Wind power equipment state identification method and device, computer equipment and storage medium

Abstract

The invention discloses a wind power equipment state identification method and device, computer equipment and a storage medium, and belongs to the technical field of artificial intelligence, so as to facilitate enabling financial risk control. The method comprises the following steps: firstly, combining texture features of an optical image with geometric features of a synthetic aperture radar image to construct a complementary fusion feature set; and local enhancement is carried out by adopting an attention mechanism, so that the recognition precision of the cloud shielding region is remarkably improved. By introducing terrain difference data, adaptive adjustment is performed on the enhanced features, so that the model can flexibly adapt to different terrain fluctuations and spatial distribution. Subsequently, the feature dimension is reduced through quantitative compression, the calculation and storage burden of the model is relieved, and efficient deployment on edge equipment is facilitated. And in combination with a transfer learning framework, rapid fine adjustment and adaptation of the model in a new area are realized, the model training period is remarkably shortened, and the recognition accuracy is improved. According to the invention, high precision, low delay and strong robustness of wind power equipment state identification are realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of artificial intelligence, and particularly relates to a wind power equipment state recognition method and device, computer equipment and a storage medium. BACKGROUND

[0002] As a core pillar of clean energy, wind power plays a crucial role in global energy transformation. Its efficient utilization and precise management are directly related to the stability of the energy system and the realization of sustainable development goals. However, the current identification and monitoring technology of wind turbines still has significant shortcomings in adaptability and accuracy under complex environments, limiting its application in diverse scenarios. Existing methods often rely on a single data source or fixed model structure, making it difficult to cope with changing weather conditions and terrain differences, resulting in unstable recognition results, especially in cloud cover or complex terrain areas, with frequent false positives and false negatives. In addition, traditional methods are inefficient in processing large-scale data, making it difficult to meet the needs of low-power deployment of edge devices and unable to achieve dynamic equipment state analysis.

[0003] In practical applications, the core challenge of wind turbine identification comes from the inefficiency of cross-modal data fusion and the insufficient adaptability of models to complex scenarios. The inefficiency of cross-modal data fusion is that optical images and synthetic aperture radar images are difficult to complement each other in feature extraction and integration, resulting in the system being unable to accurately capture the spatial position and structural features of wind turbines in cloudy weather. For example, in coastal cloudy areas, traditional methods frequently miss detection due to limited optical images, affecting the accuracy of wind farm layout planning. This inefficiency further limits the model's ability to migrate between different regions, especially in scenarios with large differences in terrain or climate, requiring retraining of the model, increasing deployment costs and time.

[0004] Therefore, how to design an efficient multi-source data fusion method that dynamically adapts to complex environments and achieves low computational load on edge devices has become a key issue in promoting the development of wind power intelligent management. SUMMARY

[0005] The purpose of the embodiments of the present application is to propose a wind power equipment state recognition method, device, computer equipment and storage medium, to design an efficient multi-source data fusion method that dynamically adapts to complex environments and achieves low computational load on edge devices.

[0006] To solve the above technical problems, the embodiments of the present application provide a wind power equipment state recognition method, which adopts the following technical solutions:

[0007] A wind power equipment state recognition method, comprising:

[0008] optical images and synthetic aperture radar images of the wind power equipment as multi-source data, extracting texture features from the optical images, and extracting geometric features from the synthetic aperture radar images to obtain a preliminary feature set;

[0009] determining a complementary feature pair in the preliminary feature set, integrating the complementary feature pair to obtain a fusion feature set, wherein each complementary feature pair includes a texture feature and a geometric feature;

[0010] determining cloud layer shielding area features from the fusion feature set, and locally enhancing the cloud layer shielding area using an attention mechanism to obtain an enhanced feature set;

[0011] extracting terrain difference data from the multi-source data, and adaptively adjusting the enhanced feature set using the terrain difference data to obtain an adaptive feature set, wherein the terrain difference data reflects the undulating change and spatial distribution characteristics of the terrain where the wind power equipment is located;

[0012] quantitatively compressing the adaptive feature set to obtain a compressed feature set, and delivering the compressed feature set to a wind power equipment state recognition model that is pre-deployed at the edge;

[0013] using a transfer learning framework to fine-tune the wind power equipment state recognition model based on the compressed feature set until the model is fitted, and outputting a state recognition result of the wind power equipment.

[0014] To solve the above technical problems, the embodiment of the application also provides a wind power equipment state recognition device, which adopts the technical scheme as follows:

[0015] A wind power equipment state recognition device comprises:

[0016] a multi-source data module configured to obtain optical images and synthetic aperture radar images of the wind power equipment as multi-source data, extract texture features from the optical images, and extract geometric features from the synthetic aperture radar images to obtain a preliminary feature set;

[0017] a feature complementarity module configured to determine a complementary feature pair in the preliminary feature set, and integrate the complementary feature pair to obtain a fusion feature set, wherein each complementary feature pair includes a texture feature and a geometric feature;

[0018] a feature enhancement module configured to determine cloud layer shielding area features from the fusion feature set, and locally enhance the cloud layer shielding area using an attention mechanism to obtain an enhanced feature set;

[0019] An adaptive adjustment module is configured to extract terrain difference data from the multi-source data, and to use the terrain difference data to adaptively adjust the enhanced feature set to obtain an adaptive feature set, wherein the terrain difference data reflects the fluctuation and spatial distribution characteristics of the terrain where the wind power equipment is located.

[0020] A quantization compression module is configured to quantize and compress the adaptive feature set to obtain a compressed feature set, and to distribute the compressed feature set to the wind power equipment state recognition model that is pre-deployed at the edge.

[0021] A model fine-tuning module is configured to use a transfer learning framework to fine-tune the parameters of the wind power equipment state recognition model based on the compressed feature set until the model is fitted, and to output the state recognition result of the wind power equipment.

[0022] To solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the technical scheme as follows:

[0023] A computer device includes a memory and a processor, the memory stores computer readable instructions, and the processor executes the computer readable instructions to realize the steps of the wind power equipment state recognition method according to any one of the above.

[0024] To solve the above technical problems, the embodiment of the present application further provides a computer readable storage medium, which adopts the technical scheme as follows:

[0025] A computer readable storage medium stores computer readable instructions, and the computer readable instructions are executed by a processor to realize the steps of the wind power equipment state recognition method according to any one of the above.

