Fan blade icing state monitoring method and system based on multistage information fusion
By using multi-level information fusion technology, combined with capacitive and image sensors, the problems of accuracy and real-time monitoring of wind turbine blade icing have been solved, realizing automated monitoring and decision support for the icing status of wind turbine blades.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-11
- Publication Date
- 2026-03-13
AI Technical Summary
In existing technologies, wind turbine blade icing monitoring relies on manual inspections and single sensors, which are labor-intensive, dangerous, have delays, and are prone to false alarms and missed alarms. It also fails to effectively integrate multi-source data, resulting in inaccurate judgment of icing status.
A multi-level information fusion method is adopted, combining capacitive sensors and image sensors. Through deep learning and Bayesian inference models, blade images and capacitive data are acquired simultaneously to estimate and fusion analyze the icing area, type and thickness, and output the icing status, average ice thickness and confidence level.
It enables accurate and reliable monitoring of the icing status of wind turbine blades, reduces false alarms and missed alarms, improves the comprehensiveness and real-time nature of icing judgment, and supports automated de-icing decision-making.
Smart Images
Figure CN121660976A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of wind power generation technology, specifically to a method, system, and storage medium for monitoring the icing status of wind turbine blades based on multi-level information fusion. Background Technology
[0002] With technological advancements, wind power is poised to become a primary source of energy for the energy revolution. However, in cold and damp winter regions, icing on wind turbine blades can have multifaceted impacts on the safety, stability, and power generation efficiency of wind turbine units. Icing alters the aerodynamic shape of the blades, leading to decreased power generation efficiency, load imbalance in the unit, and even potentially causing serious accidents such as collapse.
[0003] In existing technologies, icing monitoring largely relies on traditional manual inspections, which are labor-intensive and pose certain risks and time delays in harsh natural environments. Besides manual inspections, monitoring technologies mostly depend on single-type sensors, each with limitations that cannot comprehensively, accurately, or reliably reflect the state of icing (presence, type, thickness, distribution, etc.). To address this technical problem, existing technologies sometimes use multiple sensors, but these are merely simple data stacking or parallel alarm connections. They fail to leverage the complementary value of multi-source data from an information fusion perspective and do not include machine learning for icing recognition in natural images, leading to false alarms and missed alarms in complex operating conditions. Summary of the Invention
[0004] The purpose of this application is to provide a method, system, and storage medium for monitoring the icing status of wind turbine blades based on multi-level information fusion.
[0005] To achieve the above objectives, the first aspect of this application provides a method for monitoring the icing status of wind turbine blades based on multi-level information fusion, comprising: Simultaneously acquire raw images of the wind turbine blades and raw capacitance data collected based on a capacitive sensor array deployed at key locations on the wind turbine blades. The capacitive sensor array includes capacitive sensors installed at multiple key locations. Blade region extraction and icing analysis are performed on the original image to determine the icing area, icing type, and thickness estimate of the wind turbine blades; Temperature compensation and calibration are performed on the raw capacitance data to determine the local ice thickness value at each capacitance sensor mounting point; Based on the Bayesian inference model, the icing area, icing type, thickness estimate, and local ice thickness value at each capacitive sensor installation point are fused and analyzed to finally determine the icing data of the wind turbine blades. The icing data includes at least the icing state, average ice thickness, maximum ice thickness, and confidence level.
[0006] In this embodiment of the application, leaf region extraction and icing analysis of the original image include: A deep learning-based target detection algorithm is used to identify and extract the blade region as the region of interest from the original image; icing texture analysis is performed on the region of interest to determine the icing type, and image segmentation is performed to calculate the icing area; the baseline contour of the wind turbine blade in the non-icing state is obtained in advance; the current contour of the wind turbine blade is extracted from the original image; the distance between the current contour and the baseline contour in the normal vector direction is determined to determine the thickness estimate.
[0007] In this embodiment of the application, the target detection algorithm is the YOLO model, which adopts the transfer learning method and adjusts the model parameters of the YOLO model based on the pre-trained weights.
[0008] In this embodiment of the application, temperature compensation and calibration of the original capacitance data includes: compensating the original capacitance data based on the reading of the built-in temperature sensor and a pre-calibrated relationship curve between temperature and capacitance; and determining the local ice thickness value corresponding to the compensated capacitance value based on the compensated capacitance value, a pre-calibrated capacitance-ice-thickness conversion model, or a capacitance-ice-thickness mapping table.
