Fan blade icing degree prediction method and system based on redundant sound sensor

Abstract

The invention discloses a fan blade icing degree prediction method and system based on a redundant sound sensor. The method comprises the following steps: inputting meteorological data and sound data; noise reduction and filtering are carried out on the input data; adopting an AI algorithm to extract meteorological characteristic information from the meteorological data, and preliminarily judging basic conditions of blade icing; the AI algorithm is adopted to extract sound feature information from the sound data, meteorological feature data information is combined, the blade icing degree is comprehensively judged, and finally graded early warning information is output. Sound signals at the 1 / 3 position of the blade tip are collected, and the icing degree of the fan blade can be accurately predicted.

Description

Technical Field

[0001] The present invention relates to the field of icing prediction, and in particular to a method and system for predicting the icing degree of wind turbine blades based on redundant sound sensors. Background Art

[0002] Wind turbine blades are an important component of wind turbines. Their working environment is complex and the failure rate is high. Blades operating in an ice-covered state can easily cause major safety accidents and economic losses. Currently, there is still no mature means to predict the degree of blade icing. It is of great significance to carry out research on the method of detecting the degree of icing on wind turbine blades.

[0003] The efficient and stable operation of wind turbines has always been affected by various environmental factors, with blade icing being a particularly prominent issue. Wind turbine blades are highly susceptible to ice formation in low-temperature, high-humidity environments. Ice on the blade surface not only increases aerodynamic drag and reduces wind energy capture efficiency, but also causes blade imbalance, vibration, and even structural fatigue and damage. Prolonged ice accumulation severely impacts the mechanical properties of blades, potentially shortening their service life and increasing maintenance costs. Furthermore, abnormal vibration and load variations caused by ice accumulation can cause secondary damage to other critical components of the wind turbine, leading to more complex failures. Predicting wind turbine blade icing not only helps improve wind power system efficiency but also provides early warning to prevent catastrophic failures. Traditional maintenance methods rely primarily on regular inspections or manual patrols. However, ice accumulation is often sudden and uneven, making these methods inadequate for real-time and accurate detection. Predictive methods based on real-time data can detect early signs of ice accumulation, enabling timely de-icing measures or adjustments to operating strategies to ensure stable wind turbine operation and reduce the risk of accidents. Annual power generation losses in areas with harsh conditions can range from 20% to 50%. Developing high-precision, real-time icing prediction technology has become a key technical challenge for the wind power industry. Methods for predicting wind turbine blade icing fall into three main categories: those based on SCADA data, those based on meteorological data, and those based on numerical simulation and experimental data. None of these methods has effectively addressed the issue of predicting wind blade icing. Summary of the Invention

[0004] The present invention aims to solve one of the technical problems in the related art at least to a certain extent.

[0005] During wind turbine blade operation, acoustic waves generated by aerodynamic noise, structural vibration, and local damage can be captured by high-precision sensors. By analyzing the acoustic signals in the time, frequency, and time-frequency domains, fault signatures hidden within the signals can be uncovered. The application of acoustic methods to blade icing monitoring is still in the exploratory stage, but theoretical foundations and preliminary experimental results indicate that icing alters the vibration and acoustic properties of the blade surface, thereby affecting the acoustic signals captured by sensors. By leveraging these changes, signal processing and feature extraction techniques can be used to indirectly monitor the degree of icing. To address the shortcomings of existing wind turbine blade fault monitoring methods and their accuracy, a method for predicting the degree of icing on wind turbine blades based on redundant acoustic spectra was developed. The main technical issues addressed are as follows: To address the problem of insufficient blade sound discrimination in acoustic signal feature extraction methods, a method for predicting the degree of icing on wind turbine blades based on redundant acoustic sensors was proposed. This method collects acoustic signals from the tip 1 / 3 of the blade to accurately predict the degree of icing on the wind turbine blades.

[0006] Another object of the present invention is to provide a wind turbine blade icing degree prediction system based on redundant sound sensors.