[0026] Compared with the prior art, the embodiment of the present application has the following beneficial effects:

[0027] This application discloses a method, apparatus, computer equipment, and storage medium for wind power equipment status identification, belonging to the field of artificial intelligence technology, to empower financial risk control. First, a complementary fusion feature set is constructed by combining the texture features of optical images and the geometric features of synthetic aperture radar images. For influencing factors such as cloud cover, an attention mechanism is used for local enhancement, significantly improving the identification accuracy in cloud-covered areas. By introducing terrain difference data, the enhanced features are adaptively adjusted, enabling the model to flexibly adapt to different terrain undulations and spatial distributions. Subsequently, quantization compression reduces the feature dimensionality, alleviating the computational and storage burden of the model and facilitating efficient deployment on edge devices. Combined with a transfer learning framework, rapid fine-tuning and adaptation of the model in new regions are achieved, significantly shortening the model training cycle and improving identification accuracy. This application effectively integrates multimodal data and machine learning algorithms, achieving high accuracy, low latency, and strong robustness in wind power equipment status identification. It is suitable for complex and ever-changing real-world application scenarios, greatly improving the operation and maintenance efficiency and intelligent management level of wind farms. Attached Figure Description

[0028] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 An exemplary system architecture diagram is shown, in which this application can be applied;

[0030] Figure 2 A flowchart of one embodiment of the wind power equipment status identification method according to this application is shown;

[0031] Figure 3 It shows Figure 2 A flowchart of one embodiment of step S202;

[0032] Figure 4 A schematic diagram of a structure of an embodiment of a wind power equipment status identification device according to this application is shown;

[0033] Figure 5 It shows Figure 4 A schematic diagram of a structure of an embodiment of the complementary feature module 402;

[0034] Figure 6 A schematic diagram of the structure of one embodiment of a computer device according to this application is shown. Detailed Implementation

[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0036] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0037] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0038] like Figure 1 As shown, system architecture 100 may include terminal device 101, network 102, and server 103. Terminal device 101 may be a laptop 1011, tablet 1012, or mobile phone 1013. Network 102 is used as a medium to provide a communication link between terminal device 101 and server 103. Network 102 may include various connection types, such as wired, wireless communication links, or fiber optic cables.

[0039] Users can use terminal device 101 to interact with server 103 via network 102 to receive or send messages, etc. Various communication client applications can be installed on terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0040] Terminal device 101 can be various electronic devices with a display screen and support web browsing. In addition to laptops 1011, tablets 1012, or mobile phones 1013, terminal device 101 can also be e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 players (Moving Picture Experts Group Audio Layer IV), laptops, and desktop computers.

[0041] Server 103 can be a server that provides various services, such as a backend server that provides support for the pages displayed on terminal device 101.

[0042] It should be noted that the wind power equipment status identification method provided in this application embodiment is generally executed by a server / terminal device, and correspondingly, the wind power equipment status identification device is generally installed in the server / terminal device.

[0043] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative; the system can have any number of terminal devices, networks, and servers depending on implementation needs.

[0044] Continue to refer to Figure 2 The diagram illustrates a flowchart of an embodiment of a wind turbine status identification method according to this application. The wind turbine status identification method includes the following steps:

[0045] S201: Acquire optical images and synthetic aperture radar images of wind power equipment as multi-source data, extract texture features from optical images and geometric features from synthetic aperture radar images to obtain a preliminary feature set.

[0046] Specifically, optical imagery can be acquired from high-resolution satellites (such as WorldView and Gaofen series), with a preferred resolution of 0.3m-0.5m to ensure that structures such as wind turbine blades and towers are recognizable at the pixel level. Before input, images must undergo radiometric calibration, geometric correction, and atmospheric correction to eliminate radiative noise and geometric distortion. Texture feature extraction can employ a combination of multi-scale directional filters (such as Gabor filters) and gray-level co-occurrence matrices (GLCM) to extract energy, contrast, and homogeneity from high-contrast areas such as wind turbine blades and nacelles. Synthetic Aperture Radar (SAR) imagery can be acquired using platforms such as Sentinel-1 and TerraSAR-X, prioritizing VV or VH polarization modes with a resolution of 1m-3m. SAR data requires radiometric calibration, speckle noise suppression (e.g., based on nonlocal mean filtering (NLM) or guided filtering (GF), and georegistration).

[0047] Geometric feature extraction can utilize edge detection operators (such as Canny and Sobel) combined with Hough transform to identify the straight contours of the tower, or detect the scattering center position of the blade tip through morphological processing. Furthermore, the coherence and backscattering coefficient (σ0) of SAR can be used to characterize the high reflectivity of the metal tower and blades, and the shape, size, and orientation information of the target can be encoded in the feature vector, ultimately forming a preliminary feature set containing both optical texture features and SAR geometric features.

[0048] S202, determine the complementary feature pairs in the preliminary feature set, integrate the complementary feature pairs to obtain the fused feature set, wherein each complementary feature pair includes a texture feature and a geometric feature;

[0049] Specifically, the determination of complementary feature pairs can be accomplished through feature space correlation analysis, that is, by calculating the mutual information or Pearson correlation coefficient between optical texture features and SAR geometric features at the same geographical location to determine their complementarity in terms of information. When optical texture information is strong in a certain area, and SAR geometric features have stable shape information at the same location, the two are combined to form a complementary feature pair.

[0050] In the feature fusion stage, a feature-level fusion strategy can be adopted. Texture and geometric feature vectors are normalized (e.g., Z-score or Min-Max), and then the fused feature vector is obtained through concatenation or weighted summation. The weights can be automatically learned by a lightweight attention network (e.g., a Squeeze-and-Excitation module) in a data-driven manner, dynamically balancing the contributions of optics and SAR during the fusion process. Furthermore, to improve the spatial consistency of the fusion results, multi-scale convolutional operations (e.g., combinations of 1×1, 3×3, and 5×5 convolutional kernels) can be performed on the fused feature map to capture spatial context information at different scales. Bilateral filtering or conditional random fields (CRF) are then used to smooth the fused features, reducing noise interference from cross-modal fusion, ultimately forming a fused feature set.

[0051] S203, after determining the features of the cloud-occluded area by fusing the feature set, an attention mechanism is used to locally enhance the cloud-occluded area to obtain an enhanced feature set;

[0052] Specifically, cloud-occupied areas can be detected by jointly determining the spectral threshold and texture consistency of optical images. For example, the Normalized Cloud Index (NDSI) can be calculated using the blue band and shortwave infrared (SWIR) bands to initially segment bright areas, and texture-deficient areas can be detected through local contrast analysis and multi-scale Laplacian operators. The intersection of these two methods represents potential cloud-occupied areas. Within these areas, the reliability of optical texture features is low, thus requiring compensation using SAR geometric features. The attention mechanism can be implemented using a channel-spatial joint attention (CBAM model), assigning higher weights to SAR features in the channel dimension and locally amplifying the features at the cloud-occupied locations in the spatial dimension.

[0053] In implementation, the fused features can first be segmented using a mask to separate the cloud-occluded and unoccluded regions. Enhancement operators (such as adaptive histogram equalization and local contrast stretching) are then applied to the occluded regions. Finally, the enhanced regions are concatenated with the original unoccluded regions at the feature map level. To avoid pseudo-textures introduced during the enhancement process, regularization constraints (such as L2 norm constraints or feature sparsification) can be introduced after enhancement to control the enhancement intensity, ultimately forming an enhanced feature set containing specific optimizations for the cloud region.

[0054] S204. Extract terrain difference data from multi-source data, and use terrain difference data to adaptively adjust the enhanced feature set to obtain an adaptive feature set. The terrain difference data reflects the undulation and spatial distribution characteristics of the terrain where the wind power equipment is located.