[0009] In this embodiment of the application, the original image is acquired by an image acquisition device. Based on the Bayesian inference model, the fusion analysis of the ice area, ice type, thickness estimate, and local ice thickness value at each capacitive sensor installation point includes: setting the prior probability of the ice state based on historical data; and determining the first likelihood of the visual analysis result and the second likelihood of the capacitance measurement result. Based on the estimated ice area, ice type, thickness, and local ice thickness at each capacitive sensor installation point, the posterior probability of different ice states is calculated using Bayes' theorem. The ice state with the highest posterior probability is taken as the final fusion decision output, and this highest posterior probability value is taken as the confidence level of the ice state. Based on the local ice thickness at each capacitive sensor installation point, the average ice thickness and the maximum ice thickness are determined.
[0010] The second aspect of this application provides an online monitoring system for the icing status of wind turbine blades based on multi-level information fusion, comprising: The vision subsystem is used to acquire raw images of wind turbine blades and perform blade region extraction and icing analysis on the raw images to determine the icing area, icing type, and thickness estimate of the wind turbine blades. The capacitive sensor subsystem includes a capacitive sensor array deployed at key locations on the wind turbine blades. It is used to collect raw capacitance data from multiple key locations and perform temperature compensation and calibration on the raw capacitance data to determine the local ice thickness value at each capacitive sensor installation point. The information fusion module communicates with the vision subsystem and the capacitive sensor subsystem to receive icing area, icing type, thickness estimate, and local ice thickness value at each capacitive sensor installation point. It then performs fusion analysis based on a Bayesian inference model to ultimately determine the icing data of the wind turbine blades. The icing data includes at least the icing state, average ice thickness, maximum ice thickness, and confidence level.
[0011] In this embodiment of the application, the vision subsystem includes: Image acquisition module, used to acquire raw images containing the leaves; The ROI extraction module uses a deep learning-based target detection algorithm to identify and extract leaf regions as regions of interest from the original image; The icing analysis module is used to perform icing texture analysis on the region of interest to determine the icing type and to perform image segmentation to calculate the icing area; to obtain the reference profile of the wind turbine blade in the non-icing state in advance; to extract the current profile of the wind turbine blade from the original image; and to determine the distance between the current profile and the reference profile in the normal vector direction to determine the thickness estimate.
[0012] In this embodiment of the application, the capacitive sensor subsystem includes: A capacitive sensor array is deployed at the leading edge and tip of the blade. The signal processing module is used to perform temperature compensation, filtering and noise reduction, and calibration on the raw capacitance data. The local ice thickness determination module is used to determine the local ice thickness value corresponding to the compensated capacitance value based on the compensated capacitance value, a pre-calibrated capacitance-ice-thickness conversion model, or a capacitance-ice-thickness mapping table.
[0013] In this embodiment, the information fusion module is further configured to: set the prior probability of icing state based on historical data; determine the first likelihood of the visual analysis result and the second likelihood of the capacitance measurement result; calculate the posterior probability of different icing states based on the icing area, icing type, thickness estimate, and local ice thickness value at each capacitance sensor mounting point according to Bayes' theorem; take the icing state with the highest posterior probability as the final fusion decision output, and take the maximum posterior probability value as the confidence level of icing; and determine the average ice thickness and maximum ice thickness based on the local ice thickness value at each capacitance sensor mounting point.
[0014] A third aspect of this application provides a machine-readable storage medium storing instructions that, when executed by a processor, configure the processor to perform the aforementioned method for monitoring the icing status of wind turbine blades based on multi-level information fusion.
[0015] This method uses an information fusion center to perform decision-level fusion of two types of heterogeneous data, outputting the optimal icing state judgment. First, the blades are intelligently located, and then global information on icing is extracted from three dimensions: texture (type), pixels (area), and geometry (thickness). This information is then combined with data provided by a capacitive sensor, and finally, Bayesian inference is used to achieve better monitoring results, overcoming the inherent defects of insufficient reliability in single visual monitoring methods.
