Fan power estimation method and system based on fan identification model

Abstract

The invention discloses a fan power estimation method and system based on a fan identification model, and belongs to the technical field of new energy. A fan recognition model is constructed based on a YOLOv10 algorithm framework and a multi-scale attention mechanism, fan recognition is automatically monitored through the fan recognition model, the wind speed is reversely deduced through the angular speed and the tip speed ratio, then the fan power is estimated in real time, and the method is suitable for the scene where a wind speed sensor cannot be directly used. The operation and maintenance efficiency can be improved, the service life of equipment is prolonged, the maintenance cost is reduced, the safety of the equipment is improved, and dangerous operation of operation and maintenance personnel under severe weather conditions is avoided.

Description

Technical Field

[0001] The present invention relates to the field of new energy technology, and more specifically to a method and system for estimating wind turbine power based on a wind turbine identification model. Background Art

[0002] With the rapid development of economy, new energy has gradually become a substitute for non-renewable resources. As an important component of renewable energy, wind energy has played an increasingly important role in the global energy structure. Wind power generation converts the kinetic energy of wind into electrical energy through wind turbines. The estimation of wind turbine power is crucial for optimizing energy distribution, improving wind power generation efficiency and ensuring stable operation of the power grid.

[0003] The estimation of wind turbine power depends on wind speed. The traditional method measures wind speed through wind speed sensors. However, some wind farm sites cannot directly use wind speed sensors. The lack of real-time wind speed data leads to a decrease in the accuracy of wind turbine power, making it difficult to ensure the effective use of energy and the stable operation of the power grid.

[0004] Therefore, how to propose a wind turbine power estimation method and system based on a wind turbine identification model that can realize automated wind turbine power estimation while reducing operation and maintenance costs is an urgent problem that technical personnel in this field need to solve. Summary of the invention

[0005] In view of this, the present invention provides a wind turbine power estimation method and system based on a wind turbine identification model, which can realize low-cost identification of wind turbines and estimation of the current operating power of wind turbines based on the YOLOv10 algorithm framework, multi-scale attention mechanism and image processing algorithm.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] On the one hand, the present invention discloses a method for estimating wind turbine power based on a wind turbine identification model, comprising the following steps:

[0008] Collecting fan image data and annotating the fan image data to construct a fan identification data set;

[0009] Construct a wind turbine identification model, wherein the wind turbine identification model adds a multi-scale attention mechanism network to the YOLOv10 network framework;

[0010] The fan recognition model is trained using the fan recognition data set, and the real-time fan image data is input into the trained fan recognition model to obtain a semantic recognition result;

[0011] According to the semantic recognition result, an ID is assigned to the fan, the edge of the fan is detected, the movement trajectory of the fan blades is tracked, and the angular velocity of the fan rotation is obtained;

[0012] The wind speed is inversely calculated by combining the angular velocity with the tip speed ratio of the wind turbine, and the wind turbine power is estimated based on the wind speed.

[0013] Preferably, the YOLOv10 network framework includes a backbone network, a neck network and a head network connected in sequence;

[0014] The multi-scale attention mechanism network is applied to the head network and the neck network.

[0015] Preferably, the multi-scale attention mechanism network includes a feature extraction module, an attention module, and a detection module connected in sequence.

[0016] Preferably, using the wind turbine identification data set to train the wind turbine identification model includes:

[0017] The wind turbine identification data set includes manually annotated real boxes, and the intersection-over-union ratio between the real boxes and the detection boxes output by the wind turbine identification model is calculated;

[0018] Set a threshold, and input the detection frames whose IoU ratio exceeds the threshold into the second stage; if the IoU ratios do not exceed the threshold, then input the detection frame corresponding to the largest IoU ratio into the second stage; the second stage uses the non-maximum suppression algorithm to eliminate redundant detection frames to obtain the final detection result frame;

[0019] By comparing the coordinate data of the final detection result frame with the real frame, a relationship is established so that the output detection result frame coordinate data is equal to the coordinate data of the real result frame.

[0020] Preferably, the rotation process of the fan blades is tracked according to the optical flow pyramid method to obtain the motion trajectory of the fan blades;

[0021] The semantic recognition result includes a semantic frame, and a virtual line is set at 1 / 3 of the semantic frame, and the virtual line is used to capture the time difference t of the fan blade passing through the virtual line during the rotation process and the angle θ passed by the fan blade within the time difference t;

[0022] The calculation formula of the angular velocity ω is as follows:

[0023]

[0024] Preferably, the calculation formula of the tip speed ratio λ is:

[0025]

[0026] Where R is the rotation radius of the fan blade, which is a fixed parameter; V ω is the wind speed.