[0055] Specifically, topographic difference data can be obtained from elevation information generated by digital elevation models (DEM) or interferometric SAR (InSAR), and the data resolution can be matched with optical imagery (e.g., 0.5m–1m) to ensure spatial accuracy. Topographic feature extraction can include slope, aspect, elevation difference (ΔH), and local topographic relief index. These features can be used to characterize the geographical environment of the wind turbine, such as mountains, hills, or plains, and participate in the feature adjustment process in the model as environmental condition features.

[0056] During the adaptive adjustment phase, Conditional Batch Normalization (CBR) or Conditional Convolution (CondConv) mechanisms can be employed to dynamically adjust the channel weights of the enhanced feature set based on the terrain category. For example, when the terrain is highly undulating, it may be necessary to increase the weight of geometric features in the overall features to maintain a stable representation of the target structure under image tilt and projection distortion. Furthermore, a Multimodal Conditional Attention Module (MCAM) can be constructed to perform cross-attention calculations between DEM features and enhanced features, allowing terrain information to directly participate in the selection and emphasis of spatial features, thereby obtaining an adaptive feature set optimized for different terrain environments.

[0057] S205, quantize and compress the adaptive feature set to obtain a compressed feature set, and then send the compressed feature set to the wind power equipment status recognition model that has been pre-deployed at the edge.

[0058] Specifically, the quantization compression process needs to consider feature redundancy and the limitations of edge computing resources. Firstly, feature sparsification and low-rank decomposition methods (such as SVD or Tucker decomposition) can be used to reduce the storage dimensionality of feature tensors. Secondly, numerical precision quantization can be performed to compress floating-point features (FP32) to a low-precision format (FP16 or INT8), and quantization-aware training (QAT) can be used to ensure the model maintains consistent feature distribution even at low precision. To avoid precision degradation caused by quantization, feature rearrangement and principal component analysis (PCA) can be performed before quantization to extract the most discriminative principal component features.

[0059] Furthermore, Huffman coding or arithmetic coding can be used to losslessly compress the quantized feature vectors, further reducing the amount of data transmitted. During edge deployment, the model input structure needs to be adapted to ensure that the compressed features can be directly used as model input tensors, and that they are restored to a computable numerical range through decoding and dequantization before model inference. The entire process requires a balance between transmission latency and edge computing power to ensure that the compressed features can be efficiently utilized in resource-constrained environments.

[0060] S206 employs a transfer learning framework to fine-tune the parameters of the wind power equipment status recognition model based on a compressed feature set until the model is fitted, and then outputs the status recognition results of the wind power equipment.

[0061] Specifically, transfer learning frameworks can be built upon pre-trained deep convolutional neural networks (such as YOLOv7, ResNet50, and Swin Transformer), using compressed feature sets as input to fine-tune the network's backend classification or regression modules. Parameter fine-tuning can be achieved by freezing most layers of the backbone network and updating only the detection head and some high-level feature fusion layers to reduce computational overhead and prevent overfitting. During optimization, supervised training can be performed using a small number of labeled samples, while introducing semi-supervised learning strategies for unlabeled samples (such as pseudo-label generation and consistency regularization) to enhance generalization ability. Loss functions can be selected based on the nature of the task, such as cross-entropy loss, Focal Loss (to address class imbalance), and Smooth L1 loss (for state parameter regression).

[0062] Furthermore, domain-adversarial training can be incorporated during fine-tuning to reduce feature distribution differences across different regions or imaging conditions. During training, convergence on the validation set should be monitored, and the training process should be controlled through early stopping and a learning rate scheduler. The final output should include wind power equipment status identification results containing information such as operating status and structural integrity.

[0063] In the embodiments of this application, during the multi-source feature extraction and fusion stages S201 and S202, a scene-aware dynamic data augmentation strategy can be introduced to ensure that the input optical and SAR data have undergone perturbation simulation consistent with the real environment before entering the feature extraction network. For example, leaf motion blur is added to high-resolution samples in optical images, and edge sharpening is added to low-resolution samples to simulate changes in satellite shooting conditions; atmospheric scattering and speckle noise enhancement simulations are introduced in SAR images to improve feature robustness. In the cloud occlusion enhancement stage S203, a cross-modal feature adaptive fusion module (CMFAN) can be embedded to dynamically adjust the weight distribution of optical and SAR in different spatial locations through an attention mechanism, especially automatically increasing the SAR feature contribution in occluded areas and strengthening optical texture features in well-lit areas; in the terrain adaptation stage S204, terrain features can be linked with dynamic modal weights so that the model can prioritize retaining the most discriminative feature channels under different terrain conditions such as elevation differences and slopes, achieving cross-scene adaptive optimization.

[0064] In the S206 model transfer and fine-tuning stage, a semi-supervised pseudo-label generation system can be used to reduce the cost of manual annotation. The specific process is as follows: A wind turbine status recognition model already trained in other regions is used to make preliminary predictions on unlabeled multi-source images. High-quality pseudo-label samples are then selected based on a confidence threshold (e.g., ≥0.8). This is followed by verification using geographic coordinates and spatiotemporal consistency data extracted from S201–S204 (e.g., the positional offset of the same wind turbine in consecutive temporal images does not exceed a certain threshold), eliminating potentially mislabeled data. Validated pseudo-labels can be directly used for fine-tuning training. Simultaneously, a small amount of manually annotated data from the target area (e.g., 100 images per area) can be used as supervised samples. Domain-Adversarial Training is used to align the feature distributions of the source and target domains, enabling the model to quickly adapt to different terrains, imaging conditions, and wind turbine layouts.

[0065] After the S205 compressed feature set is generated, it can be deployed at the edge in conjunction with a lightweight wind power equipment condition recognition model (Lite-WindDet). This model can use depthwise separable convolutions instead of conventional convolutions, and combine channel attention pruning (SE Block + Pruning) to retain key feature channels, reducing the number of parameters by more than 40%. In edge deployment optimization, Quantization-Aware Training (QAT) can be introduced during the model training stage to quantize FP32 features to INT8 format, thus seamlessly integrating with the S205 compressed feature set. In actual deployment, TensorRT or OpenVINO can be used for inference acceleration, and the S205 feature compression method can be combined to reduce transmission bandwidth requirements, enabling real-time condition recognition inference on spaceborne computing modules, UAV-mounted terminals, or mobile edge servers (such as NVIDIA Jetson AGX).

[0066] In the S206 model output stage, it can be expanded into a multi-task learning framework, which not only identifies the operation / shutdown status of wind turbines but also simultaneously estimates blade rotational speed and identifies structural anomalies. For example, the blade ambiguity in the feature extraction and enhancement modules of S201-S203 is used to analyze rotational speed (RPM is inferred from the motion blur kernel), the shadow length is used to estimate blade attitude changes under the solar incidence angle, and time-series images are combined to detect anomalies such as missing blades, damage, or long-term shutdown. In terms of network structure, a rotational speed regression branch and an anomaly classification branch can be added in addition to the detection branch, and a multi-task weighting strategy (such as GradNorm dynamic weight allocation) can be used in the loss function to balance the training contributions of each task, thereby outputting complete equipment status information in a single inference.

