Wind turbine operation and maintenance decision system based on artificial intelligence
Through the operation and maintenance decision-making system of wind turbine units based on artificial intelligence, multi-dimensional information fusion and convolutional neural network are used for fault diagnosis, and through digital twin model visualization, the problem of lack of intelligent fault diagnosis and data visualization in the existing technology is solved, and the problem of improving the Operation and maintenance efficiency and accuracy of wind turbine units.
Patent Information
- Application Number
- CN202411174795.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-26
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-08-26
AI Technical Summary
The existing wind turbine operation and maintenance technology lacks an intelligent fault diagnosis system that integrates multi-dimensional information, and for amateur personnel, the data interpretation is low, the maintenance process is complex, and it is difficult to comprehensively consider various maintenance parameters, resulting in low equipment fault detection and maintenance efficiency.
Adopt the operation and maintenance decision-making system of wind turbine units based on artificial intelligence, including perception modules, analysis modules, digital twin modules and operation and maintenance modules, and multi-dimensional information fusion is carried out through voiceprint recognition, vibration recognition, temperature recognition, current and voltage data to establish convolution Neural networks are fault-diagnosed and visually displayed through digital twin models.
It realizes intelligent fault diagnosis and operation and maintenance decision-making of wind turbine units, improves the accuracy and maintenance efficiency of fault detection, provides a more intuitive equipment status display, simplifies operation and maintenance processes, and reduces maintenance costs.
Smart Images

Figure CN118690313B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of wind turbine operation and maintenance, and in particular to a wind turbine operation and maintenance decision system based on artificial intelligence. Background Art
[0002] Wind turbines convert the kinetic energy of wind into mechanical energy of the wind rotor shaft through the rotation of the wind rotor under the action of wind force, which in turn drives the generator to rotate and generate electricity. Broadly speaking, wind energy is also a form of solar energy, so wind turbines can also be regarded as a heat energy generator that uses the sun as a heat source and the atmosphere as a working medium. Wind turbines are generally composed of wind rotors, generators, direction regulators, towers and other main parts.
[0003] The operation and maintenance technology of wind turbines involves many aspects, including daily inspections, regular maintenance, fault diagnosis and repair, etc. The primary purpose of its operation and maintenance work is to ensure that the generator can operate continuously and efficiently and convert wind energy into electrical energy. Through regular maintenance and inspection, potential problems can be discovered and solved in a timely manner to prevent the decline in power generation efficiency or shutdown due to equipment failure, thereby ensuring the stable power supply of the power grid. Secondly, by regularly replacing worn parts, cleaning blades and cooling systems, checking and tightening bolts, etc., it is possible to reduce downtime caused by equipment aging or damage, reduce maintenance costs, and improve the overall economy and reliability of the equipment.
[0004] However, the existing wind turbine operation and maintenance technology still has many deficiencies and limitations in many aspects. First, traditional wind turbine maintenance often only targets a single component or a single operating parameter, and there is no intelligent fault diagnosis system with multi-dimensional information fusion; second, traditional wind turbine maintenance technology requires relevant personnel to have a high level of technical skills and knowledge reserves, so that relevant personnel can comprehensively consider various maintenance parameters during the maintenance process and then judge the operating status of the wind turbine. For amateurs, the data during the operation process lacks interpretability, and the visualization of operation and maintenance related data is low.
[0005] Therefore, it is necessary to propose an artificial intelligence-based wind turbine operation and maintenance decision-making system to solve the above problems. Summary of the invention
[0006] The purpose of the present invention is to solve the problems existing in the background technology and propose a wind turbine operation and maintenance decision system based on artificial intelligence;
[0007] The purpose of the present invention can be achieved through the following technical solutions: an artificial intelligence-based wind turbine operation and maintenance decision-making system, including a perception module, an analysis module, a digital twin module and an operation and maintenance module.
[0008] The sensing module uniquely numbers each wind turbine, represented by a number symbol i, where i=1, 2, 3, ..., n; n is the total number of wind turbines.
[0009] Furthermore, the fan operating condition is sensed based on voiceprint recognition, vibration recognition and temperature recognition, specifically:
[0010] Based on voiceprint recognition, the audio signal generated by each wind turbine during operation is obtained through a high-fidelity microphone, the sound waveform-time diagram of the audio signal is recorded, and the analytical expression e1i(t) of the sound waveform changing with time is obtained.
