the stress reflection characteristic data in the stress value training set is used as the input of the second recursive neural network, and the future stress value in the stress value training set is used as the output of the second recursive neural network, and the second recursive neural network is trained to obtain an initial second recursive neural network; The initial second recursive neural network is model verified using the stress value test set, and an initial second recursive neural network that is less than or equal to a preset second test error threshold is output as the second recursive neural network model for executing stress value prediction.

5. The wind turbine generator fault diagnosis system based on data analysis according to claim 4, characterized in that: The expression of the pre-built nonlinear function is as follows: Where: P 停运 (R) represents the outage probability of the wind turbine, σ′R represents the normalized future stress value, M′R represents the normalized future wear value, ω1 and ω2 are weight coefficients greater than zero, and b is the bias.

6. The wind turbine generator fault diagnosis system based on data analysis according to claim 5, characterized in that: The values ​​of the weight coefficient and the bias in the nonlinear function are obtained by optimizing by the least square method; Among them, the optimization method using the least squares method is as follows: a1: defines the loss function used to optimize the nonlinear function, and randomly initializes the weight coefficients ω1, ω2 and bias b; The specific expression of the loss function is as follows: Where: J(ω1,ω2,b) is the loss function, which represents the sum of the prediction errors of all sample points, P 停运，j represents the actual outage probability of the jth sample, and N is the number of samples; a2: Calculate the gradient of the loss function with respect to the weight coefficients ω1, ω2 and bias b; Among them, the calculation formula of the gradient of the weight coefficient ω1 is as follows: Among them, the calculation formula of the gradient of the weight coefficient ω2 is as follows: Among them, the calculation formula of the gradient of bias b is as follows: Where: represents the gradient of the weight coefficient ω1, represents the gradient of the weight coefficient ω2, represents the gradient of bias b; a3: Use stochastic gradient descent method to update weight coefficients ω1, ω2 and bias b. The update formula is as follows: Where: t represents the number of current iterations, η is the learning rate; a4: Repeat steps a2 and a3 above until the value of the loss function J(ω1,ω2,b) is less than or equal to the preset error threshold, or reaches a predetermined number of iterations, then end the loop and obtain the values ​​of the weight coefficient and bias in the optimized nonlinear function.

7. The wind turbine generator fault diagnosis system based on data analysis according to claim 6, characterized in that: The determining whether there is a faulty moving part in the wind turbine generator set at the future time R includes: b1: extracting the i-th outage probability from all outage probabilities, where the number of outage probabilities is K, and i and K are integers greater than zero; b2: Compare the ith outage probability with the preset outage probability threshold. If the ith outage probability is less than the outage probability threshold, it is determined that there is no faulty moving component in the wind turbine generator set R at the future time, and i=i+1 is set, and the process returns to step b1. If the ith outage probability is greater than or equal to the outage probability threshold, it is determined that there is a faulty moving component in the wind turbine generator set R at the future time, and the moving component corresponding to the ith outage probability is marked as a faulty moving component. b3: Repeat the above steps b1 to b2 until i=K, then end the loop and obtain a judgment result, which includes one of the following situations: there is a faulty moving part in the wind turbine generator set and there is no faulty moving part in the wind turbine generator set.

8. The wind turbine generator fault diagnosis system based on data analysis according to claim 7, characterized in that: The system further comprises: An information acquisition module, for acquiring a vibration signal set of the faulty moving component within a predetermined time period when a faulty moving component is found, and generating an actual vibration waveform diagram of the faulty moving component according to the vibration signal set, wherein the vibration signal set includes P vibration values, where P is an integer greater than zero; An information retrieval module, used to obtain corresponding fault analysis data of the faulty moving component according to a mapping relationship between the unique identifier and the fault analysis data, wherein the fault analysis data includes multiple fault causes and a fault vibration waveform corresponding to each fault cause; The fault diagnosis module is used to compare and analyze the actual vibration waveform diagram and the fault vibration waveform diagram to determine the fault cause of the faulty moving part.

9. The wind turbine generator fault diagnosis system based on data analysis according to claim 8, characterized in that: The similarity between the actual vibration waveform and each fault vibration waveform is calculated using a perceptual hash algorithm to obtain multiple similarities; The multiple similarities are sorted from large to small according to the numerical values, and the fault cause corresponding to the first fault vibration waveform diagram in the similarity sorting is taken as the fault cause of the faulty moving component.

Images