[0007] To achieve the above objectives, the present invention provides a method for predicting the degree of icing on wind turbine blades based on redundant acoustic sensors, comprising:

[0008] Acquire meteorological data related to ice coating on wind turbine blades and sound data generated by blade rotation respectively;

[0009] Performing noise reduction and filtering operations on the meteorological data and the sound data to obtain corresponding processed meteorological data and processed sound data;

[0010] Extracting meteorological characteristic information from the processed meteorological data using a preset first AI algorithm to preliminarily determine basic conditions for blade icing based on the meteorological characteristic information;

[0011] The preset second AI algorithm is used to extract sound feature information from the processed sound data, and the degree of blade icing is comprehensively predicted based on the sound feature information and meteorological feature information, and finally graded warning information is output.

[0012] The method for predicting the degree of icing on wind turbine blades based on redundant acoustic sensors according to an embodiment of the present invention may also have the following additional technical features:

[0013] In one embodiment of the present invention, obtaining sound data generated by blade rotation includes:

[0014] Install a sound sensor at the 1 / 3 of the blade tip on the upper part of the wind turbine tower;

[0015] Three sound sensors are arranged 120 degrees apart to form an equilateral triangle or circular symmetrical layout;

[0016] When the wind turbine blades rotate, the Doppler effect generated by the spatial position difference between the tip and the 1 / 3 of the blade tip is used to collect the sound signals generated by the blade rotation. By analyzing the frequency characteristics of the captured sound signals, the Doppler frequency shift signals related to blade icing are screened out.

[0017] In one embodiment of the present invention, obtaining meteorological data related to wind turbine blade icing includes:

[0018] Obtaining wind farm weather station parameters, wherein the wind farm weather station parameters include multiple ones of wind speed, wind direction, temperature, humidity, air pressure, precipitation, visibility, cloud cover, and sunshine hours;

[0019] Meteorological data related to icing on wind turbine blades is obtained based on the wind farm meteorological station parameters.

[0020] In one embodiment of the present invention, performing noise reduction and filtering operations on meteorological data and sound data to obtain corresponding processed meteorological data and processed sound data includes:

[0021] Utilize statistical methods or sliding window technology to detect and eliminate outliers in meteorological data, remove high-frequency noise, and perform linear interpolation or spline interpolation on missing data to obtain processed meteorological data;

[0022] The signal is decomposed using wavelet transform to remove high-frequency noise components in the sound data and then reconstructed. The median filter is applied to remove impulse noise. A band-stop filter is designed to suppress background noise in a preset frequency band, and a time-varying notch window is designed using the Doppler shift characteristics to retain the narrowband modulated signal generated by the rotation of the fan blades to obtain the processed sound data.

[0023] In one embodiment of the present invention, the method further includes:

[0024] analyzing the processed Doppler shift signal to determine a fundamental frequency component;

[0025] According to the correlation between the fundamental frequency component and the blade rotation speed, the blade rotation speed is calculated using the Doppler frequency shift formula;

[0026] The wind turbine power is calculated based on the blade speed and wind farm meteorological station parameters and using the wind turbine power curve or mathematical model.

[0027] In one embodiment of the present invention, the first AI algorithm includes a first recurrent neural network model; extracting meteorological feature information from the processed meteorological data using the preset first AI algorithm to preliminarily determine basic conditions for blade icing based on the meteorological feature information includes:

[0028] The trained first recurrent neural network model is used to perform feature extraction on the preprocessed meteorological data to extract meteorological feature information related to blade icing;

[0029] Based on the extracted meteorological characteristic information, a preliminary judgment is made on the basic conditions for blade icing and a preliminary judgment result is output; if the basic conditions for icing are met, the next step is to enter the sound feature analysis and comprehensive judgment.

[0030] In one embodiment of the present invention, the first AI algorithm includes a second recurrent neural network model; the preset second AI algorithm is used to extract sound feature information from the processed sound data, and the degree of blade icing is comprehensively predicted based on the sound feature information and meteorological feature information, and finally, graded warning information is output, including:

[0031] The trained second recurrent neural network model is used to extract features from the preprocessed sound data, and the extracted sound feature information and meteorological feature information are fused to obtain a fused feature vector;

[0032] The fused feature vector is processed using the trained second recurrent neural network model to establish a mapping relationship between sound features, meteorological features and blade icing degree to predict the blade icing degree.

