Wind power blade clearance monitoring method based on light supplement and clearance-wind speed reconstruction

Abstract

A wind power blade clearance monitoring method based on light supplement and clearance-wind speed reconstruction comprises the steps that light supplement and image acquisition equipment deployment is carried out, specifically, a light supplement system adopts a light intensity and gray level two-factor self-adaption method to switch light supplement modes according to the environment, blade and tower drum image data are acquired, and multi-dimensional data including wind speed, turbulence and blade variable pitch angles are synchronously acquired; performing preprocessing such as graying and noise reduction on the image data to extract features, performing filtering and noise reduction on data such as wind speed and the like and removing abnormal values, and completing space-time alignment on all the data according to timestamps; in the first stage, an LSTM model is used for inputting wind speed integral, turbulence integral and variable pitch angle data in historical short time, and blade basic deformation quantity is output; in the second stage, the initial clearance value is corrected in combination with a Markov state transition equation, and a final tower clearance value is obtained; clearance monitoring and early warning are conducted, specifically, a clearance safety threshold value is preset, and when the final clearance value is lower than the threshold value, an early warning signal is sent to a fan main control system; the blade deformation quantity prediction precision is improved under multi-parameter coupling; and the monitoring error of the clearance value is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of wind turbine blade clearance monitoring, and in particular to a wind turbine blade clearance monitoring method based on light supplementing and clearance-wind speed reconstruction. BACKGROUND

[0002] Under the background of rapid development of the wind power industry, the single machine capacity of wind turbine generators is continuously increasing, and the length of the blades is continuously increasing. The clearance distance monitoring between the blades and the tower has become a key link to ensure the safe operation of the wind turbine. The blades are prone to elastic deformation under the action of wind, especially near the rated wind speed, the waving deformation of the blades reaches the maximum value. If the clearance distance monitoring is not timely or the accuracy is insufficient, it is easy to cause the blade to sweep the tower, causing serious equipment failures such as blade fracture and tower damage, and further causing huge economic losses and long-term shutdown. For example, in October 2025, a Nordex N149 mainstream unit in a wind farm in Germany had a tower collapse accident, and the cabin, hub and blades fell off the top of the tower. The imbalance of the impeller further aggravates the deviation of the blade running track, making the difficulty and importance of clearance monitoring more prominent, so the development of high-precision, all-weather clearance monitoring technology has become an urgent demand in the field of wind power operation and maintenance.

[0003] Current wind turbine blade clearance monitoring technologies mainly include visual monitoring, radar ranging, laser ranging and the like. However, these technologies generally have obvious defects. The visual monitoring technology is greatly affected by light conditions. During the weak light period such as night, early morning or evening, the image clarity greatly decreases, resulting in a sharp increase in the calculation error of the clearance value. Although the radar and laser ranging technologies have improved in light adaptability, they are easily disturbed by wind speed changes. The irregular deformation of the blades caused by wind speed fluctuations cannot be effectively corrected, and the installation of the equipment is complex and the maintenance cost is high. In addition, the existing technologies rely on a single data source for clearance calculation and cannot fuse multi-dimensional working condition data such as wind speed, turbulence and variable pitch angle, so it is difficult to eliminate the monitoring error caused by the coupling of multiple factors and cannot meet the high-precision monitoring demand under complex wind conditions.

[0004] Although light supplementing technology has been tried to be applied to improve the night visual monitoring effect, the existing light supplementing equipment has a limited adjustment range and cannot adaptively adjust the light supplementing intensity according to the environmental changes, resulting in insufficient stability and reliability of the night monitoring data. At the same time, in view of the interference of wind speed on the deformation of the blades, the existing technologies mostly use a simple linear correction model and cannot fully consider the dynamic nonlinear relationship between wind speed, turbulence and blade deformation, so it is difficult to accurately eliminate the clearance calculation deviation caused by wind speed fluctuations. These technical bottlenecks make it difficult to improve the data availability of the current clearance monitoring system, so it is difficult to realize stable monitoring in all-weather and high-precision conditions, which restricts the further improvement of the safety protection level of the wind turbine generator. SUMMARY

[0005] The problem to be solved by the present application is that the existing electric blade clearance monitoring technology relies on a single data source for clearance calculation, it is difficult to eliminate the monitoring error caused by multi-factor coupling, the data availability is difficult to improve, and it is impossible to realize stable monitoring in all-weather and high-precision.