[0011] Based on vibration recognition, the vibration signal generated by each motor during operation is obtained through a vibration sensor, the frequency-time diagram of the vibration signal is recorded, and the analytical expression e2i(t) of the frequency change over time is obtained.
[0012] Based on temperature identification, the motor core temperature of each wind turbine during operation is obtained through a temperature sensor, the temperature-time diagram of the motor core temperature is recorded, and the analytical expression e3i(t) of the motor temperature change over time is obtained.
[0013] As a preferred embodiment of the present invention, the sound waveform-time diagram, frequency-time diagram and temperature-time diagram are subjected to denoising by wavelet transform, specifically: the analytical expressions e1i(t), e2i(t) and e3i(t) are subjected to wavelet transform to extract the stable signal of the low-frequency part and suppress the high-frequency signal containing the environmental noise. The expression of the wavelet transform is:
[0014] Where d = 1, 2, 3; i is the number of each wind turbine, i = 1, 2, 3, ..., n; n is the total number of wind turbines; where a is the scale factor of the wavelet transform, b is the bias factor of the wavelet transform; where δ(tb / a) is the wavelet function, and the Haar wavelet function is used for waveform denoising. Its expression is:
[0015]
[0016] Obtain the noise-reduced sound waveform-time diagram E1i(t), vibration frequency-time diagram E2i(t) and temperature-time diagram E3i(t).
[0017] As a preferred embodiment of the present invention, wind turbine circuit fault perception is performed based on current recording and voltage recording, specifically: the operating current Ii and operating voltage Vi of each wind turbine are obtained at preset time intervals, wherein i is the wind turbine number symbol, i=1, 2, 3, ..., n; n is the total number of wind turbines. The acquired current and voltage data are respectively composed of current data points (t, Ii) and voltage data points (t, Vi). The current data points and voltage data points are classified according to the wind turbine number symbol i, and are saved and recorded as a data group according to each wind turbine number symbol. In each data group, a two-dimensional rectangular coordinate system is established with time t as the horizontal axis and the operating current Ii as the vertical axis, and the current data points (t, Ii) in the same data group are filled into the coordinate system and then all points are connected with a smooth curve to generate an operating current-time diagram. In each data group, a two-dimensional rectangular coordinate system is established with time t as the horizontal axis and the operating voltage Vi as the vertical axis, and the voltage data points (t, Vi) in the same data group are filled into the coordinate system and then all points are connected with a smooth curve to generate an operating voltage-time diagram.
[0018] The operating current-time diagram and the operating voltage-time diagram are converted from the time domain to the frequency domain by using fast Fourier transform, so as to obtain the current frequency spectrum diagram and the voltage frequency spectrum diagram of each wind turbine.
[0019] As a preferred embodiment of the present invention, the analysis module establishes a convolutional neural network to extract and analyze features of the sound waveform-time graph, vibration frequency-time graph and temperature-time graph after noise reduction processing, specifically:
[0020] A five-layer neural network is established, and a set of working condition labels are preset, including: normal operation, blade damage, blade deformation, blade icing, bearing wear, high bearing temperature, gear wear, gear breakage, gearbox noise, foundation settlement, tower corrosion, winding short circuit, cable damage, etc. Each working condition label corresponds to a preset abnormal coefficient F1, and the values of two weight factors C1 and C2.
[0021] The input of convolution layer 1 is: the noise-reduced sound waveform-time graph E1i(t), vibration frequency-time graph E2i(t) and temperature-time graph E3i(t); the output formula of convolution layer 1 is: Where d = 1, 2, 3; the output formula of convolutional layer 2 is in is the output value of convolutional layer 1; and They are the convolution kernels of convolution layer 1 and convolution layer 2 respectively; and are the bias coefficients of convolutional layer 1 and convolutional layer 2 respectively.