[0027] Preferably, the estimation formula of the fan power P is as follows:

[0028] P=0.5×ρ×A×V 3 ω ×C p ;

[0029] Where ρ is the air density; A is the swept area of ​​the fan, A = πR 2 ; C p is the fan power factor.

[0030] On the other hand, the present invention further discloses a wind turbine power estimation system based on a wind turbine identification model, which is used to implement the above-mentioned wind turbine power estimation method, comprising:

[0031] A data processing module collects fan image data and annotates the fan image data to construct a fan identification data set;

[0032] A model building module is used to build a wind turbine identification model, wherein the wind turbine identification model adds a multi-scale attention mechanism network to the YOLOv10 network framework;

[0033] A training and output module, which uses the fan identification data set to train the fan identification model, and inputs the real-time fan image data into the trained fan identification model to obtain a semantic recognition result;

[0034] An angular velocity calculation module assigns an ID to the fan according to the semantic recognition result, detects the edge of the fan, tracks the motion trajectory of the fan blades, and obtains the angular velocity of the fan rotation;

[0035] The wind turbine power estimation module uses the angular velocity combined with the tip speed ratio of the wind turbine to inversely estimate the wind speed, and estimates the wind turbine power based on the wind speed.

[0036] Through the above technical solutions, it can be known that compared with the prior art, the present invention discloses a method and system for estimating wind turbine power based on a wind turbine identification model, constructs a wind turbine identification model based on the YOLOv10 algorithm framework and a multi-scale attention mechanism, uses the wind turbine identification model to automatically monitor wind turbine identification, and reversely infers wind speed through angular velocity and tip speed ratio, thereby estimating wind turbine power in real time, which is suitable for scenarios where wind speed sensors cannot be used directly. The present invention can improve operation and maintenance efficiency, extend equipment life, reduce maintenance costs, improve equipment safety, and avoid dangerous operations of operation and maintenance personnel under severe weather conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0038] Figure 1 A flow chart of a wind turbine power estimation method based on a wind turbine identification model provided by the present invention;

[0039] Figure 2 This is an overall structural diagram of the wind turbine identification model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] On the one hand, the embodiment of the present invention discloses a method for estimating wind turbine power based on a wind turbine identification model, referring to Figure 1 , the method comprises the following steps:

[0042] S1. Collect and annotate fan image data to build a fan identification dataset.

[0043] We collected different types of wind turbine image data, annotated wind turbines with specific semantic categories, and built a wind turbine identification dataset. The data collection covered different time periods, weather and environmental conditions to ensure data diversity.

[0044] When annotating wind turbine image data containing specific semantic categories, an annotation tool is used to annotate the collected wind turbine image data, and the wind turbine is selected by a rectangular frame and annotated with a specific label windturbine of the wind turbine.

[0045] S2. Construct a fan identification model, such as Figure 2 As shown in the figure, the wind turbine recognition model adds a multi-scale attention mechanism network to the YOLOv10 network framework.

[0046] The YOLOv10 network framework extracts the features of the input wind turbine image data, including the backbone network (Backbone), neck network (Neck) and head network (Head) connected in sequence. The process of generating the feature map includes:

[0047] Image data is used as input and passes through the neural network layers of Backbone, Neck, and Head for feature extraction. The main function of Backbone is to extract features from the input image. It gradually reduces the image size through multiple convolutional layers and extracts feature representations at different scales; Neck is used to fuse features from Backbone at different scales to improve the detection effect of small objects, medium objects, and large objects. Through the feature pyramid, Neck can capture and fuse information from multi-scale feature maps, thereby providing more comprehensive features for subsequent Head. Head is responsible for outputting the final detection results, namely bounding boxes and category predictions, based on the multi-scale feature maps provided by Neck.

[0048] A multi-scale attention mechanism network (BiFormer) is applied to the head network and the neck network, and the multi-scale attention mechanism network includes a feature extraction module, an attention module, and a detection module connected in sequence.

[0049] The feature extraction module is used to extract the features of the input image data and generate a feature image. The attention module can use the results extracted by the feature extraction module to highlight the detection object in the image. The detection module is used to detect the processing results of the attention module and obtain the fan feature information in the input image data.

[0050] Figure 2 The overall architecture of the wind turbine recognition model is shown. The backbone network includes PSA, SPPF, C2f_SA, SC Down and Conv modules, which are responsible for extracting multi-scale features from the input image. The neck network includes Upsample, Concat, BiFormer, SC Down and GhostConv modules. Through multi-scale feature fusion, the features generated by the backbone network are further integrated to meet the detection needs of small, medium and large scale targets. The head network contains three scale detection layers: large, medium and small. It is composed of C2fCIB and multiple BiFormer modules. Feature extraction and aggregation are performed at each scale to enhance the detection ability of the model at different scales. The BiFormer module is used in both the neck and head networks. Through the effective combination and enhancement of multi-scale features, it is seamlessly integrated with the YOLOv10 network to improve the detection performance of various targets.