[0006] To solve the above problems, the present application provides a wind power blade clearance monitoring method of light supplementing and clearance-wind speed reconstruction, comprising: Step one: light supplementing and image acquisition device deployment: install a camera and a light supplementing lamp on the cabin; the camera is adjusted at a specific angle through a support to meet the specified imaging requirements, the light supplementing lamp can supplement light for the camera and is calibrated with the camera through a calibration camera, and a fixed calibration coordinate is configured; Step two: all-weather multi-source data acquisition: the control system of the light supplementing lamp switches the light supplementing mode according to the environment by using a light intensity and gray double-factor adaptive method; the image data of the blade and the tower drum are acquired through the camera, and multi-dimensional data including wind speed, turbulence, and blade variable pitch angle are acquired synchronously through other sensors; Step three: multi-modal data preprocessing: the image data acquired in step two are preprocessed including gray scaling and noise reduction to extract features, the wind speed and other data are filtered and denoised and abnormal values are eliminated, and all data are time and space aligned according to the time stamp; Step four: based on the clearance-wind speed reconstruction calculation of the two-stage dynamic decoupling, the data processed in step three are used: in the first stage, the LSTM model inputs the wind speed integral, turbulence integral, and variable pitch angle data in a short time, and outputs the blade basic deformation variable; in the second stage, the initial clearance value is corrected by combining the Markov state transition equation to obtain the final tower clearance value; Step five: clearance monitoring and early warning: preset a clearance safety threshold, and when the final tower clearance value is lower than the threshold, an early warning signal is sent to the wind turbine main control system.

[0007] The specified imaging requirements are that the tower bottom position in the picture is located at the upper left corner of the whole picture, the upper end position is 1 / 8 to 3 / 8 of the picture width, and the left end position is 1 / 6 to 1 / 2 of the picture length; at this position, the overall appearance of the blade path can be seen, and the position where the blade is closest to the tower drum can be seen, which is a good observation angle.

[0008] Further, the method of switching the light supplementing mode by using the light intensity and gray double-factor adaptive method in step two is that the light supplementing lamp is equipped with a light intensity sensor + image gray mean sensor double-channel sensor, which acquires the ambient light intensity and the imaging gray mean of the camera in real time, controls the light supplementing power based on the light intensity and gray double-factor adaptive method , and specifically: ; in To maximize the power of the fill light, the overlap between the fill light area and the camera's field of view is calibrated by calibrating the camera to ensure that the light spot covers no areas without omissions.

[0009] Furthermore, the low-light images captured by the camera ( First, wavelet transform and guided filtering are used to enhance the image in layers: (1) the image is decomposed into low-frequency components (lighting information) and high-frequency components (edge ​​details) by wavelet transform; (2) gamma correction is used to enhance the overall brightness of the low-frequency components and adaptive threshold enhancement is used to enhance the high-frequency components; (3) the enhanced high and low frequency components are fused by guided filtering to suppress high-frequency noise.

[0010] Furthermore, in step four, the specific method for using an LSTM model to input historical wind speed integrals, turbulence integrals, and pitch angle data within a short period of time to output the basic blade deformation is as follows: LSTM blade deformation mapping model construction and output: Construct a 3-layer Long Short-Term Memory (LSTM) network and input the historical wind speed integrals within 20ms. Turbulence integral and pitch angle data The wind speed integral Real-time wind speed during this period The cumulative value is calculated as follows: ,in For integration time variable, The current monitoring time; the turbulence integral Real-time turbulence intensity during this period cumulative value The pitch angle data The input data is the average pitch angle acquired by the pitch encoder at a frequency of 1 ms / time. After min-max normalization, the input data is fed into an LSTM network, where the ReLU activation function is used to output the basic blade deformation through a fully connected layer. The output formula is ,in Let be the hidden state vector of the LSTM network at the last time step. This is the weight matrix.