[0022] The full-link layer completes the abstract feature matrix extraction of the convolutional neural network through full-link calculation. The output formula of the full-link layer 1 is: in is the output value of convolutional layer 2; is the weight coefficient from the jth unit in convolutional layer 2 to the j1th unit in fully connected layer 1, where is the bias coefficient of the j1th unit in the fully connected layer 1;
[0023] The output formula of the fully connected layer 2 is: in is the output value of the fully connected layer 1; is the weight coefficient from the j1th unit in the fully connected layer 1 to the j2th unit in the fully connected layer 2, where is the bias coefficient of the j2th unit in the fully connected layer 2;
[0024] The output formula of the fully connected layer 3 is: in is the output value of the fully connected layer 2; is the weight coefficient from the j2th unit in the fully connected layer 2 to the j3th unit in the fully connected layer 3, where is the bias coefficient of the j3th unit in the fully connected layer;
[0025] The values of all weight coefficients and bias coefficients described above depend on the learning and training of the convolutional neural network. Let the output of the convolutional neural network be the abnormal coefficient F1i.
[0026] For normal operation of wind turbines, blade damage, blade deformation, blade icing, bearing wear, high bearing temperature, gear wear, gear breakage, gearbox noise, foundation settlement, tower corrosion, winding short circuit, cable damage and other working conditions, the acoustic waveform-time diagram, frequency-time diagram and temperature-time diagram under each working condition are collected as training data of the convolutional neural network, and the convolutional neural network is trained. The trained convolutional neural network and its convolution kernel, bias coefficient and weight coefficient are saved.
[0027] At every preset time interval t, the sound waveform-time diagram, vibration frequency-time diagram and temperature-time diagram of each wind turbine are input into the trained convolutional neural network to obtain the output abnormal coefficient F1i, and the corresponding working condition label is matched according to the value of the abnormal coefficient.
[0028] As a preferred embodiment of the present invention, the current frequency spectrum and voltage frequency spectrum recorded by the perception module are obtained. In the current frequency spectrum and voltage frequency spectrum, each peak image is called a frequency peak, the highest point of the frequency peak is called a frequency peak, and the frequency corresponding to the frequency peak is called a peak frequency. In the current frequency spectrum, the number of current frequency peaks n1 is obtained, and each current frequency peak is uniquely numbered, and the number symbol is k1, k1=1, 2, 3, ..., n1. The current frequency peak values A1, A2, A3, ..., An1 and the corresponding current peak frequencies f1, f2, f3, ..., fn1 are obtained. In the voltage frequency spectrum, the number of voltage frequency peaks n2 is obtained, and each voltage frequency peak is uniquely numbered, and the number symbol is k2, k2=1, 2, 3, ..., n2. The voltage frequency peak values B1, B2, B3, ..., Bn2 and the corresponding voltage peak frequencies V1, V2, V3, ..., Vn2 are obtained.
[0029] By formula Calculate the current abnormal coefficient F2i and voltage abnormal coefficient F3i of each wind turbine i. It is a set of preset standard current frequency peak values. is a set of preset standard current peak frequencies; It is a set of preset standard voltage frequency peak values. It is a set of preset standard voltage peak frequencies.
[0030] As a preferred embodiment of the present invention, by formula ψi = F1i C1×F2i+C2×F3i Calculate the alarm coefficient ψi of each wind turbine i, where F1i is the abnormal coefficient, F2i is the current abnormal coefficient, and F3i is the voltage abnormal coefficient; c1 and c2 are weight factors. If there is an alarm coefficient ψi greater than the preset threshold, the corresponding alarm signal is output. The alarm signal contains the time t and the unique number i of the wind turbine.
[0031] As a preferred embodiment of the present invention, the digital twin module obtains the topographic map where the wind turbine is located and the position coordinates of each wind turbine, generates a virtual model of the wind turbine on the topographic map according to the position coordinates, and displays the sound waveform-time graph, frequency-time graph, temperature-time graph, current frequency spectrum graph and voltage frequency spectrum graph obtained by the perception module on the virtual model, thereby generating a digital twin model of the wind turbine. In the digital twin model of the wind turbine, the wind turbine that generates the alarm signal is highlighted.
[0032] When the operation and maintenance module senses the generation of an alarm signal n, it stops the operation of the nth wind turbine and informs the maintenance personnel through the display screen.