[0016] Other features and advantages of the embodiments of this application will be described in detail in the following detailed description section. Attached Figure Description
[0017] The accompanying drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the following detailed description to explain the embodiments of this application, but do not constitute a limitation on the embodiments of this application. In the drawings: Figure 1 The schematic diagram illustrates a flowchart of a wind turbine blade icing status monitoring method based on multi-level information fusion according to an embodiment of this application; Figure 2 This illustration shows a schematic diagram of the processing flow of a wind turbine blade icing status monitoring system based on multi-level information fusion according to an embodiment of this application; Figure 3 A schematic diagram illustrating a Bayesian inference flowchart according to an embodiment of this application is shown. Figure 4 The diagram illustrates the internal structure of a computer device according to an embodiment of this application. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for illustration and explanation of the embodiments of this application and are not intended to limit the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.
[0019] Figure 1 The illustration shows a schematic flowchart of a wind turbine blade icing status monitoring method based on multi-level information fusion according to an embodiment of this application. Figure 1 As shown in one embodiment of this application, a method for monitoring the icing status of wind turbine blades based on multi-level information fusion is provided, including the following steps: Step 102: Simultaneously acquire the original image of the wind turbine blade and the original capacitance data collected based on the capacitance sensor array deployed at key locations on the wind turbine blade. The capacitance sensor array includes capacitance sensors installed at multiple key locations. Step 104: Extract blade regions and perform icing analysis on the original image to determine the icing area, icing type, and thickness estimate of the wind turbine blades. Step 106: Perform temperature compensation and calibration on the original capacitance data to determine the local ice thickness value at each capacitance sensor installation point; Step 108: Based on the Bayesian inference model, the icing area, icing type, thickness estimate, and local ice thickness value at each capacitive sensor installation point are fused and analyzed to finally determine the icing data of the wind turbine blades. The icing data includes at least the icing state, average ice thickness, maximum ice thickness, and confidence level. First, for the image acquisition device and the capacitive sensor, timestamps can be aligned to ensure that the data acquired by these two devices are synchronized in time. This allows for the simultaneous acquisition of raw images of the wind turbine blades and raw capacitance data collected by an array of capacitive sensors deployed at key locations on the blades. The image acquisition device can be a camera, video camera, or other image acquisition equipment with image capture capabilities, or a vision sensor, etc. No specific limitation is made here. The raw images of the wind turbine blades can be images of the turbine directly captured by high-definition cameras or vision sensors deployed on the nacelle or tower. The capacitive sensor array includes capacitive sensors installed at multiple key locations. Each capacitive sensor can acquire raw capacitance data at its designated installation point.
[0020] Furthermore, the processor can perform blade region extraction and icing analysis on the original image using deep learning-based object detection algorithms, such as the YOLO algorithm. It automatically identifies the location of the wind turbine blades from the original image and crops them out from complex backgrounds (such as the sky, clouds, and ground). This cropped region can be called the Region of Interest (ROI). Then, a series of image processing algorithms can be applied to the extracted blade ROI image to determine the icing area, icing type, and thickness estimate of the wind turbine blades. Preliminary classification of icing types can be performed based on texture features. Icing types include rime, hoarfrost, etc. Specifically, image segmentation algorithms can be used to separate the ice layer region from the blade background region in the image, and then the percentage or absolute area of ice pixels relative to the total blade pixels can be calculated to determine the icing area of the wind turbine blades.
[0021] Furthermore, after acquiring the raw capacitance data, the processor can preprocess the data. For example, it can use low-pass filtering or moving average algorithms to smooth the data and suppress electronic noise and transient interference. Then, temperature compensation and calibration are performed on the preprocessed raw capacitance data to determine the local ice thickness value at each capacitance sensor mounting point. Since the results of the capacitance path are local and precise, the processor can output a set of numerical matrices C after analyzing and processing each capacitance data point. Where C refers to the ice thickness value, This is a coordinate used to uniquely identify the installation location of each capacitive sensor on the blade surface.
[0022] Then, based on a Bayesian inference model, the processor can perform a fusion analysis of the icing area, icing type, thickness estimate, and local ice thickness values at each capacitive sensor installation point to ultimately determine the icing data for the wind turbine blades. This icing data includes at least the icing state, average ice thickness, maximum ice thickness, and confidence level. Icing states include multiple states such as "no ice," "light icing," and "severe icing." The average ice thickness reflects the overall ice load on the blades, and the maximum ice thickness can be used to assess the most severe local risk points. The confidence level is the maximum posterior probability value calculated by the Bayesian model.