[0011] Furthermore, the specific method for correcting the initial clearance value and obtaining the final tower clearance value in the second stage of step four, combining the Markov state transition equation, is as follows: (2) Initial clearance value acquisition: the camera collected blade and tower image is grayed, Gaussian filter denoising, Canny edge detection preprocessing, the blade tip and tower edge feature coordinates are extracted, the pixel coordinates are converted into actual three-dimensional space coordinates combined with the camera internal parameter matrix and distortion coefficient obtained through Zhang Zhengyou calibration method, and the initial clearance value is calculated ; (3) Markov dynamic correction calculation: the initial clearance value is dynamically corrected based on the Markov state transition equation , wherein is the current final tower clearance value, , , are respectively a blade deformation correction coefficient, a wind speed integral correction coefficient and a turbulence integral correction coefficient, and the values are 0.85, 0.02 and 0.015 respectively, is the historical clearance value at the previous 10 ms and participates in the state transition function iteration; (4) Model training optimization: 50,000 sets of full working condition measured samples covering wind speed, turbulence intensity and variable pitch angle are used to train the model, the mean square error is used as the loss function, the network parameters and correction coefficients are iteratively updated through the Adam optimizer until the model validation set error is less than the set value.

[0012] Further, in step one, the camera and the light supplement lamp are sleeved with fireproof cloth when installed, the camera support is fixed through the through hole with a bolt, the bolt head is located outside the cabin, the nut is located inside the cabin, and a preset bolt torque is applied after installation; the auxiliary plate is provided when the light supplement lamp support is installed, the auxiliary plate can rotate around the bolt, and a light supplement lamp debugging operation space is reserved.

[0013] Further, in step one, the edge of the cabin opening needs to be rounded and polished, and if the surface of the cabin is not flat, the installation surface is slightly polished and flattened by an angle grinder; the cables of the light supplement lamp and the camera are wrapped through a snake skin pipe, and are deployed along the cable bridge wiring in the cabin.

[0014] Further, the camera is also provided with a lens washing device, including a washing water tank and a washing water tank protection cover, a water spraying pipeline is arranged on the washing water tank, and the water spraying pipeline can spray water to automatically wash the camera lens regularly.

[0015] Further, in step three, the image noise reduction adopts a Gaussian filter method, and the blade tip and tower edge feature extraction adopts edge detection and texture analysis technology.

[0016] Further, in step five, the clearance safety threshold is pre-set according to the fan model parameters, blade length and operating conditions, and the warning signal is transmitted to the fan main control system through the CANOPEN communication mode.

[0017] Technical effects of the application: (1) The light supplementing and clearance-wind speed reconstructing wind power blade clearance monitoring method of the present application, The light intensity and gray double-factor infrared light supplementing self-adaptive adjusting method is adopted to achieve the effects of accurate light supplementing power matching in dim light environment and avoiding image distortion caused by over-supplementing light / under-supplementing light.

[0018] (2) The light supplementing and clearance-wind speed reconstructing wind power blade clearance monitoring method of the present application adopts the wavelet transform and guided filter layered enhancement method to process dim light images, thereby improving the overall brightness and edge detail contrast of the images and suppressing high-frequency noise.

[0019] (3) The light supplementing and clearance-wind speed reconstructing wind power blade clearance monitoring method of the present application adopts the LSTM blade deformation mapping model integrated with light supplementing power and image gray mean value to realize the improvement of blade deformation variable prediction accuracy under multi-parameter coupling.