[0033] Compared with the prior art, the present invention has the following beneficial effects:
[0034] (1) The present invention constructs a convolutional neural network to conduct a comprehensive analysis of the soundprint, vibration and temperature of the wind turbine, and establishes an intelligent fault diagnosis engine based on multi-dimensional information fusion and voiceprint technology. The diagnosis of the presence and severity of the fault is made according to the detection data, which can assist the operation and maintenance personnel in formulating the operation and maintenance plan;
[0035] (2) The present invention visualizes the operating status of the wind turbine through a digital twin model, establishes a three-dimensional model of the equipment structure, and projects the status of the physical equipment on the digital twin model after real-time analysis, thereby achieving a more intuitive and vivid display effect;
[0036] (3) The present invention comprehensively considers the voltage, current, sound waves, vibration and temperature data of the wind turbine, and through data fusion and data noise reduction, specifically determines whether to generate an alarm signal, thereby improving the efficiency of data processing and the efficient implementation of wind turbine maintenance. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.
[0038] Figure 1 is a system block diagram of the present invention;
[0039] Figure 2 A topological diagram of a five-layer neural network structure of the present invention;
[0040] Figure 3 Schematic diagram of the digital twin model of the present invention. DETAILED DESCRIPTION
[0041] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. 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.
[0042] See also Figure 1 As shown in the figure, the wind turbine operation and maintenance decision-making system based on artificial intelligence includes a perception module, an analysis module, a digital twin module and an operation and maintenance module.
[0043] The sensing module uniquely numbers each wind turbine, represented by a number symbol i, where i=1, 2, 3, ..., n; n is the total number of wind turbines.
[0044] Furthermore, the fan operating condition is sensed based on voiceprint recognition, vibration recognition and temperature recognition, specifically:
[0045] Based on voiceprint recognition, the audio signal generated by each wind turbine during operation is obtained through a high-fidelity microphone, the sound waveform-time diagram of the audio signal is recorded, and the analytical expression e1i(t) of the sound waveform changing with time is obtained.
[0046] Based on vibration recognition, the vibration signal generated by each motor during operation is obtained through a vibration sensor, the frequency-time diagram of the vibration signal is recorded, and the analytical expression e2i(t) of the frequency change over time is obtained.
[0047] Based on temperature identification, the motor core temperature of each wind turbine during operation is obtained through a temperature sensor, the temperature-time diagram of the motor core temperature is recorded, and the analytical expression e3i(t) of the motor temperature change over time is obtained.
[0048] Furthermore, the sound waveform-time graph, frequency-time graph and temperature-time graph are subjected to noise reduction processing by wavelet transform, specifically: wavelet transform is performed on the analytical expressions e1i(t), e2i(t) and e3i(t), the stable signal of the low-frequency part is extracted, and the high-frequency signal containing environmental noise is suppressed. The expression of the wavelet transform is:
[0049] Where d = 1, 2, 3; i is the number of each wind turbine, i = 1, 2, 3, ..., n; n is the total number of wind turbines; where a is the scale factor of the wavelet transform, b is the bias factor of the wavelet transform; where δ(tb / a) is the wavelet function, and the Haar wavelet function is used for waveform denoising. Its expression is:
[0050]
[0051] Obtain the noise-reduced sound waveform-time diagram E1i(t), vibration frequency-time diagram E2i(t) and temperature-time diagram E3i(t).
[0052] It should be noted that the operating environment of wind turbines is complex, and a large amount of data noise is contained in the acoustic waveform-time diagram, frequency-time diagram and temperature-time diagram. The filtering realized by wavelet transform can effectively filter out noise and interference signals by selectively transmitting or suppressing signals of specific frequencies, extracting useful information components from complex signals, providing a basis for subsequent signal processing and analysis, and improving the signal-to-noise ratio and reliability of the signal.
[0053] Further, based on the current record and the voltage record, the wind turbine circuit fault perception is performed, specifically: the operating current Ii and the operating voltage Vi of each wind turbine are obtained at every preset time interval, wherein i is the wind turbine number symbol, i=1, 2, 3, ..., n; n is the total number of wind turbines. The acquired current and voltage data are respectively composed of current data points (t, Ii) and voltage data points (t, Vi). The current data points and voltage data points are classified according to the wind turbine number symbol i, and are saved and recorded as a data group according to each wind turbine number symbol. In each data group, a two-dimensional rectangular coordinate system is established with time t as the horizontal axis and the operating current Ii as the vertical axis, and the current data points (t, Ii) in the same data group are filled into the coordinate system and then all points are connected with a smooth curve to generate an operating current-time diagram. In each data group, a two-dimensional rectangular coordinate system is established with time t as the horizontal axis and the operating voltage Vi as the vertical axis, and the voltage data points (t, Vi) in the same data group are filled into the coordinate system and then all points are connected with a smooth curve to generate an operating voltage-time diagram.