[0067] The entire S201-S206 process can be embedded into a three-tiered cloud-edge-device collaborative inference system: the cloud is responsible for batch processing of massive satellite imagery, dynamic data augmentation generation, model training and updates; edge nodes are responsible for rapid inference of compressed feature sets and local model fine-tuning (S205, S206), especially for real-time monitoring of wind farms under extreme weather paths such as typhoons and blizzards; and terminal devices receive inference results through lightweight APIs for on-site operation and maintenance decisions. Furthermore, an adaptive resolution inference mechanism can be introduced, dividing the feature extraction and fusion processes in S201 and S204 into a low-resolution fast mode and a high-resolution fine mode. This allows for rapid scanning of sparse areas and automatic switching to fine detection for dense areas, reducing computational load while maintaining accuracy.

[0068] In a specific embodiment of this application, after obtaining the status identification result of the output wind turbine, the wind turbine power generation efficiency can be predicted by combining meteorological data (such as wind speed), further empowering actuarial and risk assessment scenarios. Specifically, historical meteorological data and actual power generation efficiency data of the area where the wind turbine is located can be collected to establish a mapping model between meteorological data (wind speed, etc.) and power generation efficiency. After obtaining the current status identification result of the wind turbine, real-time meteorological data is simultaneously acquired and input into the constructed mapping model to predict the power generation efficiency of the wind turbine under the current meteorological conditions such as wind speed. Based on these prediction results, the actuarial field can more accurately assess the power generation revenue and potential losses of wind turbines under different meteorological conditions, and reasonably formulate insurance rates and compensation strategies. In risk assessment scenarios, it can predict in advance the fluctuation of power generation efficiency caused by meteorological changes, identify potentially high-risk periods and areas, provide risk warnings for wind farm operators, help them adjust their operation and maintenance strategies in a timely manner, reduce equipment failure risks and economic losses, and improve the operational stability and economic benefits of the entire wind power industry.

[0069] Further, please refer to Figure 3 The steps of identifying complementary feature pairs in the initial feature set and integrating these complementary feature pairs to obtain the fused feature set specifically include:

[0070] S301, for each texture feature and each geometric feature, perform feature complementarity pairing in sequence and calculate the complementarity pairing value;

[0071] S302, if the complementarity pairing value is greater than or equal to the preset complementarity threshold, then the texture feature and the geometric feature are determined to be a complementary feature pair;

[0072] S303, perform feature fusion on the texture features and geometric features that are complementary feature pairs in the preliminary feature set to obtain a fused feature set.

[0073] In this embodiment, the feature complementarity pairing in S301 can be achieved based on the statistical correlation or information entropy difference between optical texture features and SAR geometric features at the same spatial location. Specifically, the two types of features can be normalized first, and then their mutual information, Pearson correlation coefficient, or similarity measure based on feature distribution overlap can be calculated to obtain the complementarity pairing value. In S302, feature pairs with large information differences and strong complementarity are selected by setting a complementarity threshold (e.g., 0.65-0.8) to avoid redundant feature fusion. In S303, the fusion method can use feature concatenation, weighted summation, or attention mechanism fusion (e.g., SE module or CBAM module), dynamically adjusting the weight allocation of different modal features during the fusion process. To improve the spatial consistency of the fused features, multi-scale convolution can be introduced after fusion to extract contextual information, and conditional random fields (CRF) can be used for boundary smoothing, ultimately forming a high-quality fused feature set to provide multimodal information support for subsequent state recognition.

[0074] For example, suppose we extract the texture features of the blade edges from satellite imagery of a wind farm area, and simultaneously extract the geometric features of the wind turbine towers from synthetic aperture radar (SAR) imagery. First, these two types of features are normalized at the same spatial location to eliminate the influence of scale differences. Then, their mutual information value is calculated to measure the correlation between the texture and geometric features in terms of information content. If the mutual information value exceeds a preset threshold, the pair of features is considered to have strong complementarity and can jointly describe the different physical properties of the wind turbine. Next, these complementary features are fused, forming a comprehensive feature vector containing both texture and geometric information through splicing or weighted fusion methods.

[0075] If the complementarity pairing value is less than a preset complementarity threshold, the system will determine that the complementarity between the texture feature and the geometric feature is weak, and consider that they have little information overlap or insufficient correlation in describing the state of wind power equipment. In this case, the system will temporarily exclude the feature pair from the fusion feature set to avoid introducing redundant or noisy features that could affect the overall model performance. Simultaneously, the system will further analyze feature pairs that fail to meet the threshold, potentially employing alternative feature matching strategies or adjusting the threshold parameters to explore potential nonlinearities or implicit correlations. Furthermore, the system can store these low-complementarity features separately as partial input to feature enhancement or deep fusion modules, or combine them with dynamic data augmentation strategies for reprocessing to enhance their informational value. Through this screening and dynamic adjustment mechanism, the high quality and effectiveness of the fusion feature set are ensured, improving the accuracy and stability of wind power equipment state identification.

[0076] By following the steps above, the complementary information advantages of optical and SAR images can be effectively preserved, and the discriminative power and stability of the fused features can be improved, thereby providing a more comprehensive and robust feature input for wind power equipment status identification.

[0077] Furthermore, for each texture feature and each geometric feature, the step of sequentially performing feature complementarity pairing and calculating the complementarity pairing value specifically includes:

[0078] The correlation coefficient between texture features and geometric features is obtained by using the mutual information method.

[0079] Based on the comparison results between the correlation coefficient and the preset correlation threshold, the strength of complementarity between texture features and geometric features is determined.

[0080] The strength of complementarity is quantified to obtain a complementarity pairing value. The larger the complementarity pairing value, the stronger the complementarity between texture features and geometric features.

[0081] In this embodiment, the mutual information method is used to characterize the degree of information sharing between texture features and geometric features. Its calculation process includes: first, normalizing the two types of feature vectors to eliminate dimensional differences; then, calculating the mutual information value by statistically analyzing their joint probability distribution and their respective marginal probability distributions. This value can be further normalized to the 0-1 range for direct comparison with a preset correlation threshold. When the mutual information value is below the threshold, it indicates that the two features have significant information differences and strong complementarity; conversely, the complementarity is weak. To achieve quantitative representation, linear or nonlinear mapping functions (such as Sigmoid or Min-Max mapping) can be used to convert the correlation coefficient into a complementary pairing value, which is recorded in the feature pairing matrix for use in feature pair selection in S302 and fusion in S303. This process can be performed not only on single-scale features but also in parallel on multi-scale feature maps, thus taking into account both local and global complementarity evaluation.

[0082] Through the above steps, this scheme can accurately quantify the complementary relationship between optical texture and SAR geometric features, ensuring that subsequent fusion steps prioritize the use of feature combinations with strong information differences, thereby improving the discriminability and adaptability of the fusion results.

[0083] Furthermore, the steps of determining cloud-occluded region features by fusing the feature set and using an attention mechanism to locally enhance the cloud-occluded region to obtain the enhanced feature set specifically include:

[0084] The fused feature set is traversed to identify feature identifiers of areas obscured by clouds.

[0085] Based on feature identifiers, cloud-obscured areas are located within the fused feature set;

[0086] For cloud-occupied areas, a matching attention weight matrix is ​​designed, which is used to emphasize key features within the cloud-occupied areas.