[0020] (4) The light supplementing and clearance-wind speed reconstructing wind power blade clearance monitoring method of the present application adopts the improved Markov state transition equation to reduce the clearance value monitoring error. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is the overall flow schematic diagram of the clearance monitoring method of the preferred embodiment of the present application; Figure 2 is the installation schematic diagram of the camera of the preferred embodiment of the present application (the red circle at the lower right is the installation position); Figure 3 is the light supplementing lamp installation structure schematic diagram of the preferred embodiment of the present application; Figure 4 is the tower bottom installation position schematic diagram of the preferred embodiment of the present application; Figure 5 is the water tank and protective cover installation effect schematic diagram of the video monitoring monitoring system of the preferred embodiment of the present application; Figure 6 is the clearance distance graph of the video monitoring monitoring of the preferred embodiment of the present application ((a) is the daytime imaging graph; (b) is the nighttime imaging graph); Figure 7 is the clearance efficiency calculation graph of the preferred embodiment of the present application ((a) is the 3# unit clearance efficiency calculation graph; (b) is the 4# unit clearance efficiency calculation graph).

[0022] Figure 8 is the clearance efficiency calculation comparison graph of the preferred embodiment of the present application after adding a radar (the left side is the clearance efficiency calculation graph without adding a radar, and the right side is the clearance efficiency calculation graph after adding a radar).

[0023] The reference signs in the drawings represent: 1 - cabin, 2 - U-shaped mounting base plate, 3 - auxiliary plate, 4 - mounting bolt, 5 - long hole, 6 - water tank, 7 - water injection pipeline. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0025] The embodiment of the present application provides a wind turbine blade clearance monitoring method for light compensation and clearance-wind speed reconstruction, comprising: Step one: please refer to Figures 1-4 , first light compensation and image acquisition device deployment: install the camera and the light compensation lamp by opening corresponding size mounting holes and reserved interfaces on the bottom front side and the front right side of the cabin; the camera is adjusted to a specific angle through a U-shaped support to meet the specified imaging requirements (i.e. the image clearly shows the distance between the blade tip and the tower barrel), the light compensation lamp can be adjusted in angle and calibrated with the camera through the calibration camera, and the fixed calibration coordinates are configured; Wherein, the specified imaging requirements are: the tower bottom position in the picture is located at the upper left corner of the whole picture, the upper end position is 1 / 8 to 3 / 8 of the picture width, and the left end position is 1 / 6 to 1 / 2 of the picture length (see Figure 4 , Figure 4 for details, the columnar object in the upper left corner of the picture is the tower bottom); here, the whole picture of the blade path can be seen, and the position where the blade is closest to the tower barrel can be seen, which is a good observation angle and position.

[0026] In the actual implementation process, the edge of the cabin 1 after opening needs to be rounded and polished, and if the surface of the cabin 1 is not flat, the mounting surface is slightly polished flat by an angle grinder; the cables of the light compensation lamp and the camera are wrapped by a snake skin pipe, and are deployed along the cable bridge wiring in the cabin.

[0027] The light supplement lamp mounting interface is divided into two cases: for new units, an interface is reserved on the right side of the cabin front end (facing the impeller), the center of the hole is 600-800 mm away from the center plane of the cabin, and is not less than 100 mm away from the front edge; for units with video monitoring needs but no reserved interface, post-hole drilling is required, the center point of the hole is 60-800 mm away from the center plane, and attention should be paid to not cutting the pre-buried lightning protection line during operation. The general installation steps are as follows: first, outline the pre-drilled hole with a marker pen, then use an angle grinder to install a cutting piece to cut the edge of the hole and clean up the fiberglass fragments in time, hollow out the middle area, and then use a drill with a grinding head to round the corners and edge of the hole, and if the surface of the cabin is not flat, it can be slightly polished flat with an angle grinder and an angle grinding piece; then put the bracket into the hole and drill 4 holes on the mounting bracket, then install the light supplement lamp bracket, pre-install the bolts and operate from the hole, screw the bolts in (the bolt head is outside the cabin and the nut is inside the cabin) and tighten the nut, without applying torque, to reserve space for light supplement lamp debugging and operation.