[0033] Set graded warning rules based on the predicted degree of icing, and output corresponding graded warning information based on the prediction results.

[0034] To achieve the above-mentioned object, the present invention further provides a wind turbine blade icing degree prediction system based on redundant acoustic sensors, comprising:

[0035] A raw data acquisition module, used to respectively acquire meteorological data related to ice coating on wind turbine blades and sound data generated by blade rotation;

[0036] A data preprocessing module is used to perform noise reduction and filtering operations on meteorological data and sound data to obtain corresponding processed meteorological data and processed sound data;

[0037] a basic condition judgment module, configured to extract meteorological characteristic information from the processed meteorological data using a preset first AI algorithm, so as to preliminarily judge the basic conditions for blade icing based on the meteorological characteristic information;

[0038] The icing degree prediction module is used to extract sound feature information from the processed sound data using a preset second AI algorithm, and comprehensively predict the blade icing degree based on the sound feature information and meteorological feature information, and finally output graded warning information.

[0039] The wind turbine blade icing degree prediction method and system based on redundant sound sensors in the embodiments of the present invention have significant advantages in coverage, positioning accuracy, reliability and engineering implementation through the layout of three redundant sensors with a mutual difference of 120 degrees.

[0040] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0042] Figure 1 is a flow chart of a method for predicting the degree of icing on wind turbine blades based on redundant sound sensors according to an embodiment of the present invention;

[0043] Figure 2 is an architectural diagram of a method for predicting icing degree of wind turbine blades based on redundant sound sensors according to an embodiment of the present invention;

[0044] Figure 3 2. It is a schematic diagram of measuring sound using a Doppler frequency shift algorithm according to an embodiment of the present invention;

[0045] Figure 4 is a schematic diagram of a wind turbine tower according to an embodiment of the present invention;

[0046] Figure 5 is a diagram of a redundant sound sensor arrangement according to an embodiment of the present invention;

[0047] Figure 6 is a wind turbine blade icing degree warning diagram according to an embodiment of the present invention;

[0048] Figure 7 4 is a structural diagram of a wind turbine blade icing degree prediction system based on redundant sound sensors according to an embodiment of the present invention. DETAILED DESCRIPTION

[0049] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments of the present invention can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0050] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described 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 making creative efforts should fall within the scope of protection of the present invention.

[0051] The following describes a method and system for predicting the degree of icing on wind turbine blades based on redundant acoustic sensors according to an embodiment of the present invention with reference to the accompanying drawings.

[0052] Figure 1 FIG. 1 is a flow chart of a method for predicting the degree of icing on wind turbine blades based on redundant sound sensors according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0053] S1, respectively acquiring meteorological data related to ice coating on wind turbine blades and sound data generated by blade rotation;

[0054] S2, performing noise reduction and filtering operations on the meteorological data and the sound data to obtain corresponding processed meteorological data and processed sound data;

[0055] S3, extracting meteorological characteristic information from the processed meteorological data using a preset first AI algorithm, and preliminarily determining basic conditions for blade icing based on the meteorological characteristic information;

[0056] S4, uses the preset second AI algorithm to extract sound feature information from the processed sound data, and comprehensively predicts the degree of blade icing based on the sound feature information and meteorological feature information, and finally outputs graded warning information.

[0057] Figure 2 This is the architecture diagram of the redundant acoustic sensor-based wind turbine blade icing prediction method. First, meteorological data and acoustic data are input; the input data is then subjected to noise reduction and filtering; an AI algorithm is used to extract meteorological characteristics from the meteorological data to preliminarily determine the basic conditions for blade icing; an AI algorithm is used to extract acoustic characteristics from the acoustic data, and combined with the meteorological characteristics, a comprehensive determination of the blade icing degree is made, and finally, a graded warning message is output. The three redundant acoustic sensors of the present invention are installed 120 degrees apart (in an equilateral triangle or circular symmetrical layout), offering numerous advantages in applications such as acoustic monitoring, positioning, and noise analysis.