[0087] The features within the cloud-occluded area are weighted according to the attention weight matrix to achieve local enhancement of the cloud-occluded area, resulting in an enhanced feature set.

[0088] In this embodiment, the traversal of the fused feature set can be performed using a sliding window or block scanning method. Potential cloud areas are identified by combining the spectral characteristics of optical images (e.g., high reflectivity in the blue band and low reflectivity in the infrared band) and the lack of texture consistency. Simultaneously, regions with decreased signal-to-noise ratio in SAR features are referenced to form preliminary cloud occlusion feature identifiers. After locating cloud occlusion regions based on these feature identifiers, the corresponding feature maps can be extracted and input into the attention weight generation module. The weight matrix can be generated using a mechanism combining channel attention and spatial attention (e.g., the CBAM module). In the channel dimension, the contribution of SAR geometric features is increased, while in the spatial dimension, higher weights are assigned to locally significant regions. During weighted processing, the weight matrix is ​​multiplied with the target region features through element-wise multiplication, thereby highlighting key feature responses and suppressing irrelevant noise. This process can be combined with regularization constraints to prevent excessive weight concentration. Finally, the weights are concatenated with the features of the unoccluded regions to restore the complete enhanced feature set.

[0089] By following the steps above, key feature responses can be effectively amplified under cloud cover conditions, reducing the impact of missing information on recognition accuracy and thus improving the stability of state recognition under complex weather conditions.

[0090] Furthermore, the steps of extracting terrain difference data from multi-source data and adaptively adjusting the enhanced feature set using the terrain difference data to obtain the adapted feature set specifically include:

[0091] Multi-source data is preprocessed, and terrain feature points are identified in the preprocessed multi-source data;

[0092] Based on terrain feature points, a terrain difference model is constructed, and terrain difference data between different terrain feature points are calculated based on the terrain difference model;

[0093] The terrain difference data is fused with the enhanced feature set to obtain fused feature data;

[0094] Based on the fused feature data, an adaptive algorithm is used to adjust the enhanced feature set to obtain an adaptive feature set, which is a set of features that can adapt to different terrain conditions.

[0095] In this embodiment, preprocessing of multi-source data includes atmospheric correction, geometric correction, and noise suppression for optical images, and radiometric calibration, speckle noise filtering, and georegistration for SAR images to ensure the accuracy and consistency of subsequent terrain feature extraction. Terrain feature points can be extracted using digital elevation models (DEMs), interferometric SAR (InSAR) data, or stereo mapping techniques. Feature point types include elevation peaks, depressions, slope change points, and terrain boundary points. Based on these feature points, a terrain difference model can be constructed using methods such as triangular mesh (TIN) interpolation, kriging interpolation, or multi-resolution fractal analysis to calculate indicators such as elevation difference, slope difference, aspect difference, and local undulation between feature points, forming a terrain difference data matrix. During the fusion stage, the terrain difference data is modally aligned with the enhanced feature set. Feature mapping can be used to encode the terrain features into tensors of the same dimension as the enhanced feature set, and cross-modal attention is used to dynamically fuse them in both channel and spatial dimensions to obtain fused feature data. During the adaptive adjustment phase, Conditional Batch Normalization, Conditional Convolution (CondConv), or Meta-learning methods can be used to dynamically adjust the feature channel weights and convolution kernel parameters based on the terrain category labels (such as plains, hills, and mountains) of the fused feature data, so that the final generated adaptive feature set maintains high feature stability and discriminative ability in different terrain environments.

[0096] Based on fused feature data, an adaptive algorithm is employed to adjust the enhanced feature set, aiming to enable the final adaptive feature set to dynamically adapt to the wind power equipment status recognition requirements under different terrain conditions. Specifically, the adaptive algorithm enhances the responsiveness to key features of wind power equipment in specific terrain environments by adjusting the weights and selecting features that incorporate terrain difference information. For example, in mountainous and undulating areas, the algorithm may automatically increase the feature weights related to slope changes and elevation differences to more accurately reflect the geometric and textural changes in the wind turbine structure caused by terrain influences; while in relatively flat areas such as plains or the sea, the weights of terrain features are reduced, relying more on the information from the texture and geometric features themselves. This adjustment process can employ conditional convolution (CondConv), attention mechanisms, or meta-learning strategies, dynamically adjusting the parameters of the convolution kernel or feature channels to achieve sensitive adaptation to different terrain features.

[0097] For example, if a wind farm is located in a mountainous area, the model will automatically enhance the feature response related to terrain undulations to avoid feature confusion caused by complex terrain and improve detection accuracy. In offshore wind farms, the terrain feature response is weakened, and the model relies more on optical textures and SAR geometric features to achieve more efficient state recognition. This step ensures that the model has good generalization performance and stable recognition capabilities in diverse terrain environments.

[0098] Furthermore, identifying terrain differences is beneficial for actuarial calculations and risk assessments of wind power equipment insurance. In mountainous wind farms, complex terrain may expose wind turbines to more challenging weather conditions such as strong wind shear and turbulence, increasing the risk of equipment damage. Insurance companies can use models to identify the condition of wind power equipment in such terrains and assess potential risks, allowing them to reasonably increase insurance rates for wind power equipment in those areas to balance the risk of potentially high payouts.

[0099] For offshore wind farms, the effects of corrosion and wave impact from the marine environment differ from those on land. By identifying marine topographic features, models can more accurately assess the long-term wear and tear and failure probability of offshore wind turbines. Insurance companies can then develop targeted insurance plans, such as setting specific deductibles and payout caps for offshore wind power equipment. Simultaneously, wind farm operators can leverage these risk assessment results based on topographic differences to plan equipment maintenance and upgrades in advance, reducing equipment failure rates and minimizing power generation losses due to equipment damage, thereby ensuring their economic benefits and the stability of insurance payouts.

[0100] Through the above steps, features can be optimized in a targeted manner under different terrain conditions, reducing the interference of terrain undulations and perspective changes on the identification of wind power equipment status, and improving the generalization ability of the model in cross-scenario deployment.

[0101] Furthermore, the steps of constructing a terrain difference model based on terrain feature points and calculating terrain difference data between different terrain feature points based on the terrain difference model specifically include:

[0102] Geographic Information System (GIS) technology is used to convert terrain feature points into three-dimensional coordinates;

[0103] Based on the three-dimensional coordinates, the elevation difference and orientation angle between adjacent terrain feature points are calculated to obtain preliminary terrain difference data;

[0104] Based on preliminary terrain difference data, spatial interpolation methods are used to estimate terrain differences at non-terrain feature points, resulting in a complete terrain difference dataset.

[0105] The terrain difference dataset is smoothed to obtain the final terrain difference data;

[0106] Obtain terrain difference data between different terrain feature points from the final terrain difference data.