[0028] Wherein, the camera and the light supplement lamp are sleeved with fireproof cloth during installation, the camera bracket is fixed through the through hole with a bolt, the bolt head is located outside the cabin, and the nut is located inside the cabin, and a predetermined bolt torque is applied after installation; the light supplement lamp bracket is installed with a U-shaped mounting plate 2 and an auxiliary plate 3, the U-shaped mounting plate 2 is provided with a plurality of long hole 5 for sliding installation of mounting bolts, one end of the auxiliary plate can rotate around the light supplement lamp mounting bolt 4, and the other end is provided with a U-shaped socket that can be clamped on the mounting bolt.

[0029] Please refer to Figure 5 In some preferred embodiments, the camera is also equipped with a lens washing device, including a washing water tank and a washing water tank protective cover, a water spraying pipeline is provided on the washing water tank, and the water spraying pipeline can spray water to automatically wash the camera lens regularly.

[0030] Step two: all-weather multi-source data acquisition: the light supplement system switches the light supplement mode according to the environment using the light intensity and gray double-factor adaptive method, acquires image data of the blade and the tower through the camera, and synchronously acquires multi-dimensional data including wind speed, turbulence, and blade pitch angle; Specifically, the method for switching the light supplement mode using the light intensity and gray double-factor adaptive method is as follows: the light supplement lamp is equipped with a light intensity sensor + image gray mean sensor double-channel sensor, which acquires the ambient light intensity and the gray mean value of the camera imaging based on the light intensity and gray double-factor adaptive method to control the light supplement power Specifically: ; Wherein, 0≤L<100 lux corresponds to "extremely dark / dim light environment" (such as night, overcast indoor), which belongs to the typical scene of machine vision requiring light compensation; 100≤L<300 lux corresponds to weak light / ordinary indoor lighting, and the demand for light compensation is weakened; L≥300 lux corresponds to sufficient light environment (such as overcast outdoor, bright indoor), and no additional light compensation is required.

[0031] For the maximum power of the light compensation lamp, calibrate the camera to calibrate the light compensation area and the camera field of view to ensure that the light spot covers no missing area.

[0032] In some preferred embodiments, for the dim light image collected by the camera, a wavelet transform and guided filter layered enhancement method is used: (1) the image is decomposed into low-frequency components (illumination information) and high-frequency components (edge details) by wavelet transform; (2) the low-frequency component is enhanced by gamma correction to improve the overall brightness, and the high-frequency component is enhanced by adaptive threshold; (3) the enhanced high and low frequency components are fused by guided filtering to suppress high frequency noise.

[0033] Step three: multi-modal data preprocessing: image data is preprocessed for feature extraction, such as grayscale and noise reduction, and wind speed data is filtered and denoised to remove outliers, and all data is time and space aligned according to the timestamp; wherein, the noise reduction uses Gaussian filtering method, and the blade tip and tower edge feature extraction uses edge detection and texture analysis technology.

[0034] Step four: clearance-wind speed reconstruction calculation based on two-stage dynamic decoupling: the first stage uses an LSTM model to input the wind speed integral, turbulence integral and variable pitch angle data in a short period of history, and outputs the blade basic deformation variable; the second stage combines the Markov state transition equation to correct the initial clearance value to obtain the final tower clearance value; In some preferred embodiments, the specific method of the first stage is: LSTM blade deformation mapping model construction and output: a 3-layer long short-term memory network (LSTM) is constructed, and the wind speed integral , turbulence integral and variable pitch angle data in the past 20 ms are input; The wind speed integral is the cumulative value of the real-time wind speed in this period, and the calculation method is , where is the integral time variable, is the current monitoring time; the turbulence integral is the cumulative value of the real-time turbulence intensity in this period; The average value of the pitch angle of the period collected by the pitch encoder at a frequency of 1 ms / time; after the input data are subjected to min-max standardization processing, the data are sent to an LSTM network, preferably, the number of nodes in the network hidden layer is set to 64, a ReLU activation function is used, and a blade basic deformation variable is output through a fully connected layer , the output formula is , wherein is the hidden state vector of the last time step of the LSTM network, at this time, is a 64x1-dimensional weight matrix.