[0051] BiFormer's feature extraction module, attention module, and detection module play a key role in the model. The feature extraction module integrates information from different levels of the image through multi-scale feature fusion to capture rich spatial and semantic information. The attention module introduces a self-attention mechanism to adaptively assign higher weights to important features and filter out irrelevant information, thereby enhancing the expressiveness and anti-interference ability of the features. The detection module is used in detection layers of different scales, combined with the output of the previous feature extraction and attention modules, to effectively improve the detection accuracy of small, medium, and large-scale targets. This modular design enables BiFormer to be efficiently integrated into YOLOv10, significantly improving the detection performance of the model.

[0052] S3. Use the fan recognition data set to train the fan recognition model, and input the real-time fan image data into the trained fan recognition model to obtain the semantic recognition result.

[0053] The wind turbine identification model is trained using the wind turbine identification dataset, including:

[0054] The wind turbine identification dataset includes manually annotated true boxes, and the intersection-over-union ratio between the true boxes and the detection boxes output by the wind turbine identification model is calculated;

[0055] Set a threshold, and input the detection frames whose IoU ratio exceeds the threshold into the second stage; if the IoU ratios do not exceed the threshold, then input the detection frame corresponding to the largest IoU ratio into the second stage; the second stage uses the non-maximum suppression algorithm to eliminate redundant detection frames to obtain the final detection result frame;

[0056] By comparing the coordinate data of the final detection result frame with the real frame, a relationship is established so that the output detection result frame coordinate data is equal to the coordinate data of the real result frame.

[0057] S4. Assign an ID to the fan according to the semantic recognition result, detect the edge of the fan, track the movement trajectory of the fan blades, and obtain the angular velocity of the fan rotation.

[0058] Each identified wind turbine is assigned a unique ID.

[0059] The edge image of the fan is identified through the frame difference method and the Canny edge detection algorithm, and the motion trajectory of the fan blades is tracked using the optical flow pyramid technology. The optical flow technology can detect the motion changes of pixels in the image sequence, and obtain the motion direction and speed of each pixel in the continuous frames through analysis.

[0060] The semantic recognition result includes a semantic box. A virtual line is set at 1 / 3 of the semantic box. The virtual line is used to capture the time difference t when the fan blade passes through the virtual line during the rotation process and the angle θ that the fan blade passes within the time difference t.

[0061] The calculation formula of angular velocity ω is as follows:

[0062]

[0063] T is the time period for the fan blade to rotate one circle. Usually, the complete period can be calculated by observing the time difference t of multiple fan blades passing through the virtual line.

[0064] S5. The wind speed is inferred from the angular velocity combined with the tip speed ratio of the wind turbine, and the wind turbine power is estimated based on the wind speed.

[0065] The tip speed ratio is the ratio of the linear velocity at the tip of the fan blade to the wind speed. It reflects the matching of the fan blade rotation speed with the ambient wind speed and is an important parameter for estimating fan performance.

[0066] The calculation formula of tip speed ratio λ is:

[0067]

[0068] Where R is the rotation radius of the fan blade. The rotation radius R of the fan defines the length of the fan blade. It is usually a fixed value when the fan is designed. In actual scenarios, it can be obtained from the technical specifications of the fan. V ω is the wind speed.

[0069] Inferring wind speed through angular velocity and tip speed ratio is a non-traditional method of wind speed estimation, which gets rid of the dependence on physical sensors.

[0070] Different types of wind turbines have different optimal tip speed ratios, usually ranging from 6 to 8. Too high a tip speed ratio will lead to reduced wind energy utilization efficiency, while too low a tip speed ratio may cause the wind turbine to operate inefficiently.

[0071] Power output and power coefficient C of fan p The power coefficient is a measure of the efficiency of a wind turbine in converting wind energy into mechanical energy. According to Betz's Law, the maximum power coefficient of a wind turbine is theoretically 0.593, which means that the wind turbine can only convert 59.3% of wind energy into mechanical energy. p , usually between 0.4 and 0.5.

[0072] The estimated formula for fan power P is as follows:

[0073]

[0074] Where ρ is the air density; A is the swept area of ​​the fan, A = πR 2 ; C p is the fan power factor.