[0023] In one embodiment, leaf region extraction and icing analysis of the original image includes: A deep learning-based target detection algorithm is used to identify and extract the blade region as the region of interest from the original image; icing texture analysis is performed on the region of interest to determine the icing type, and image segmentation is performed to calculate the icing area; the baseline contour of the wind turbine blade in the non-icing state is obtained in advance; the current contour of the wind turbine blade is extracted from the original image; the distance between the current contour and the baseline contour in the normal vector direction is determined to determine the thickness estimate.
[0024] In this embodiment, the image can be processed using the YOLO deep learning object detection model to automatically identify and accurately define the blade region, while eliminating interference from irrelevant backgrounds such as the sky, ground, and tower. Then, depth analysis can be performed on the cropped blade ROI image. An image segmentation algorithm (based on thresholding and texture) is used to distinguish the ice layer from the blade background, and the proportion of iced pixels is calculated to determine the icing area. Since different types of ice have different visual texture characteristics, the type of ice can also be determined by analyzing the texture features of the ice layer in the ROI image (such as roughness, graininess, and uniformity). For example: rime ice: usually relatively transparent and smooth, with a darker color and uniform texture after imaging. Hoarfrost: usually white and opaque, exhibiting a granular or feather-like rough texture. Furthermore, the baseline contour of the wind turbine blade in an un-iced state can be obtained in advance, and then the current contour of the wind turbine blade can be extracted from the original image. The distance between the current contour and the baseline contour in the normal vector direction is then determined to determine the thickness estimate.
[0025] Furthermore, in one embodiment, the object detection algorithm is the YOLO model, employing transfer learning to adjust the model parameters of the YOLO model based on pre-trained weights. Specifically, the YOLO model can be trained first, with the following steps: 1) Label the sampled images. In each image, select every visible leaf with a rectangle and assign a label in the format (object-class, x_center, y_center, width, height) (where object-class represents category 0 or 1, the center point of its bounding box is located at the image size (x_center, y_center), and the width and height are the image width and height, respectively. All coordinates are normalized values relative to the image size). Save this data in a .txt file.
[0026] 2) Use the "transfer learning" method and train the YOLO model. Here, transfer learning can use the pre-trained model weights (.pt file) with strong feature extraction capabilities on a large dataset, and the model has already learned how to detect general objects.
[0027] 3) After loading the pre-trained weights, fine-tune them using existing wind turbine blade data to determine the parameters of the final model. Deploy the model to the monitoring system processor for field application and output a high-confidence bounding box [x1, y1, x2, y2, confidence, class], where: (x1, y1) are the pixel coordinates of the upper left corner of the bounding box; (x2, y2) are the pixel coordinates of the lower right corner of the bounding box; confidence: confidence; class: class.
[0028] In practical applications of the YOLO model, it can first be used to determine the icing texture of samples after ROI region extraction. Global thresholding is used to convert the ROI-extracted image from RGB to grayscale, and a fixed grayscale value is selected as the threshold T. Pixels with grayscale values greater than T are identified as ice, and those less than T are identified as leaves, then the icing area is calculated. Next, the Canny edge detection algorithm can be used to extract the leaf contours without icing as the baseline template contour. Then, contour extraction is performed on the real-time sampled data. The ice thickness d is calculated as follows:
[0029] in Let i be the unit normal vector of the reference profile at point i.
[0030] In one embodiment, temperature compensation and calibration of the original capacitance data includes: compensating the original capacitance data based on a pre-calibrated temperature-capacitance relationship curve according to the reading of the built-in temperature sensor; and determining the local ice thickness value corresponding to the compensated capacitance value based on the compensated capacitance value, a pre-calibrated capacitance-ice-thickness conversion model, or a capacitance-ice-thickness mapping table.
[0031] Capacitive sensor readings are susceptible to drift due to ambient temperature. Temperature compensation corrects the original capacitance value based on the built-in temperature sensor readings, using a pre-calibrated temperature-capacitance curve to eliminate temperature interference. The processed capacitance value is then substituted into a pre-calibrated capacitance-ice thickness conversion model or capacitance-ice thickness mapping table to convert the capacitance value into a specific physical thickness value. In other words, after temperature compensation and calibration, the precise physical ice thickness at each sensor's installation point can be determined—the local ice thickness value at each installation point.