[0035] The specific method of the second stage is as follows: (1) Initial clearance value acquisition: the blade and tower image collected by the camera is subjected to grayscale, Gaussian filter denoising, Canny edge detection preprocessing, the blade tip and tower edge feature coordinates are extracted, the pixel coordinates are converted into actual three-dimensional space coordinates in combination with the camera intrinsic matrix and distortion coefficient obtained through Zhang Zhengyou calibration method, and the initial clearance value is calculated ; (2) Markov dynamic correction calculation: the initial clearance value is dynamically corrected based on a Markov state transition equation, the state transition equation is: , wherein is the current final tower clearance value, , , are respectively a blade deformation correction coefficient, a wind speed integral correction coefficient and a turbulence integral correction coefficient, and the values can be, for example, 0.85, 0.02 and 0.015 respectively, is the historical clearance value at the time of 10 ms ago and participates in the iteration of the state transition function; (3) Model training and optimization: a plurality of groups of full-condition measured samples covering wind speed, turbulence intensity and pitch angle are used to train the model, the mean square error is used as the loss function, the network parameters and correction coefficients are iteratively updated through the Adam optimizer until the model validation set error is less than a set value.

[0036] Step five: clearance monitoring and early warning: a preset clearance safety threshold is set, when the final clearance value is lower than the threshold, an early warning signal is sent to the fan main control system. The clearance safety threshold is preset according to the fan model parameters, blade length and operating conditions, and the early warning signal is transmitted to the fan main control system through the CANOPEN communication mode.

[0037] Please refer to Figure 6 , Figure 7The embodiments of the present application achieve good imaging effects in the daytime and at night. According to the actual implementation of the unit, the 3# unit has a full-day clearance efficiency of 99.93%, a daytime clearance efficiency of 99.88%, and a night clearance efficiency of 99.96%; the 4# unit has a full-day clearance efficiency of 99.9%, a daytime clearance efficiency of 99.82%, and a night clearance efficiency of 99.95%.

[0038] The clearance efficiency in the daytime and at night is more than 99.8%. The prediction accuracy of the blade deformation under the coupling of multiple parameters is improved; and the clearance monitoring error is reduced.

[0039] In some preferred embodiments, the influence of fog, heavy fog, rain, snow and other environments on camera imaging is considered (for example, the wind field in some areas of Guizhou), and a millimeter wave radar device can also be added to monitor the clearance. At this time, a data processing method fusing radar data is also proposed: The filtered radar data , image data collected by the camera, and time stamps of working condition data such as wind speed, turbulence, and variable pitch angle are converted into UTC standard time with a time accuracy of 1ms; the image data collection time is taken as the reference, radar data and working condition data with a time difference of less than 5ms are searched, the average value is taken if there are multiple sets of matching data, and the corresponding time data is generated through linear interpolation if there is no matching data; the aligned visual feature data (blade tip and tower edge coordinates), radar distance data , and working condition data (wind speed integral , turbulence integral , and variable pitch angle ) are stored in time sequence, forming a three-dimensional data matrix of “vision + radar + working condition” , wherein is the visual feature matrix, is the radar distance matrix, is the working condition parameter matrix.

[0040] Based on the ambient light intensity collected by the light intensity sensor carried by the camera, the light scene is divided: (1) sufficient light scene: and the average gray value of camera imaging ; (2) weak light / bad weather scene: or (including light, heavy fog, dense fog, rain, snow and other environments).

[0041] Among them, for the sufficient light scene, the visual feature matrix at the corresponding time is extracted from the three-dimensional data matrix to calculate the visual clearance value. Radar distance data in radar distance matrix is extracted as the main part Cross-validation is performed as the auxiliary part, and the intermediate value of the fused clearance is : ; If , trigger data validity verification, discard data with large deviation and reacquire; Among them, for weak light / poor weather scene: extract radar distance data in radar distance matrix from three-dimensional data matrix ; Extract the visual feature matrix as the main part , and the visual clearance value calculated is modified as the auxiliary part, and the intermediate value of the fused clearance is : If the visual data is invalid (such as unable to extract edge features in heavy fog weather), directly use the radar distance data as the intermediate value of the clearance.