[0107] In this embodiment, after the terrain feature points are acquired, the latitude and longitude information of these feature points is first converted into three-dimensional coordinates (X, Y, Z) in a unified reference system using Geographic Information System (GIS) software or API interfaces (such as ArcGIS, QGIS, GDAL), where Z represents the elevation value. To ensure accuracy, the three-dimensional coordinate conversion process requires coordinate projection transformation (such as WGS84 to UTM) and correction of measurement errors. Subsequently, based on the three-dimensional coordinate data, the elevation difference and orientation angle between adjacent terrain feature points are calculated to form preliminary terrain difference data, which is used to characterize the local terrain undulations and directional change characteristics. To cover areas without terrain feature points, spatial interpolation methods are used for estimation based on the preliminary terrain difference data. Methods such as Kriging interpolation, inverse distance weighted (IDW), or spline interpolation can be selected to extend the discrete difference values ​​to a continuous spatial range, thereby generating a complete terrain difference dataset. To avoid noise and discontinuities introduced by interpolation, the dataset is smoothed using Gaussian filtering, low-pass filtering, or Bézier surface fitting to preserve the main terrain trends and suppress local outliers. Finally, the difference values ​​between different terrain feature points are extracted from the smoothed terrain difference data. These difference values ​​will serve as important reference inputs in the subsequent terrain adaptive adjustment stage, guiding the adaptive optimization of enhancement features under different terrain conditions.

[0108] Through the above steps, the terrain differences between different regions can be accurately and continuously described, providing high-resolution, low-noise terrain input support for feature adaptive adjustment, thereby enhancing the stability of the model under complex terrain.

[0109] Furthermore, the step of fusing terrain difference data with the enhanced feature set to obtain fused feature data specifically includes:

[0110] The terrain difference data is concatenated with each feature vector in the enhanced feature set as an additional feature vector to obtain a fused feature vector containing terrain information;

[0111] The fused feature vectors are normalized to obtain standardized fused feature data;

[0112] A feature selection algorithm is used to filter the standardized fused feature data to obtain the fused feature data.

[0113] In this embodiment, when fusing terrain difference data with the enhanced feature set, the terrain difference data is first encoded into additional feature vectors compatible with the dimensions of the enhanced feature set. These vectors may include multi-dimensional indicators such as elevation difference, slope, aspect, and terrain undulation. These feature vectors are then concatenated with each feature vector in the enhanced feature set along the channel dimension to form a fused feature vector containing terrain information. To avoid the impact of differences in the dimensions and distribution of features from different sources on the fusion effect, the fused feature vector is normalized, preferably using Z-score normalization or Min-Max normalization, to ensure that the features of each dimension are within a uniform numerical range. Subsequently, feature selection algorithms (such as Principal Component Analysis (PCA), Maximum Relevance Minimum Redundancy Ratio (mRMR), and L1 regularized feature filtering) are used to filter the normalized fused feature data, retaining feature dimensions that contribute significantly to wind power equipment status identification and removing redundant or noisy features to reduce computational load and improve the model's discrimination efficiency.

[0114] Through the above steps, terrain information can be effectively embedded in the fusion process, and irrelevant variables can be removed through feature filtering, thereby providing a more refined and discriminative feature representation for state recognition under different terrain conditions.

[0115] In the above embodiments, this application discloses a wind power equipment status recognition method, belonging to the field of artificial intelligence technology, to empower financial risk control. First, a complementary fusion feature set is constructed by combining the texture features of optical images and the geometric features of synthetic aperture radar images. For influencing factors such as cloud cover, an attention mechanism is used for local enhancement, significantly improving the recognition accuracy in cloud-covered areas. By introducing terrain difference data, the enhanced features are adaptively adjusted, enabling the model to flexibly adapt to different terrain undulations and spatial distributions. Subsequently, quantization compression reduces the feature dimension, alleviating the computational and storage burden of the model and facilitating efficient deployment on edge devices. Combined with a transfer learning framework, rapid fine-tuning and adaptation of the model in new regions are achieved, significantly shortening the model training cycle and improving recognition accuracy. This application effectively integrates multimodal data and machine learning algorithms, achieving high precision, low latency, and strong robustness in wind power equipment status recognition. It is suitable for complex and ever-changing real-world application scenarios, greatly improving the operation and maintenance efficiency and intelligent management level of wind farms.

[0116] In this embodiment, the wind power equipment status identification method operates on electronic devices (e.g., Figure 1The server shown can receive instructions or acquire data via wired or wireless connection. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra-wideband) connections, and other currently known or future wireless connection methods.

[0117] It should be emphasized that, in order to further ensure the privacy and security of the aforementioned wind power equipment status information, the aforementioned wind power equipment status information can also be stored in a blockchain node.

[0118] The blockchain referred to in this application is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0119] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0120] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0121] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware with computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When executed, the program can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0122] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures 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, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0123] Further reference Figure 4 As a response to the above Figure 2 The implementation of the method shown in this application provides an embodiment of a wind power equipment status identification device, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0124] like Figure 4 As shown, the wind power equipment status identification device 400 described in this embodiment includes:

[0125] The multi-source data module 401 is used to acquire optical images and synthetic aperture radar images of wind power equipment as multi-source data, extract texture features from the optical images, and extract geometric features from the synthetic aperture radar images to obtain a preliminary feature set.

[0126] The feature complementarity module 402 is used to determine complementary feature pairs in the preliminary feature set, integrate the complementary feature pairs to obtain a fused feature set, wherein each complementary feature pair includes a texture feature and a geometric feature.

[0127] Feature enhancement module 403 is used to determine the features of the cloud-occluded area in the fused feature set, and to perform local enhancement on the cloud-occluded area using an attention mechanism to obtain an enhanced feature set;

[0128] The adaptive adjustment module 404 is used to extract terrain difference data from multi-source data, and use the terrain difference data to adaptively adjust the enhanced feature set to obtain an adaptive feature set. The terrain difference data reflects the undulation and spatial distribution characteristics of the terrain where the wind power equipment is located.

[0129] The quantization and compression module 405 is used to quantize and compress the adaptive feature set to obtain a compressed feature set, and then send the compressed feature set to the wind power equipment status recognition model that has been pre-deployed at the edge.

[0130] The model fine-tuning module 406 is used to fine-tune the parameters of the wind power equipment status recognition model based on the compressed feature set using a transfer learning framework until the model is fitted, and output the status recognition results of the wind power equipment.

[0131] Further, please refer to Figure 5 The feature complementation module 402 specifically includes:

[0132] The complementary pairing unit 501 is used to perform feature complementary pairing sequentially for each texture feature and each geometric feature, and calculate the complementary pairing value.

[0133] The complementary feature determination unit 502 is used to determine the texture feature and the geometric feature as a complementary feature pair if the complementary pairing value is greater than or equal to a preset complementary threshold.

[0134] The complementary feature fusion unit 503 is used to fuse texture features and geometric features that are complementary feature pairs in the preliminary feature set to obtain a fused feature set.

[0135] Furthermore, the complementary pairing units specifically include:

[0136] The correlation calculation subunit is used to calculate the correlation between texture features and geometric features using the mutual information method, and obtain the correlation coefficient.

[0137] The correlation comparison subunit is used to determine the strength of complementarity between texture features and geometric features based on the comparison results of the correlation coefficient and the preset correlation threshold.

[0138] The complementarity quantization subunit is used to quantify the strength of complementarity to obtain a complementarity pairing value. The larger the complementarity pairing value, the stronger the complementarity between the texture features and the geometric features.