[0032] In one embodiment, the original image is acquired using an image acquisition device. The fusion analysis based on a Bayesian inference model of icing area, icing type, thickness estimate, and local ice thickness at each capacitive sensor mounting point includes: setting a prior probability of the icing state based on historical data; The first likelihood of the visual analysis results and the second likelihood of the capacitance measurement results are determined. Based on the icing area, icing type, thickness estimate, and local ice thickness value at each capacitance sensor mounting point, the posterior probability of different icing states is calculated according to Bayes' theorem. The icing state with the highest posterior probability is taken as the final fusion decision output, and this maximum posterior probability value is taken as the confidence level of icing. Based on the local ice thickness value at each capacitance sensor mounting point, the average ice thickness and the maximum ice thickness are determined.
[0033] In this embodiment, a fusion algorithm based on Bayesian inference is used to compare and evaluate the confidence level of the thickness estimate output by the vision subsystem with the precise thickness value output by the capacitive sensor. Simultaneously, by integrating the icing area information, the joint icing probability and the final ice thickness are calculated. The assumed probability of icing is as follows: .
[0034] Wherein, P(V|I) is the likelihood of the visual sensor, i.e., the first likelihood, which is the prior probability of determining whether there is ice or not at a certain ice thickness. P(C|I) is the likelihood of the capacitive sensor, i.e., the second likelihood. As a normalization factor, it is necessary to calculate all states. The numerators and summation:
[0035] Then, the final decision results after fusion can be output, including the overall icing status, average ice thickness, maximum ice thickness, and icing distribution confidence, providing a control basis for the wind turbine de-icing system and operation strategy.
[0036] This method uses an information fusion center to perform decision-level fusion of two types of heterogeneous data, outputting the optimal icing state judgment. First, the blades are intelligently located, and then global information on icing is extracted from three dimensions: texture (type), pixels (area), and geometry (thickness). This information is then combined with data provided by a capacitive sensor, and finally, Bayesian inference is used to achieve better monitoring results, overcoming the inherent defects of insufficient reliability in single visual monitoring methods.
[0037] Figure 1 This is a flowchart illustrating a wind turbine blade icing status monitoring method based on multi-level information fusion in one embodiment. It should be understood that, although... Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise explicitly stated herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.
[0038] In one embodiment, an online monitoring system for wind turbine blade icing status based on multi-level information fusion is provided (not shown in the figure). The system includes: The vision subsystem is used to acquire raw images of wind turbine blades and perform blade region extraction and icing analysis on the raw images to determine the icing area, icing type, and thickness estimate of the wind turbine blades. The capacitive sensor subsystem includes a capacitive sensor array deployed at key locations on the wind turbine blades. It is used to collect raw capacitance data from multiple key locations and perform temperature compensation and calibration on the raw capacitance data to determine the local ice thickness value at each capacitive sensor installation point. The information fusion module communicates with the vision subsystem and the capacitive sensor subsystem to receive icing area, icing type, thickness estimate, and local ice thickness value at each capacitive sensor installation point. It then performs fusion analysis based on a Bayesian inference model to ultimately determine the icing data of the wind turbine blades. The icing data includes at least the icing state, average ice thickness, maximum ice thickness, and confidence level.
[0039] In one embodiment, the vision subsystem includes: Image acquisition module, used to acquire raw images containing the leaves; The ROI extraction module uses a deep learning-based target detection algorithm to identify and extract leaf regions as regions of interest from the original image; The icing analysis module is used to perform icing texture analysis on the region of interest to determine the icing type and to perform image segmentation to calculate the icing area; to obtain the reference profile of the wind turbine blade in the non-icing state in advance; to extract the current profile of the wind turbine blade from the original image; and to determine the distance between the current profile and the reference profile in the normal vector direction to determine the thickness estimate.
[0040] In one embodiment, the capacitive sensor subsystem includes: A capacitive sensor array is deployed at the leading edge and tip of the blade. The signal processing module is used to perform temperature compensation, filtering and noise reduction, and calibration on the raw capacitance data. The local ice thickness determination module is used to determine the local ice thickness value corresponding to the compensated capacitance value based on the compensated capacitance value, a pre-calibrated capacitance-ice-thickness conversion model, or a capacitance-ice-thickness mapping table.