[0058] In one embodiment of the present invention, Figure 3 The schematic diagram of Doppler frequency shift algorithm for measuring sound is shown as follows: Figure 3 As shown, 1 is the fan blade and 2 is the sound sensor.

[0059] In one embodiment of the present invention, Figure 4 is a schematic diagram of the wind turbine tower, as shown in Figure 4 As shown, Figure 4 In the middle, 1 is the blade, 2 is 1 / 3 of the tower, 3 is the wind turbine, and 4 is the tower base.

[0060] As you can understand, the redundant sensor arrangement was designed based on the Doppler effect, with sensors installed at the upper third of the blade tip. The Doppler effect is generated by the spatial difference between the blade tip and the lower third, capturing and filtering sound signals.

[0061] In one embodiment of the present invention, Figure 5 Arrange a topology for redundant sound sensors, such as Figure 5 As shown, 1, 2, and 3 are sound sensors arranged redundantly with a difference of 120, and 4 is a weather station.

[0062] It's understandable that the signals captured by the sound sensors are redundant and overlap, forming a complete sound capture system. A single sensor can only determine whether an abnormal signal exists, but an array of three sensors can achieve three-dimensional spatial positioning through arrival time difference positioning, leveraging the superposition of the Doppler effect for more accurate positioning. Through adaptive beamforming technology, the three-sensor system can actively suppress interference. The three-sensor architecture has a triple redundancy mechanism. It can include several steps: arrival time difference positioning, Doppler effect superposition positioning, and adaptive beamforming technology:

[0063] Among them, arrival time difference positioning includes:

[0064] Signal acquisition and timestamp recording: Three acoustic sensors, positioned 120 degrees apart, collect the acoustic signals generated by the blade's rotation and record the arrival time of each signal. Because the sensor positions are known and the blade's sound source is relatively fixed, the signals received by the three sensors exhibit time differences.

[0065] The application of time difference calculation and positioning algorithms: Calculate the time difference between the signals received by the three sensors, use the time difference of arrival (TDOA) positioning method combined with geometric relationships, and use algorithms (such as least squares and direction of arrival estimation) to calculate the position of the sound source in three-dimensional space. For example, if the three sensors are A, B, and C, by calculating tAB = tB - tA and tAC = tC - tA, combined with the geometric positions of the sensors, the three-dimensional coordinates of the sound source can be calculated.

[0066] Among them, Doppler effect superposition positioning includes:

[0067] Doppler shift calculation: For each sensor's collected sound signal, the corresponding Doppler shift is calculated using the Doppler effect formula. Because the three sensors are located in different locations, the received Doppler shift will also vary.

[0068] Doppler Effect Superposition: This method uses the Doppler frequency shifts of the three sensors to analyze and superimpose, further improving positioning accuracy. For example, by analyzing the Doppler frequency shift trends and amplitude differences of the three sensors, the direction and speed of the sound source can be more accurately determined, achieving more precise three-dimensional positioning.

[0069] Among them, adaptive beamforming technology includes:

[0070] Beamforming technology application: Adaptive beamforming technology is used to process the sound signals collected by the three sensors. Beamforming technology adjusts the phase and amplitude of the sensor signals to form a beam pointing in a specific direction, enhancing the signal in that direction while suppressing noise and interference in other directions.

[0071] Interference Suppression and Signal Enhancement: Adaptive algorithms (such as minimum mean square error and linearly constrained minimum variance) adjust the beam direction and shape in real time to proactively suppress background noise and other interference sources. For example, when strong interference is detected in a certain direction, beamforming technology can adjust the beam direction to avoid the interference source, thereby improving the signal-to-noise ratio and enhancing the ability to capture blade sound signals.

[0072] In this embodiment of the present invention, a sound sensor is installed at the tip 1 / 3 of the upper portion of the wind turbine tower. Three sound sensors are arranged 120 degrees apart to form an equilateral triangle or circular symmetrical layout. As the wind turbine blades rotate, the Doppler effect generated by the spatial positional difference between the tip and the tip 1 / 3 collects the sound signals generated by the blades. The frequency characteristics of the captured sound signals are analyzed to filter out Doppler frequency shift signals associated with blade ice accumulation. This is how the sound signals are collected.