[0042] The proportion of heavy fog weather during the data analysis period of the wind farm in Hunan and the wind farm in Guizhou is high, and the utilization rate of the radar rear clearance system is obviously improved, and the average utilization rate is improved by more than 6% (as shown in Figure 8 ).

[0043] Those skilled in the art can easily understand that the above description is only a preferred embodiment of the present application, and is not intended to limit the present application, and any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for monitoring the air clearance of wind turbine blades based on supplemental lighting and air clearance-wind speed reconstruction, characterized in that, Step 1: Deployment of supplementary lighting and image acquisition equipment: Install cameras and supplementary lights on the cabin; the camera can be adjusted to a specific angle via a bracket to achieve the specified imaging requirements, and the supplementary lights can provide supplementary lighting for the camera and are calibrated with the camera via a calibration camera, and fixed calibration coordinates are configured; Step 2: All-weather multi-source data acquisition: The supplementary lighting control system switches the supplementary lighting mode according to the environment using a dual-factor adaptive method of light intensity and grayscale; it acquires image data of the blades and tower through cameras, and simultaneously acquires multi-dimensional data including wind speed, turbulence, and blade pitch angle through other sensors; Step 3: Multimodal data preprocessing: The image data acquired in Step 2 is preprocessed, including grayscale conversion and noise reduction, to extract features. Data such as wind speed is filtered to remove noise and outliers are eliminated. All data is spatiotemporally aligned according to timestamps. Step 4: For the data processed in Step 3, calculate the headroom-wind speed reconstruction based on two-stage dynamic decoupling: In the first stage, use the LSTM model to input the wind speed integral, turbulence integral and pitch angle data within a short historical period, and output the basic blade deformation. The second stage combines the Markov state transition equation to correct the initial clearance value and obtain the final tower clearance value. Step 5: Clearance monitoring and early warning: A preset clearance safety threshold is set. When the final tower clearance value is lower than the threshold, an early warning signal is sent to the wind turbine main control system.

2. The wind turbine blade clearance monitoring method based on supplementary lighting and clearance-wind speed reconstruction according to claim 1, characterized in that, The specified imaging requirements are as follows: the base of the tower in the image is located at the upper left corner of the overall image, the upper position is 1 / 8 to 3 / 8 of the image width, and the left position is 1 / 6 to 1 / 2 of the image length.

3. The wind turbine blade clearance monitoring method based on supplementary lighting and clearance-wind speed reconstruction according to claim 1, characterized in that, The method for switching the supplementary lighting mode using a dual-factor adaptive approach based on light intensity and grayscale in step two is as follows: the supplementary light is equipped with a dual-channel sensor consisting of a light intensity sensor and an image grayscale mean sensor to collect ambient light intensity data in real time. With the average grayscale value of the camera image Controlling the supplementary light power based on a dual-factor adaptive method of light intensity and grayscale Specifically: ； in To maximize the power of the fill light, the overlap between the fill light area and the camera's field of view is calibrated by calibrating the camera to ensure that the light spot covers no areas without omissions.

4. The wind turbine blade clearance monitoring method based on supplementary lighting and clearance-wind speed reconstruction according to claim 1, characterized in that: For low-light images captured by the camera, a wavelet transform and guided filtering layered enhancement method is first used: (1) the image is decomposed into low-frequency components (light information) and high-frequency components (edge ​​details) by wavelet transform; (2) gamma correction is used to enhance the overall brightness of the low-frequency components, and adaptive threshold enhancement is used to enhance the high-frequency components; (3) the enhanced high and low frequency components are fused by guided filtering to suppress high-frequency noise.