[0041] In one embodiment, the information fusion module is further configured to: set the prior probability of icing state based on historical data; determine the first likelihood of the visual analysis result and the second likelihood of the capacitance measurement result; calculate the posterior probability of different icing states based on the icing area, icing type, thickness estimate, and local ice thickness value at each capacitance sensor mounting point according to Bayes' theorem; take the icing state with the highest posterior probability as the final fusion decision output, and take the maximum posterior probability value as the confidence level of icing; and determine the average ice thickness and maximum ice thickness based on the local ice thickness value at each capacitance sensor mounting point.
[0042] In a specific embodiment, such as Figure 2 As shown in the diagram, the visual sensor pathway includes: acquiring raw images of the wind turbine blades using a visual sensor; extracting the Region of Interest (ROI) from the raw images to locate the blade area; and then using visual algorithms to analyze the ROI area for icing. Specifically, this includes judging the icing texture of the ROI area to determine the icing type and performing image segmentation to calculate the icing area. Contour analysis can also be performed to determine the estimated thickness of the wind turbine blades. The capacitance sensor pathway includes: first, acquiring raw capacitance data at multiple key locations using a capacitance sensor array deployed at key locations on the wind turbine blades; then performing data preprocessing on the raw capacitance data, including temperature compensation, filtering and noise reduction, and calibration; and finally, determining the local ice thickness value corresponding to the compensated capacitance value based on the compensated capacitance value and a pre-calibrated capacitance-ice-thickness conversion model or capacitance-ice-thickness mapping table.
[0043] Further, the process can proceed to the fusion decision-making stage. Specifically, the information fusion module receives data processed by the vision subsystem and the capacitive sensor subsystem, including icing area, icing type, thickness estimate, and local ice thickness values at each capacitive sensor mounting point. Then, the information fusion module performs fusion analysis based on a Bayesian inference model to ultimately determine the icing data for the wind turbine blades. This icing data includes at least the icing state, average ice thickness, maximum ice thickness, and confidence level.
[0044] Specifically, the information fusion module can employ a fusion algorithm based on Bayesian inference. Figure 3 The Bayesian inference flowchart is shown. The thickness estimate output by the vision subsystem is compared and confidence-evaluated with the precise thickness value output by the capacitive sensor. Simultaneously, by integrating icing area information, the joint icing probability and final ice thickness are calculated, where the assumed icing probability is: .
[0045] Where P(V|I) is the likelihood of the visual sensor, which is the prior probability of determining whether there is ice or not under a certain ice thickness; P(C|I) is the likelihood of the capacitive sensor. As a normalization factor, it is necessary to calculate all states. The numerators and summation:
[0046] Output: The output is the final decision result after fusion, including the overall icing status, average ice thickness, maximum ice thickness, and icing distribution confidence level. This provides a control basis for the wind turbine de-icing system and operation strategy. Specifically, the final decision result can be transmitted to the SCADA system to trigger de-icing and power control of the de-icing device to initiate the de-icing operation. SCADA stands for Supervisory Control and Data Acquisition.
[0047] Specifically, when fusing and analyzing the data obtained from the vision subsystem and the capacitive sensor subsystem, a pre-established Bayesian network model can be used to calculate the confidence levels of the information provided by different sensors under different ice conditions. For example, if the visually estimated thickness is highly consistent with the capacitive thickness, the confidence levels of both will increase; if there is a conflict (such as visual misjudgment due to heavy fog), the system will automatically reduce the weight of the visual data. Then, all input information and their confidence levels can be combined, and the final joint icing state can be calculated according to the weighted decision rules. The final output decision is: generating a fusion result including the overall icing state, average ice thickness, maximum ice thickness, and system confidence level. The specific algorithm is as follows: 1. Parameter Definition Set discrete probability values, icing conditions, sensor parameters, and the output of the capacitive sensor array, etc. The icing conditions include: Ice-free Light icing (frost) Severe icing (visible ice). Sensor parameter V represents the output of the vision sensor. This is indicated by visual judgment as "no ice". Visual judgment: "There is ice." The output of the capacitive sensor array can... , .in For average thickness <1mm, The average thickness is >= 1 mm.