[0073] Specifically, according to the Doppler shift algorithm, the Doppler shift signal generated by the rotation of wind turbine blades exhibits typical narrowband modulation characteristics. Its fundamental frequency component is strictly constrained by the blade rotational speed, and the harmonic groups are concentrated near integer multiples of the rotational frequency. The background noise in the open space of wind farms is mostly broadband white noise. Motion trajectory decoupling requires the linear velocity of the blade tip, which causes a continuous frequency deviation in the echo signal. The band-stop filter bank is designed with a time-varying notch window to achieve 8-12dB of fixed-point noise suppression while maintaining the integrity of the Doppler sideband information. The analysis of the blade sound signal at the 1 / 3 of the blade tip results in sound signal acquisition.

[0074] Understandably, the acoustic sensor was installed at the upper third of the blade tip on the wind turbine tower. This location is based on the spatial difference between the blade tip and the lower third, ensuring that it can effectively capture the acoustic signal changes caused by the Doppler effect when the blades rotate.

[0075] Three sound sensors are positioned 120 degrees apart to form an equilateral triangle or circular symmetrical layout. This layout enables 360-degree monitoring without blind spots, while also improving the accuracy of sound source localization and the system's redundancy and fault tolerance.

[0076] When a wind turbine blade rotates, the difference in spatial position between the tip and the tip-third causes a Doppler shift in the sound signal as it reaches the sensor. The sound sensor captures the sound signals generated by the rotating blades, which exhibit specific frequency shift characteristics due to the Doppler effect.

[0077] By analyzing the frequency characteristics of the captured sound signals, Doppler shift signals related to blade icing can be screened out. Blade icing changes the blade's vibration characteristics, which in turn affects the Doppler shift characteristics of the sound signals, enabling preliminary screening of icing signals.

[0078] Furthermore, the processed Doppler frequency shift signal is analyzed to determine the fundamental frequency component; based on the correlation between the fundamental frequency component frequency and the blade speed, the blade speed is calculated using the Doppler frequency shift formula; based on the blade speed and wind farm meteorological station parameters, the wind turbine power is calculated using the wind turbine power curve or mathematical model.

[0079] In one embodiment of the present invention, wind farm meteorological station parameters are obtained, and the wind farm meteorological station parameters include multiple ones of wind speed, wind direction, temperature, humidity, air pressure, precipitation, visibility, cloud cover and sunshine hours; meteorological data related to icing of wind turbine blades are obtained based on the wind farm meteorological station parameters.

[0080] Furthermore, statistical methods or sliding window technology are used to detect and eliminate outliers in meteorological data, remove high-frequency noise, and perform linear interpolation or spline interpolation on missing data to obtain processed meteorological data; wavelet transform is used to decompose the signal to remove high-frequency noise components in the sound data and then reconstruct the signal, and median filtering is applied to remove impulse noise; a band-stop filter is designed to suppress background noise in a preset frequency band, and the Doppler frequency shift characteristics are used to design a time-varying notch window to retain the narrowband modulated signal generated by the rotation of the fan blades to obtain processed sound data.

[0081] It is understandable that for meteorological data, the main purpose is to remove outliers or high-frequency noise caused by sensor errors or environmental interference. Common processing methods include:

[0082] Outlier detection and removal: Use statistical methods to identify and remove data points that deviate significantly from the normal range.

[0083] Low-pass filtering: Apply a low-pass filter to remove unnecessary high-frequency components and retain the trend changes of meteorological parameters.

[0084] Interpolation to fill missing values: Use linear interpolation, spline interpolation and other methods to fill gaps in the data to ensure data integrity.

[0085] Noise reduction and filtering of sound data: For sound data, the focus is on removing interference factors such as background noise and mechanical vibration to more accurately capture signals related to the status of wind turbine blades. The specific steps for measuring sound using the Doppler frequency shift algorithm are as follows:

[0086] Band-stop filter: Design a band-stop filter to suppress specific frequency bands (such as broadband white noise) in the background noise of the wind farm to suppress these interferences.