[0054] The operating current-time diagram and the operating voltage-time diagram are converted from the time domain to the frequency domain by using fast Fourier transform, so as to obtain the current frequency spectrum diagram and the voltage frequency spectrum diagram of each wind turbine.
[0055] The analysis module establishes a convolutional neural network to extract and analyze the features of the noise-reduced sound waveform-time graph, vibration frequency-time graph, and temperature-time graph. Specifically:
[0056] See also Figure 2 As shown, a five-layer neural network is established
[0057] A set of operating condition labels are preset, including: normal operation, blade damage, blade deformation, blade icing, bearing wear, high bearing temperature, gear wear, gear breakage, gearbox noise, foundation settlement, tower corrosion, winding short circuit, cable damage, etc. Each operating condition label corresponds to a preset abnormal coefficient F1, and the values of two weight factors C1 and C2.
[0058] The input of convolution layer 1 is: the noise-reduced sound waveform-time graph E1i(t), vibration frequency-time graph E2i(t) and temperature-time graph E3i(t); the output formula of convolution layer 1 is: Where d = 1, 2, 3; the output formula of convolutional layer 2 is in is the output value of convolutional layer 1; and They are the convolution kernels of convolution layer 1 and convolution layer 2 respectively; and are the bias coefficients of convolutional layer 1 and convolutional layer 2 respectively.
[0059] The full-link layer completes the abstract feature matrix extraction of the convolutional neural network through full-link calculation. The output formula of the full-link layer 1 is: in is the output value of convolutional layer 2; is the weight coefficient from the jth unit in convolutional layer 2 to the j1th unit in fully connected layer 1, where is the bias coefficient of the j1th unit in the fully connected layer 1;
[0060] The output formula of the fully connected layer 2 is: in is the output value of the fully connected layer 1; is the weight coefficient from the j1th unit in the fully connected layer 1 to the j2th unit in the fully connected layer 2, where is the bias coefficient of the j2th unit in the fully connected layer 2;
[0061] The output formula of the fully connected layer 3 is: in is the output value of the fully connected layer 2; is the weight coefficient from the j2th unit in the fully connected layer 2 to the j3th unit in the fully connected layer 3, where is the bias coefficient of the j3th unit in the fully connected layer;
[0062] The values of all weight coefficients and bias coefficients described above depend on the learning and training of the convolutional neural network. Let the output of the convolutional neural network be the abnormal coefficient F1i.
[0063] For normal operation of wind turbines, blade damage, blade deformation, blade icing, bearing wear, high bearing temperature, gear wear, gear breakage, gearbox noise, foundation settlement, tower corrosion, winding short circuit, cable damage and other working conditions, the acoustic waveform-time diagram, frequency-time diagram and temperature-time diagram under each working condition are collected as training data of the convolutional neural network, and the convolutional neural network is trained. The trained convolutional neural network and its convolution kernel, bias coefficient and weight coefficient are saved.
[0064] At every preset time interval t, the sound waveform-time diagram, vibration frequency-time diagram and temperature-time diagram of each wind turbine are input into the trained convolutional neural network to obtain the output abnormal coefficient F1i, and the corresponding working condition label is matched according to the value of the abnormal coefficient.
[0065] Further, the current frequency spectrum and voltage frequency spectrum recorded by the perception module are obtained. In the current frequency spectrum and voltage frequency spectrum, each peak image is called a frequency peak, the highest point of the frequency peak is called a frequency peak, and the frequency corresponding to the frequency peak is called a peak frequency. In the current frequency spectrum, the number of current frequency peaks n1 is obtained, and each current frequency peak is uniquely numbered, and the number symbol is k1, k1 = 1, 2, 3, ..., n1. The current frequency peaks A1, A2, A3, ..., An1 and the corresponding current peak frequencies f1, f2, f3, ..., fn1 are obtained. In the voltage frequency spectrum, the number of voltage frequency peaks n2 is obtained, and each voltage frequency peak is uniquely numbered, and the number symbol is k2, k2 = 1, 2, 3, ..., n2. The voltage frequency peaks B1, B2, B3, ..., Bn2 and the corresponding voltage peak frequencies V1, V2, V3, ..., Vn2 are obtained.