[0075] On the other hand, the present invention further discloses a wind turbine power estimation system based on a wind turbine identification model, which is used to implement the above-mentioned wind turbine power estimation method, comprising:

[0076] The data processing module collects and annotates the fan image data to construct a fan identification data set;

[0077] Model building module, building a wind turbine identification model. The wind turbine identification model adds a multi-scale attention mechanism network to the YOLOv10 network framework;

[0078] The training and output module uses the fan recognition data set to train the fan recognition model, and inputs the real-time fan image data into the trained fan recognition model to obtain the semantic recognition result;

[0079] The angular velocity calculation module assigns an ID to the fan according to the semantic recognition result, detects the fan edge, tracks the movement trajectory of the fan blades, and obtains the angular velocity of the fan rotation;

[0080] The wind turbine power estimation module uses the angular velocity combined with the tip speed ratio of the wind turbine to infer the wind speed and estimates the wind turbine power based on the wind speed.

[0081] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0082] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for estimating wind turbine power based on a wind turbine identification model, characterized in that: The following steps are involved: Collecting fan image data and annotating the fan image data to construct a fan identification data set; Construct a wind turbine identification model, wherein the wind turbine identification model adds a multi-scale attention mechanism network to the YOLOv10 network framework; The fan recognition model is trained using the fan recognition data set, and the real-time fan image data is input into the trained fan recognition model to obtain a semantic recognition result; According to the semantic recognition result, an ID is assigned to the fan, the edge of the fan is detected, the movement trajectory of the fan blades is tracked, and the angular velocity of the fan rotation is obtained; The wind speed is inversely calculated by combining the angular velocity with the tip speed ratio of the wind turbine, and the wind turbine power is estimated based on the wind speed.

2. A method for estimating wind turbine power based on a wind turbine identification model according to claim 1, characterized in that: The YOLOv10 network framework includes a backbone network, a neck network and a head network connected in sequence; The multi-scale attention mechanism network is applied to the head network and the neck network.

3. A method for estimating wind turbine power based on a wind turbine identification model according to claim 2, characterized in that: The multi-scale attention mechanism network includes a feature extraction module, an attention module, and a detection module connected in sequence.

4. The method for estimating wind turbine power based on a wind turbine identification model according to claim 1, characterized in that: Using the wind turbine identification data set to train the wind turbine identification model includes: The wind turbine identification data set includes manually annotated real boxes, and the intersection-over-union ratio between the real boxes and the detection boxes output by the wind turbine identification model is calculated; Set a threshold, and input the detection frames whose IoU ratio exceeds the threshold into the second stage; if the IoU ratios do not exceed the threshold, then input the detection frame corresponding to the largest IoU ratio into the second stage; the second stage uses the non-maximum suppression algorithm to eliminate redundant detection frames to obtain the final detection result frame; By comparing the coordinate data of the final detection result frame with the real frame, a relationship is established so that the output detection result frame coordinate data is equal to the coordinate data of the real result frame.

5. The method for estimating wind turbine power based on a wind turbine identification model according to claim 1, characterized in that: The rotation process of the fan blades is tracked according to the optical flow pyramid method to obtain the motion trajectory of the fan blades; The semantic recognition result includes a semantic frame, and a virtual line is set at 1 / 3 of the semantic frame, and the virtual line is used to capture the time difference t of the fan blade passing through the virtual line during the rotation process and the angle θ passed by the fan blade within the time difference t; The calculation formula of the angular velocity ω is as follows:

6. A method for estimating wind turbine power based on a wind turbine identification model according to claim 5, characterized in that: The calculation formula of tip speed ratio λ is: Where R is the rotation radius of the fan blade, which is a fixed parameter; V ω is the wind speed.

7. A method for estimating wind turbine power based on a wind turbine identification model according to claim 6, characterized in that: The estimated formula for fan power P is as follows: Where ρ is the air density; A is the swept area of ​​the fan, A = πR 2 ; C p is the fan power factor.

8. A wind turbine power estimation system based on a wind turbine identification model, characterized in that: include: A data processing module collects fan image data and annotates the fan image data to construct a fan identification data set; A model building module is used to build a wind turbine identification model, wherein the wind turbine identification model adds a multi-scale attention mechanism network to the YOLOv10 network framework; A training and output module, which uses the fan identification data set to train the fan identification model, and inputs the real-time fan image data into the trained fan identification model to obtain a semantic recognition result; An angular velocity calculation module assigns an ID to the fan according to the semantic recognition result, detects the edge of the fan, tracks the motion trajectory of the fan blades, and obtains the angular velocity of the fan rotation; The wind turbine power estimation module uses the angular velocity combined with the tip speed ratio of the wind turbine to inversely estimate the wind speed, and estimates the wind turbine power based on the wind speed.

Images