[0087] Doppler shift analysis: The Doppler shift signal generated by the rotation of wind turbine blades has typical narrowband modulation characteristics. Through a time-varying notch window design, we achieve 8-12dB of fixed-point noise suppression while maintaining the integrity of Doppler sideband information. This step is particularly suitable for extracting sound characteristics directly related to blade motion from complex acoustic environments.

[0088] Therefore, both meteorological and acoustic data undergo noise reduction and filtering, though the specific methods and techniques depend on the data type and application scenario. For acoustic data, Doppler shift algorithms are used to accurately measure and analyze acoustic signals, enabling better prediction of wind turbine blade ice coverage. This approach effectively improves the accuracy and reliability of the monitoring system.

[0089] Furthermore, the present invention uses a pre-defined first AI algorithm (such as a convolutional neural network (CNN), a recurrent neural network (RNN), or its variant, an LSTM, in deep learning) to extract features from the pre-processed meteorological data. The AI ​​algorithm automatically extracts meteorological feature information related to blade icing by learning from a large amount of meteorological data samples.

[0090] The AI ​​algorithm identifies patterns and regularities in meteorological data related to blade icing, such as combinations of temperature below a certain threshold, humidity above a certain threshold, and wind speed within a specific range. These characteristics can indicate the likelihood of blade icing under current meteorological conditions.

[0091] Based on the extracted meteorological characteristics, the recurrent neural network (RNN) makes a preliminary assessment of the basic conditions for blade icing. For example, when the temperature is below 0°C and the humidity is above 80%, the blades are initially judged to have a high risk of icing. When the wind speed is between 5-10 m / s and the temperature is between -5°C and -10°C, the icing risk increases further.

[0092] The recurrent neural network (RNN) outputs a preliminary judgment, indicating whether the basic conditions for blade icing are met under the current meteorological conditions. If so, the next step is sound feature analysis and comprehensive judgment. If not, the current blade icing risk is considered low and no further analysis is required.

[0093] Furthermore, the present invention utilizes a pre-defined second AI algorithm (such as a convolutional neural network (CNN) or a recurrent neural network (RNN) or its variant, LSTM, in deep learning) to perform feature extraction on the pre-processed sound data. By learning from a large number of sound data samples, the AI ​​algorithm automatically extracts characteristic information related to blade icing, such as Doppler frequency shift, harmonic distribution, and signal amplitude changes.

[0094] The sound feature information extracted by the second AI algorithm is combined with the meteorological feature information extracted by the first AI algorithm. For example, the Doppler shift feature in the sound signal is combined with the temperature and humidity features in the meteorological data to form a comprehensive feature vector.

[0095] The fused feature vector is then used to perform a comprehensive prediction using a second AI algorithm. By learning from a large number of sound and meteorological data samples, the AI ​​algorithm establishes a mapping relationship between sound and meteorological characteristics and the degree of blade icing, thereby accurately predicting the degree of blade icing. For example, when the Doppler shift amplitude in the sound signal increases and the temperature in the meteorological data is below 0°C, the predicted degree of icing is high.

[0096] Based on the predicted degree of icing, graded warning rules are set. For example, icing can be categorized into three levels: light, moderate, and severe. Light icing may not affect normal wind turbine operation, while moderate icing requires preventive measures. Severe icing requires immediate shutdown and de-icing.

[0097] Based on the prediction results, the system outputs a graded warning message. For example, if the predicted icing level is light, the system outputs "Light icing warning, continuous monitoring recommended"; if the predicted icing level is moderate, the system outputs "Moderate icing warning, preventive measures recommended"; and if the predicted icing level is heavy, the system outputs "Heavy icing warning, immediate shutdown and de-icing recommended."

[0098] The wind turbine blade ice coverage warning diagram of the present invention is as follows: Figure 6 shown.