[0066] By formula Calculate the current abnormal coefficient F2i and voltage abnormal coefficient F3i of each wind turbine i. It is a set of preset standard current frequency peak values. is a set of preset standard current peak frequencies; It is a set of preset standard voltage frequency peak values. It is a set of preset standard voltage peak frequencies.
[0067] Furthermore, through the formula ψi = F1i C1×F2i+C2×F3i Calculate the alarm coefficient ψi of each wind turbine i, where F1i is the abnormal coefficient, F2i is the current abnormal coefficient, and F3i is the voltage abnormal coefficient; c1 and c2 are weight factors. If there is an alarm coefficient ψi greater than the preset threshold, the corresponding alarm signal is output. The alarm signal contains the time t and the unique number i of the wind turbine.
[0068] See also Figure 3 As shown, the digital twin module obtains the topographic map where the wind turbine is located and the position coordinates of each wind turbine, generates a virtual model of the wind turbine on the topographic map according to the position coordinates, and displays the sound waveform-time graph, frequency-time graph, temperature-time graph, current frequency spectrum graph and voltage frequency spectrum graph obtained by the perception module on the virtual model, thereby generating a digital twin model of the wind turbine. In the digital twin model of the wind turbine, the wind turbine that generates the alarm signal is highlighted.
[0069] When the operation and maintenance module senses the generation of an alarm signal n, it stops the operation of the nth wind turbine and informs the maintenance personnel through the display screen.
[0070] It should be understood that the terms "include" and "comprising" used in the specification and claims of the present disclosure indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.
[0071] It should also be understood that the terms used in this disclosure are only for the purpose of describing specific embodiments and are not intended to limit the disclosure. As used in this disclosure and the claims, the singular forms "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise. It should also be further understood that the term "and / or" used in this disclosure and the claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes these combinations;
[0072] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.
Claims
1. Wind turbine operation and maintenance decision-making system based on artificial intelligence, including perception module and analysis module, Features: The sensing module uniquely numbers each wind turbine; Based on voiceprint recognition, vibration recognition and temperature recognition, the fan operating condition is sensed to obtain the noise-reduced sound waveform-time diagram E1i(t), vibration frequency-time diagram E2i(t) and temperature-time diagram E3i(t); Based on the current and voltage records, the wind turbine circuit fault is sensed to obtain the current frequency spectrum and voltage frequency spectrum of each wind turbine; The analysis module presets a set of working condition labels, including normal operation, blade damage, blade deformation, blade icing, bearing wear, bearing overtemperature, gear wear, gear breakage, gearbox noise, foundation settlement, tower corrosion, winding short circuit and cable damage; each working condition label corresponds to a preset abnormal coefficient F1, as well as the values of two weight factors C1 and C2; a five-layer convolutional neural network is established to extract and analyze the features of the noise-reduced sound waveform-time graph, vibration frequency-time graph and temperature-time graph, and match them to the corresponding abnormal coefficient F1i: The current frequency spectrum and the voltage frequency spectrum recorded by the sensing module are obtained. In the current frequency spectrum and the voltage frequency spectrum, each peak image is called a frequency peak, the highest point of the frequency peak is called a frequency peak, and the frequency corresponding to the frequency peak is called a peak frequency; in the current frequency spectrum, the number of current frequency peaks n1 is obtained, and each current frequency peak is uniquely numbered, and the numbering symbol is k1, k1=1, 2, 3, ..., n1; the current frequency peak values A1, A2, A3, ..., An1 and the corresponding current peak frequencies f1, f2, f3, ... , fn1; in the voltage frequency spectrum, obtain the number of voltage frequency peaks n2, and uniquely number each voltage frequency peak, the numbering symbol is k2, k2 = 1, 2, 3, ..., n2; obtain the voltage frequency peak values B1, B2, B3, ..., Bn2 and the corresponding voltage peak frequencies V1, V2, V3, ..., Vn2; obtain the current abnormality coefficient F2i and the voltage abnormality coefficient F3i of each wind turbine i through calculation, and comprehensively consider the abnormality coefficient F1i, the current abnormality coefficient F2i and the voltage abnormality coefficient F3i to determine whether to generate an alarm signal.