[0048] 2. Construct a Bayesian network Prior probability P(I): Based on historical meteorological data and operational records of the wind field, the frequency of occurrence of various states is statistically analyzed, P( P( P( The options can be customized based on prior knowledge, namely no ice, light ice, and heavy ice. Calculate the likelihoods P(V|I) and P(C|I) (conditional probability table) of the visual sensor and the capacitive array under icing conditions to describe the reliability of different sensors under different real-world conditions (this data was obtained through laboratory experiments and learning and statistics of historical data). The likelihood probability table P(V|I) for visual sensors and the likelihood probability table P(C|I) for capacitive sensors are shown in Tables 1 and 2 below: Table 1
[0049] Table 2
[0050] 3. Fusion reasoning When the system is running, the sensor data received in real time serves as new evidence, and the posterior probability of all possible states needs to be calculated: Calculate P( | V= ) and P( | V= ), P( | ) to calculate P( | For example, the result is obtained by looking up the table above:
[0051] =
[0052] Normalize the results as follows:
[0053]
[0054] =
[0055] =
[0056] Where P( ) is the normalization factor, which needs to be calculated for all states. , , The numerators of the equations are summed. All posterior probabilities are compared, and the state with the highest posterior probability is selected as the output of the final fusion decision. Then, the final decision result is uploaded to the wind farm SCADA (Supervisory Control and Data Acquisition) system or the central control room via a communication interface.
[0057] Based on this result, the SCADA system can automatically or prompt the administrator to trigger the de-icing device (such as electric / gas-thermal de-icing) and adjust the fan's electric / gas-thermal de-icing system, thereby forming a complete "perception-decision-control" closed loop.
[0058] In one embodiment, a wind turbine blade icing status monitoring device (not shown in the figure) based on multi-level information fusion is provided, comprising: The memory is configured to store instructions; The processor is configured to retrieve instructions from memory and, when executing the instructions, to implement the wind turbine blade icing status monitoring method based on multi-level information fusion according to any of the above embodiments.
[0059] The processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and adjusting kernel parameters enables a multi-level information fusion-based method for monitoring the icing status of wind turbine blades.
[0060] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.
[0061] This application provides a storage medium storing a program that, when executed by a processor, implements the above-described method for monitoring the icing status of wind turbine blades based on multi-level information fusion.
[0062] This application provides a processor for running a program, wherein the program executes the above-described method for monitoring the icing status of wind turbine blades based on multi-level information fusion.
[0063] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor A01, a network interface A02, a memory (not shown), and a database (not shown) connected via a system bus. The processor A01 provides computing and control capabilities. The memory includes internal memory A03 and a non-volatile storage medium A04. The non-volatile storage medium A04 stores an operating system B01, a computer program B02, and a database (not shown). The internal memory A03 provides an environment for the operation of the operating system B01 and the computer program B02 stored in the non-volatile storage medium A04. The network interface A02 is used for communication with external terminals via a network connection. When the computer program B02 is executed by the processor A01, it implements a method for monitoring the icing status of wind turbine blades based on multi-level information fusion.
[0064] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0065] This application provides a computer (electronic) device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of any of the above-mentioned methods for monitoring the icing status of wind turbine blades based on multi-level information fusion.
[0066] This application also provides a computer program product that, when executed on a data processing device, is suitable for executing a program that initializes a method for monitoring the icing status of wind turbine blades based on multi-level information fusion.
[0067] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0068] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0069] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0070] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0071] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0072] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.
[0073] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.
[0074] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.
[0075] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.
Claims
1. A method for monitoring the icing status of wind turbine blades based on multi-level information fusion, characterized in that, The method includes: Simultaneously acquire raw images of the wind turbine blades and raw capacitance data collected based on a capacitive sensor array deployed at key locations on the wind turbine blades, the capacitive sensor array including capacitive sensors installed at multiple key locations; The original image is subjected to blade region extraction and icing analysis to determine the icing area, icing type, and thickness estimate of the wind turbine blades. Temperature compensation and calibration are performed on the raw capacitance data to determine the local ice thickness value at each capacitance sensor mounting point; Based on a Bayesian inference model, the icing area, icing type, thickness estimate, and local ice thickness value at each capacitive sensor installation point are fused and analyzed to finally determine the icing data of the wind turbine blades. The icing data includes at least the icing state, average ice thickness, maximum ice thickness, and confidence level.