[0099] According to the wind turbine blade icing degree prediction method based on redundant sound sensors according to an embodiment of the present invention, the three sensors are 120 degrees apart and can evenly cover all directions (each sensor covers an area of ​​approximately 120°), avoiding blind spots in a single direction, and are suitable for omnidirectional perception sound monitoring systems. By comparing the time difference or phase difference of the signals received by different sensors, the direction of the sound source can be judged more accurately and the positioning accuracy can be improved. The three sensors form a geometric baseline, and combined with algorithms (such as least squares method and wave direction estimation), the position of the sound source in a two-dimensional plane can be calculated, which is more reliable than dual-sensor positioning. Beamforming technology can be used to enhance signals in specific directions and suppress the contribution of background noise or interference sources. If one sensor fails, the remaining two can still provide partial functions (such as 180° coverage) to avoid complete paralysis of the system. Multi-sensor data can be cross-validated to reduce the risk of false alarms or missed alarms and improve system robustness. The symmetrical layout with a 120-degree difference simplifies the mathematical modeling of the signal phase relationship and facilitates real-time calculation of the sound wave arrival time difference or frequency characteristics. In wind turbine blade monitoring, a 120-degree layout matches common three-phase mechanical structures, effectively capturing periodic acoustic signatures. A circular or triangular layout provides uniform force distribution during physical installation, securing the blade tip at the top third of the wind turbine tower surface to reduce mechanical stress. A symmetrical layout reduces the complexity of acoustic wave reflection paths and mitigates multipath interference on signal analysis.

[0100] In summary, the redundant three-sensor layout with a 120-degree difference has significant advantages in coverage, positioning accuracy, reliability and engineering implementation.

[0101] In order to implement the above embodiment, Figure 7 As shown, this embodiment also provides a wind turbine blade icing degree prediction system 10 based on redundant sound sensors, including:

[0102] A raw data acquisition module 100 is used to respectively acquire meteorological data related to ice coating on wind turbine blades and sound data generated by blade rotation;

[0103] The data preprocessing module 200 is used to perform noise reduction and filtering operations on the meteorological data and the sound data to obtain corresponding processed meteorological data and processed sound data;

[0104] A basic condition determination module 300 is configured to extract meteorological characteristic information from the processed meteorological data using a preset first AI algorithm, so as to preliminarily determine the basic conditions for blade icing based on the meteorological characteristic information;

[0105] The icing degree prediction module 400 is used to extract sound feature information from the processed sound data using a preset second AI algorithm, and comprehensively predict the blade icing degree based on the sound feature information and meteorological feature information, and finally output graded warning information.

[0106] According to the wind turbine blade icing degree prediction system based on redundant sound sensors according to an embodiment of the present invention, multi-sensor data can be cross-verified, reducing the risk of false alarms or missed alarms and improving system robustness. The symmetrical layout with a mutual difference of 120 degrees simplifies the mathematical modeling of the signal phase relationship and facilitates the real-time calculation of the sound wave arrival time difference or frequency characteristics. In wind turbine blade monitoring, the 120-degree layout can match the common three-phase mechanical structure and effectively capture periodic acoustic characteristics. The annular or triangular layout is evenly stressed during physical installation and is fixed at the blade tip 1 / 3 of the wind turbine tower surface to reduce mechanical stress. The symmetrical layout can reduce the complexity of the sound wave reflection path and alleviate the interference of the multipath effect on signal analysis.

[0107] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0108] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.

Claims

1. A method for predicting the degree of icing on wind turbine blades based on redundant acoustic sensors, characterized in that: include: Acquire meteorological data related to ice coating on wind turbine blades and sound data generated by blade rotation respectively; Performing noise reduction and filtering operations on the meteorological data and the sound data to obtain corresponding processed meteorological data and processed sound data; Extracting meteorological characteristic information from the processed meteorological data using a preset first AI algorithm to preliminarily determine basic conditions for blade icing based on the meteorological characteristic information; The preset second AI algorithm is used to extract sound feature information from the processed sound data, and the degree of blade icing is comprehensively predicted based on the sound feature information and meteorological feature information, and finally graded warning information is output.