2. According to the artificial intelligence-based wind turbine operation and maintenance decision system of claim 1, It is characterized in that It also includes a digital twin module and an operation and maintenance module; The digital twin module obtains the topographic map where the wind turbine is located and the position coordinates of each wind turbine, generates a virtual model of the wind turbine on the topographic map according to the position coordinates, and displays the sound waveform-time graph, frequency-time graph, temperature-time graph, current frequency spectrum graph and voltage frequency spectrum graph obtained by the perception module on the virtual model, thereby generating a digital twin model of the wind turbine; in the digital twin model of the wind turbine, the wind turbine generating the alarm signal is highlighted; When the operation and maintenance module senses the generation of an alarm signal n, it stops the operation of the nth wind turbine and informs the maintenance personnel through the display screen.
3. According to the artificial intelligence-based wind turbine operation and maintenance decision-making system of claim 1, It is characterized in that The specific process of sensing the fan operating condition based on voiceprint recognition, vibration recognition and temperature recognition is as follows: Based on voiceprint recognition, the audio signal generated by each wind turbine during operation is obtained through a high-fidelity microphone, the sound waveform-time diagram of the audio signal is recorded, and the analytical expression e1i(t) of the sound waveform changing with time is obtained; Based on vibration recognition, the vibration signal generated by each motor during operation is obtained through a vibration sensor, the frequency-time diagram of the vibration signal is recorded, and the analytical expression e2i(t) of the frequency change over time is obtained; Based on temperature identification, the core temperature of each wind turbine during operation is obtained through a temperature sensor, the temperature-time diagram of the core temperature of the motor is recorded, and the analytical expression e3i(t) of the change of the motor temperature over time is obtained; The sound waveform-time graph, frequency-time graph and temperature-time graph are subjected to noise reduction processing by wavelet transform, specifically: wavelet transform is performed on the analytical expressions e1i(t), e2i(t) and e3i(t), the stable signal of the low-frequency part is extracted, and the high-frequency signal containing the environmental noise is suppressed. The expression of the wavelet transform is: Where d = 1, 2, 3; i is the number of each wind turbine, i = 1, 2, 3, ..., n; n is the total number of wind turbines; where a is the scale factor of the wavelet transform, and b is the bias factor of the wavelet transform; where δ(tb / a) is the wavelet function, and the Haar wavelet function is used for waveform denoising; its expression is: Obtain the noise-reduced sound waveform-time diagram E1i(t), vibration frequency-time diagram E2i(t) and temperature-time diagram E3i(t).
4. According to the artificial intelligence-based wind turbine operation and maintenance decision system of claim 1, It is characterized in that The specific process of fan circuit fault perception based on current and voltage records is as follows: The operating current Ii and the operating voltage Vi of each wind turbine are obtained at preset time intervals; the obtained current and voltage data are respectively formed into current data points (t, Ii) and voltage data points (t, Vi); the current data points and voltage data points are classified according to the wind turbine number symbol i, and are saved according to each wind turbine number symbol and recorded as a data group; in each data group, a two-dimensional rectangular coordinate system is established with time t as the horizontal axis and the operating current Ii as the vertical axis, the current data points (t, Ii) in the same data group are filled into the coordinate system, and then all points are connected with a smooth curve to generate an operating current-time diagram; In each data group, a two-dimensional rectangular coordinate system is established with time t as the horizontal axis and the operating voltage Vi as the vertical axis. The voltage data points (t, Vi) in the same data group are filled into the coordinate system and then all points are connected with a smooth curve to generate an operating voltage-time graph; The operating current-time diagram and the operating voltage-time diagram are converted from the time domain to the frequency domain by using fast Fourier transform, so as to obtain the current frequency spectrum diagram and the voltage frequency spectrum diagram of each wind turbine.