[0139] Furthermore, the feature enhancement module 403 specifically includes:

[0140] The feature set traversal unit is used to traverse the fused feature set and identify feature identifiers of areas with cloud cover.

[0141] The occlusion area localization unit is used to locate cloud-occluded areas in a fused feature set based on feature identifiers.

[0142] Attention weight unit, used to design matching attention weight matrix for cloud-occluded areas, where attention weight matrix is ​​used to emphasize key features within cloud-occluded areas;

[0143] The weighted processing unit is used to perform weighted processing on the features in the cloud-occluded area according to the attention weight matrix, so as to achieve local enhancement of the cloud-occluded area and obtain the enhanced feature set.

[0144] Furthermore, the adaptation module 404 specifically includes:

[0145] The preprocessing unit is used to preprocess multi-source data and identify terrain feature points in the preprocessed multi-source data;

[0146] The terrain difference calculation unit is used to construct a terrain difference model based on terrain feature points, and to calculate terrain difference data between different terrain feature points based on the terrain difference model;

[0147] The enhanced feature fusion unit is used to fuse terrain difference data with the enhanced feature set to obtain fused feature data;

[0148] The adaptive adjustment unit is used to adjust the enhanced feature set according to the fused feature data using an adaptive algorithm to obtain an adaptive feature set, wherein the adaptive feature set is a feature set that can adapt to different terrain conditions.

[0149] Furthermore, the terrain difference calculation unit specifically includes:

[0150] The coordinate construction sub-unit is used to convert terrain feature points into three-dimensional coordinates using geographic information system technology;

[0151] The elevation difference and orientation angle calculation subunit is used to calculate the elevation difference and orientation angle between adjacent terrain feature points based on three-dimensional coordinates, and obtain preliminary terrain difference data.

[0152] The spatial interpolation subunit is used to estimate the terrain differences of non-terrain feature points based on preliminary terrain difference data using spatial interpolation methods, thereby obtaining a complete terrain difference dataset.

[0153] The smoothing subunit is used to smooth the terrain difference dataset to obtain the final terrain difference data;

[0154] The terrain difference identification subunit is used to obtain terrain difference data between different terrain feature points from the final terrain difference data.

[0155] Furthermore, the enhanced feature fusion unit specifically includes:

[0156] The vector concatenation subunit is used to concatenate the terrain difference data with each feature vector in the enhanced feature set as an additional feature vector to obtain a fused feature vector containing terrain information.

[0157] The normalization processing subunit is used to normalize the fused feature vector to obtain standardized fused feature data;

[0158] The feature data filtering subunit is used to filter the standardized fused feature data using a feature selection algorithm to obtain fused feature data.

[0159] In the above embodiments, this application discloses a wind power equipment status recognition device, belonging to the field of artificial intelligence technology, to empower financial risk control. First, a complementary fusion feature set is constructed by combining the texture features of optical images and the geometric features of synthetic aperture radar images. For influencing factors such as cloud cover, an attention mechanism is used for local enhancement, significantly improving the recognition accuracy in cloud-covered areas. By introducing terrain difference data, the enhanced features are adaptively adjusted, enabling the model to flexibly adapt to different terrain undulations and spatial distributions. Subsequently, quantization compression reduces the feature dimension, alleviating the computational and storage burden of the model and facilitating efficient deployment on edge devices. Combined with a transfer learning framework, rapid fine-tuning and adaptation of the model in new regions are achieved, significantly shortening the model training cycle and improving recognition accuracy. This application effectively integrates multimodal data and machine learning algorithms, achieving high precision, low latency, and strong robustness in wind power equipment status recognition. It is suitable for complex and ever-changing practical application scenarios, greatly improving the operation and maintenance efficiency and intelligent management level of wind farms.

[0160] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 6 , Figure 6 This is a basic structural block diagram of the computer device in this embodiment.

[0161] The computer device 6 includes a memory 61, a processor 62, and a network interface 63 that are interconnected via a system bus. It should be noted that only the computer device 6 with memory 61, processor 62, and network interface 63 is shown in the figure; however, it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0162] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0163] The memory 61 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 61 may be an internal storage unit of the computer device 6, such as the hard disk or memory of the computer device 6. In other embodiments, the memory 61 may also be an external storage device of the computer device 6, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 6. Of course, the memory 61 may include both the internal storage unit and its external storage device of the computer device 6. In this embodiment, the memory 61 is typically used to store the operating system and various application software installed on the computer device 6, such as computer-readable instructions for wind power equipment status identification methods. In addition, the memory 61 can also be used to temporarily store various types of data that have been output or will be output.

[0164] In some embodiments, the processor 62 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 62 is typically used to control the overall operation of the computer device 6. In this embodiment, the processor 62 is used to execute computer-readable instructions stored in the memory 61 or to process data, for example, to execute computer-readable instructions for the wind power equipment status identification method.

[0165] The network interface 63 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 6 and other electronic devices.

[0166] This application also provides an embodiment, namely, a computer device including a memory and a processor. The memory stores computer-readable instructions, and the processor, when executing the computer-readable instructions, implements the steps of the wind power equipment status identification method described above, that is, it implements:

[0167] A method for identifying the status of wind power equipment, comprising:

[0168] Optical images and synthetic aperture radar images of wind power equipment are acquired as multi-source data. Texture features are extracted from the optical images and geometric features are extracted from the synthetic aperture radar images to obtain a preliminary feature set.

[0169] The complementary feature pairs in the initial feature set are identified, and the complementary feature pairs are integrated to obtain the fused feature set, wherein each complementary feature pair includes a texture feature and a geometric feature.

[0170] The features of the cloud-occluded region are determined by fusing the feature set, and an attention mechanism is used to locally enhance the cloud-occluded region to obtain the enhanced feature set.

[0171] Topographic difference data is extracted from multi-source data, and the enhanced feature set is adaptively adjusted using the topographic difference data to obtain an adaptive feature set. The topographic difference data reflects the undulation and spatial distribution characteristics of the terrain where the wind power equipment is located.

[0172] The adaptive feature set is quantized and compressed to obtain a compressed feature set, and the compressed feature set is distributed to the wind power equipment status recognition model that has been pre-deployed at the edge.

[0173] A transfer learning framework is used to fine-tune the parameters of the wind power equipment status recognition model based on a compressed feature set until the model fits the model, and then output the status recognition results of the wind power equipment.

[0174] This application also provides another embodiment, namely, a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the wind power equipment status identification method described above, i.e., to implement:

[0175] A method for identifying the status of wind power equipment, comprising:

[0176] Optical images and synthetic aperture radar images of wind power equipment are acquired as multi-source data. Texture features are extracted from the optical images and geometric features are extracted from the synthetic aperture radar images to obtain a preliminary feature set.

[0177] The complementary feature pairs in the initial feature set are identified, and the complementary feature pairs are integrated to obtain the fused feature set, wherein each complementary feature pair includes a texture feature and a geometric feature.

[0178] The features of the cloud-occluded region are determined by fusing the feature set, and an attention mechanism is used to locally enhance the cloud-occluded region to obtain the enhanced feature set.