2. The method according to claim 1, characterized in that Acquire the sound data generated by blade rotation, including: Install a sound sensor at the 1 / 3 of the blade tip on the upper part of the wind turbine tower; Three sound sensors are arranged 120 degrees apart to form an equilateral triangle or circular symmetrical layout; When the wind turbine blades rotate, the Doppler effect generated by the spatial position difference between the tip and the 1 / 3 of the blade tip is used to collect the sound signals generated by the blade rotation. By analyzing the frequency characteristics of the captured sound signals, the Doppler frequency shift signals related to blade icing are screened out.

3. The method according to claim 1, characterized in that Obtain meteorological data related to wind turbine blade icing, including: Obtaining wind farm weather station parameters, wherein the wind farm weather station parameters include multiple ones of wind speed, wind direction, temperature, humidity, air pressure, precipitation, visibility, cloud cover, and sunshine hours; Meteorological data related to icing on wind turbine blades is obtained based on the wind farm meteorological station parameters.

4. The method according to claim 1, wherein Performing noise reduction and filtering operations on the meteorological data and the sound data to obtain corresponding processed meteorological data and processed sound data includes: Utilize statistical methods or sliding window technology to detect and eliminate outliers in meteorological data, remove high-frequency noise, and perform linear interpolation or spline interpolation on missing data to obtain processed meteorological data; The signal is decomposed using wavelet transform to remove high-frequency noise components in the sound data and then reconstructed. The median filter is applied to remove impulse noise. A band-stop filter is designed to suppress background noise in a preset frequency band, and a time-varying notch window is designed using the Doppler shift characteristics to retain the narrowband modulated signal generated by the rotation of the fan blades to obtain the processed sound data.

5. The method according to claim 2, characterized in that The method further comprises: analyzing the processed Doppler shift signal to determine a fundamental frequency component; According to the correlation between the fundamental frequency component and the blade rotation speed, the blade rotation speed is calculated using the Doppler frequency shift formula; The wind turbine power is calculated based on the blade speed and wind farm meteorological station parameters and using the wind turbine power curve or mathematical model.

6. The method according to claim 1, wherein The first AI algorithm includes a first recurrent neural network model; using the preset first AI algorithm to extract meteorological characteristic information from the processed meteorological data to preliminarily determine basic conditions for blade icing based on the meteorological characteristic information, including: The trained first recurrent neural network model is used to perform feature extraction on the preprocessed meteorological data to extract meteorological feature information related to blade icing; Based on the extracted meteorological characteristic information, a preliminary judgment is made on the basic conditions for blade icing and a preliminary judgment result is output; if the basic conditions for icing are met, the next step is to enter the sound feature analysis and comprehensive judgment.

7. The method according to claim 1, characterized in that The first AI algorithm includes a second recurrent neural network model; the preset second AI algorithm is used to extract sound feature information from the processed sound data, and the degree of blade icing is comprehensively predicted based on the sound feature information and meteorological feature information, and finally, graded warning information is output, including: The trained second recurrent neural network model is used to extract features from the preprocessed sound data, and the extracted sound feature information and meteorological feature information are fused to obtain a fused feature vector; The fused feature vector is processed using the trained second recurrent neural network model to establish a mapping relationship between sound features, meteorological features and blade icing degree to predict the blade icing degree. Set graded warning rules based on the predicted degree of icing, and output corresponding graded warning information based on the prediction results.

8. A wind turbine blade icing degree prediction system based on redundant sound sensors, characterized in that: include: A raw data acquisition module, used to respectively acquire meteorological data related to ice coating on wind turbine blades and sound data generated by blade rotation; A data preprocessing module is used to perform noise reduction and filtering operations on meteorological data and sound data to obtain corresponding processed meteorological data and processed sound data; a basic condition judgment module, configured to extract meteorological characteristic information from the processed meteorological data using a preset first AI algorithm, so as to preliminarily judge the basic conditions for blade icing based on the meteorological characteristic information; The icing degree prediction module is used to extract sound feature information from the processed sound data using a preset second AI algorithm, and comprehensively predict the blade icing degree based on the sound feature information and meteorological feature information, and finally output graded warning information.

Images