5. According to the artificial intelligence-based wind turbine operation and maintenance decision system of claim 1, It is characterized in that The five-layer convolutional neural network is specifically: The input of convolution layer 1 is: the noise-reduced sound waveform-time graph E1i(t), vibration frequency-time graph E2i(t) and temperature-time graph E3i(t); the output formula of convolution layer 1 is: Where d = 1, 2, 3; the output formula of convolutional layer 2 is in is the output value of convolutional layer 1; and They are the convolution kernels of convolution layer 1 and convolution layer 2 respectively; and are the bias coefficients of convolutional layer 1 and convolutional layer 2 respectively; The full-link layer completes the abstract feature matrix extraction of the convolutional neural network through full-link calculation. The output formula of the full-link layer 1 is: in is the output value of convolutional layer 2; is the weight coefficient from the jth unit in convolutional layer 2 to the j1th unit in fully connected layer 1, where is the bias coefficient of the j1th unit in the fully connected layer 1; The output formula of the fully connected layer 2 is: in is the output value of the fully connected layer 1; is the weight coefficient from the j1th unit in the fully connected layer 1 to the j2th unit in the fully connected layer 2, where is the bias coefficient of the j2th unit in the fully connected layer 2; The output formula of the fully connected layer 3 is: in is the output value of the fully connected layer 2; is the weight coefficient from the j2th unit in the fully connected layer 2 to the j3th unit in the fully connected layer 3, where is the bias coefficient of the j3th unit in the fully connected layer; The values of all the weight coefficients and bias coefficients described above depend on the learning and training of the convolutional neural network; let the output of the convolutional neural network be the abnormal coefficient F1i; For normal operation, blade damage, blade deformation, blade icing, bearing wear, excessive bearing temperature, gear wear, gear breakage, gear box noise, foundation settlement, tower corrosion, winding short circuit and cable damage of the wind turbine, the sound waveform-time diagram, frequency-time diagram and temperature-time diagram under each working condition are collected as training data of the convolutional neural network, and the convolutional neural network is trained; Save the trained convolutional neural network and its convolution kernel, bias coefficient and weight coefficient; At every preset time interval t, the sound waveform-time diagram, vibration frequency-time diagram and temperature-time diagram of each wind turbine are input into the trained convolutional neural network to obtain the output abnormal coefficient F1i, and the corresponding working condition label is matched according to the value of the abnormal coefficient.
6. The wind turbine operation and maintenance decision system based on artificial intelligence according to claim 1, It is characterized in that The specific process of obtaining the current abnormal coefficient F2i and voltage abnormal coefficient F3i of each wind turbine i through calculation is as follows: By formula Calculate the current abnormal coefficient F2i and voltage abnormal coefficient F3i of each wind turbine i; It is a set of preset standard current frequency peak values. is a set of preset standard current peak frequencies; It is a set of preset standard voltage frequency peak values. It is a set of preset standard voltage peak frequencies; the abnormal coefficient F1i, the current abnormal coefficient F2i and the voltage abnormal coefficient F3i are comprehensively considered to determine whether to generate an alarm signal.
7. According to the artificial intelligence-based wind turbine operation and maintenance decision system of claim 6, It is characterized in that The specific process of comprehensively considering the abnormal coefficient F1i, the current abnormal coefficient F2i and the voltage abnormal coefficient F3i to determine whether to generate an alarm signal is as follows: Through the formula ψi = F1i C1×F2i+C2×F3i Calculate the alarm coefficient ψi of each wind turbine i, where F1i is the abnormal coefficient, F2i is the current abnormal coefficient, and F3i is the voltage abnormal coefficient; c1 and c2 are weight factors; if there is an alarm coefficient ψi greater than a preset threshold, output a corresponding alarm signal; the alarm signal includes the time t and the unique number i of the wind turbine.
Citation Information
Patent Citations
Offshore wind plant operation and maintenance method and system based on digital twinning
CN115935737A
Equipment fault diagnosis method and system based on deep neural network
CN117708542A
Preventive maintenance and fault diagnosis system and method for offshore wind turbine generator
CN118462501A
Method and system for evaluating health state of power distribution equipment
CN118503636A
System, method and computer-readable medium for evaluating structural integrity of a gradient coil
US20220146608A1
Cited By
Power source anomaly detection method and system
TWI906067B