5. The wind turbine blade clearance monitoring method based on supplementary lighting and clearance-wind speed reconstruction according to claim 1, characterized in that, In step four, the first stage of using an LSTM model to input wind speed integral, turbulence integral, and pitch angle data within a short historical period, and outputting the basic blade deformation variables, is specifically as follows: LSTM blade deformation mapping model construction and output: Construct at least a 3-layer long short-term memory network, and input the wind speed integral within a historical 20ms period. Turbulence integral and pitch angle data The wind speed integral Real-time wind speed during this period The cumulative value is calculated as follows: ,in For integration time variable, The current monitoring time; the turbulence integral Real-time turbulence intensity during this period The cumulative value, The pitch angle data The input data is the average pitch angle acquired by the pitch encoder at a frequency of 1 ms / time. After min-max normalization, the input data is fed into an LSTM network, where the ReLU activation function is used to output the basic blade deformation through a fully connected layer. The output formula is ,in Let be the hidden state vector of the LSTM network at the last time step. This is the weight matrix.

6. The wind turbine blade clearance monitoring method based on supplementary lighting and clearance-wind speed reconstruction according to claim 5, characterized in that, The specific method for correcting the initial clearance value and obtaining the final tower clearance value in the second stage of step four is as follows: Initial clearance value acquisition: The images of the blades and tower acquired by the camera are preprocessed by grayscale conversion, Gaussian filtering for noise reduction, and Canny edge detection to extract the feature coordinates of the blade tips and tower edges. Combined with the camera intrinsic parameter matrix and distortion coefficients obtained by Zhang Zhengyou calibration method, the pixel coordinates are converted into actual three-dimensional spatial coordinates, and the initial clearance value is calculated. ; Markov dynamic correction calculation: The initial net value is dynamically corrected based on the Markov state transition equation, wherein the state transition equation is as follows: ,in This represents the current final tower clearance value. , , These are the blade deformation correction factor, wind speed integral correction factor, and turbulence integral correction factor, respectively. The historical net value for the first 10ms is used and participates in the state transition function iteration; Model training and optimization: The model is trained using multiple sets of real-world samples covering wind speed, turbulence intensity, and pitch angle. The mean squared error is used as the loss function, and the network parameters and correction coefficients are iteratively updated using the Adam optimizer until the model validation set error is less than the set value.

7. The wind turbine blade clearance monitoring method based on supplementary lighting and clearance-wind speed reconstruction according to claim 1, characterized in that, In step one, both the camera and the fill light are covered with fireproof cloth during installation. The camera bracket is fixed with bolts through the through holes, with the bolt heads outside the cabin and the nuts inside. A preset bolt torque is applied after installation. When installing the fill light bracket, a U-shaped mounting base plate and an auxiliary plate are provided. The mounting base plate has multiple elongated holes for sliding installation of the mounting bolts. One end of the auxiliary plate can rotate around the fill light mounting bolts, and the other end has a U-shaped locking slot that can be locked onto the mounting bolts.

8. The wind turbine blade clearance monitoring method based on supplementary lighting and clearance-wind speed reconstruction according to claim 1, characterized in that, In step one, after the nacelle is opened, the edges need to be rounded and polished. If the nacelle surface is uneven, the mounting surface is lightly polished and smoothed using an angle grinder. The cables for the fill light and camera are wrapped in braided tubing and deployed along the cable bridge inside the nacelle.

9. The wind turbine blade clearance monitoring method based on supplementary lighting and clearance-wind speed reconstruction according to any one of claims 1-8, characterized in that, The camera is also equipped with a lens washing device, which includes a washing water tank and a water spray pipe on the washing water tank. The water spray pipe can spray water to automatically wash the camera lens at regular intervals.

10. The wind turbine blade clearance monitoring method based on supplementary lighting and clearance-wind speed reconstruction according to any one of claims 1-8, characterized in that, In step five, the airspace safety threshold is preset based on the wind turbine model parameters, blade length, and operating conditions, and the warning signal is transmitted to the wind turbine main control system via CANOPEN communication.

Images