2. The method according to claim 1, characterized in that, The leaf region extraction and icing analysis of the original image includes: A deep learning-based target detection algorithm is used to identify and extract the leaf region as the region of interest from the original image; The region of interest is subjected to icing texture analysis to determine the icing type, and image segmentation is performed to calculate the icing area. The baseline profile of the wind turbine blades in the non-icing state is obtained in advance; Extract the current outline of the wind turbine blades from the original image; The distance between the current profile and the reference profile in the normal vector direction is determined to determine the thickness estimate.
3. The method according to claim 2, characterized in that, The target detection algorithm is the YOLO model, which uses transfer learning to adjust the model parameters of the YOLO model based on pre-trained weights.
4. The method according to claim 1, characterized in that, Temperature compensation and calibration of the raw capacitance data includes: The original capacitance data is compensated based on the readings from the built-in temperature sensor and a pre-calibrated temperature-capacitance relationship curve. Based on the compensated capacitance value, the pre-calibrated capacitance-ice-thickness conversion model, or the capacitance-ice-thickness mapping table, the local ice-thickness value corresponding to the compensated capacitance value is determined.
5. The method according to claim 1, characterized in that, The original image was acquired using an image acquisition device. Based on a Bayesian inference model, a fusion analysis was performed on the ice coverage area, ice type, thickness estimate, and local ice thickness value at each capacitive sensor installation point, including: The prior probability of icing status is set based on historical data; Determine the first likelihood of the visual analysis results and the second likelihood of the capacitance measurement results; Based on the ice-covered area, ice-covered type, thickness estimate, and local ice thickness at each capacitive sensor installation point, the posterior probability of different ice-covered states is calculated according to Bayes' theorem. The icing state with the highest posterior probability is taken as the final fusion decision output, and the highest posterior probability value is taken as the confidence level of icing. The average ice thickness and the maximum ice thickness are determined based on the local ice thickness value at each capacitive sensor installation point.
6. An online monitoring system for the icing status of wind turbine blades based on multi-level information fusion, characterized in that, The system includes: The vision subsystem is used to acquire raw images of wind turbine blades and perform blade region extraction and icing analysis on the raw images to determine the icing area, icing type, and thickness estimate of the wind turbine blades. The capacitive sensor subsystem includes a capacitive sensor array deployed at key locations on the wind turbine blades, used to collect raw capacitance data at multiple key locations and perform temperature compensation and calibration on the raw capacitance data to determine the local ice thickness value at each capacitive sensor installation point. The information fusion module is communicatively connected to the vision subsystem and the capacitive sensor subsystem. It is used to receive the icing area, icing type, thickness estimate, and local ice thickness value at each capacitive sensor installation point. Based on the Bayesian inference model, it performs fusion analysis to finally determine the icing data of the wind turbine blades. The icing data includes at least the icing state, average ice thickness, maximum ice thickness, and confidence level.
7. The system according to claim 6, characterized in that, The vision subsystem includes: Image acquisition module, used to acquire raw images containing the leaves; The ROI extraction module uses a deep learning-based target detection algorithm to identify and extract leaf regions as regions of interest from the original image. The icing analysis module is used to perform icing texture analysis on the region of interest to determine the icing type, and to perform image segmentation to calculate the icing area; to obtain the reference profile of the wind turbine blade in the non-icing state in advance; to extract the current profile of the wind turbine blade from the original image; and to determine the distance between the current profile and the reference profile in the normal vector direction to determine the thickness estimate.
8. The system according to claim 6, characterized in that, The capacitive sensor subsystem includes: A capacitive sensor array is deployed at the leading edge and tip of the blade. The signal processing module is used to perform temperature compensation, filtering and noise reduction, and calibration on the raw capacitance data; The local ice thickness determination module is used to determine the local ice thickness value corresponding to the compensated capacitance value based on the compensated capacitance value, a pre-calibrated capacitance-ice thickness conversion model, or a capacitance-ice thickness mapping table.
9. The system according to claim 6, characterized in that, The information fusion module is also used for: The prior probability of icing status is set based on historical data; Determine the first likelihood of the visual analysis results and the second likelihood of the capacitance measurement results; Based on the ice-covered area, ice-covered type, thickness estimate, and local ice thickness at each capacitive sensor installation point, the posterior probability of different ice-covered states is calculated according to Bayes' theorem. The icing state with the highest posterior probability is taken as the final fusion decision output, and the highest posterior probability value is taken as the confidence level of icing. The average ice thickness and the maximum ice thickness are determined based on the local ice thickness value at each capacitive sensor installation point.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 5.