[0179] Topographic difference data is extracted from multi-source data, and the enhanced feature set is adaptively adjusted using the topographic difference data to obtain an adaptive feature set. The topographic difference data reflects the undulation and spatial distribution characteristics of the terrain where the wind power equipment is located.

[0180] The adaptive feature set is quantized and compressed to obtain a compressed feature set, and the compressed feature set is distributed to the wind power equipment status recognition model that has been pre-deployed at the edge.

[0181] A transfer learning framework is used to fine-tune the parameters of the wind power equipment status recognition model based on a compressed feature set until the model fits the model, and then output the status recognition results of the wind power equipment.

[0182] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0183] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0184] It should be noted that the software tools or components not belonging to this company that appear in the various embodiments of this application are merely illustrative examples and do not represent actual use.

[0185] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A method for identifying the status of wind power equipment, characterized in that, include: Optical images and synthetic aperture radar images of wind power equipment are acquired as multi-source data. Texture features are extracted from the optical images and geometric features are extracted from the synthetic aperture radar images to obtain a preliminary feature set. The complementary feature pairs in the preliminary feature set are determined, and the complementary feature pairs are integrated to obtain a fused feature set, wherein each complementary feature pair includes a texture feature and a geometric feature; The cloud-occluded region features are determined from the fused feature set, and an attention mechanism is used to locally enhance the cloud-occluded region to obtain an enhanced feature set. Topographic difference data is extracted from the multi-source data, and the enhanced feature set is adaptively adjusted using the topographic difference data to obtain an adaptive feature set, wherein the topographic difference data reflects the undulation and spatial distribution characteristics of the terrain where the wind power equipment is located. The adaptive feature set is quantized and compressed to obtain a compressed feature set, and the compressed feature set is sent to the wind power equipment status recognition model that has been pre-deployed at the edge. Using a transfer learning framework, the parameters of the wind power equipment status recognition model are fine-tuned based on the compressed feature set until the model is fitted, and the status recognition result of the wind power equipment is output.

2. The wind power equipment status identification method as described in claim 1, characterized in that, The step of determining complementary feature pairs in the preliminary feature set and integrating the complementary feature pairs to obtain a fused feature set specifically includes: For each texture feature and each geometric feature, feature complementarity pairing is performed sequentially, and a complementarity pairing value is calculated. If the complementarity pairing value is greater than or equal to a preset complementarity threshold, then the texture feature and the geometric feature are determined to be a complementary feature pair; The texture features and geometric features that are complementary feature pairs in the preliminary feature set are fused to obtain the fused feature set.

3. The wind power equipment status identification method as described in claim 2, characterized in that, The step of sequentially performing feature complementarity pairing and calculating the complementarity pairing value for each texture feature and each geometric feature specifically includes: The correlation coefficient between the texture features and the geometric features is obtained by calculating the correlation coefficient using the mutual information method. Based on the comparison result between the correlation coefficient and the preset correlation threshold, the complementarity between the texture feature and the geometric feature is determined. The complementarity strength is quantified to obtain the complementarity pairing value, wherein the larger the complementarity pairing value, the stronger the complementarity between the texture feature and the geometric feature.

4. The wind power equipment status identification method as described in claim 1, characterized in that, The step of determining cloud-occluded region features from the fused feature set and using an attention mechanism to locally enhance the cloud-occluded region to obtain an enhanced feature set specifically includes: The fused feature set is traversed to identify feature identifiers of areas obscured by clouds. Based on the feature identifiers, the cloud-obscured area is located in the fused feature set; For the cloud-occupied area, a matching attention weight matrix is ​​designed, wherein the attention weight matrix is ​​used to emphasize key features within the cloud-occupied area; The features within the cloud-occluded region are weighted according to the attention weight matrix to achieve local enhancement of the cloud-occluded region, resulting in the enhanced feature set.

5. The wind power equipment status identification method as described in claim 1, characterized in that, The steps of extracting terrain difference data from the multi-source data and adaptively adjusting the enhanced feature set using the terrain difference data to obtain an adapted feature set specifically include: The multi-source data is preprocessed, and terrain feature points are identified in the preprocessed multi-source data. Based on the terrain feature points, a terrain difference model is constructed, and terrain difference data between different terrain feature points are calculated based on the terrain difference model. The terrain difference data is fused with the enhanced feature set to obtain fused feature data; Based on the fused feature data, an adaptive algorithm is used to adjust the enhanced feature set to obtain the adaptive feature set, wherein the adaptive feature set is a feature set that can adapt to different terrain conditions.

6. The wind power equipment status identification method as described in claim 5, characterized in that, The steps of constructing a terrain difference model based on the terrain feature points and calculating terrain difference data between different terrain feature points based on the terrain difference model specifically include: Geographic Information System (GIS) technology is used to convert the terrain feature points into three-dimensional coordinates; Based on the three-dimensional coordinates, the elevation difference and orientation angle between adjacent terrain feature points are calculated to obtain preliminary terrain difference data; Based on the preliminary terrain difference data, the terrain differences of non-terrain feature points are estimated using spatial interpolation methods to obtain a complete terrain difference dataset. The terrain difference dataset is smoothed to obtain the final terrain difference data; The terrain difference data between different terrain feature points is obtained from the final terrain difference data.

7. The wind power equipment status identification method as described in claim 5, characterized in that, The step of fusing the terrain difference data with the enhanced feature set to obtain fused feature data specifically includes: The terrain difference data is concatenated with each feature vector in the enhanced feature set as an additional feature vector to obtain a fused feature vector containing terrain information; The fused feature vector is normalized to obtain standardized fused feature data; The standardized fused feature data is filtered using a feature selection algorithm to obtain the fused feature data.

8. A wind power equipment status identification device, characterized in that, include: The multi-source data module is used to acquire optical images and synthetic aperture radar images of wind power equipment as multi-source data, extract texture features from the optical images, and extract geometric features from the synthetic aperture radar images to obtain a preliminary feature set. A feature complementarity module is used to determine complementary feature pairs in the preliminary feature set, integrate the complementary feature pairs to obtain a fused feature set, wherein each complementary feature pair includes a texture feature and a geometric feature; The feature enhancement module is used to determine the features of the cloud-occluded area in the fused feature set, and to perform local enhancement on the cloud-occluded area using an attention mechanism to obtain an enhanced feature set. An adaptive adjustment module is used to extract terrain difference data from the multi-source data, and use the terrain difference data to adaptively adjust the enhanced feature set to obtain an adaptive feature set, wherein the terrain difference data reflects the undulation and spatial distribution characteristics of the terrain where the wind power equipment is located. The quantization and compression module is used to quantize and compress the adaptive feature set to obtain a compressed feature set, and then send the compressed feature set to the wind power equipment status recognition model that has been pre-deployed at the edge. The model fine-tuning module is used to fine-tune the parameters of the wind power equipment status recognition model based on the compressed feature set using a transfer learning framework until the model is fitted, and then outputs the status recognition result of the wind power equipment.

9. A computer device, characterized in that, The device includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the wind power equipment status identification method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the wind power equipment status identification method as described in any one of claims 1 to 7.

Images