Energy storage power station overheat point dynamic monitoring method and device based on high-precision machine vision

Abstract

The invention discloses an energy storage power station overheat point dynamic monitoring method and device based on high-precision machine vision, and the method comprises the steps: arranging an image collection unit based on the layout of energy storage power station equipment, and synchronously collecting the visible light and multispectral images of the equipment in an energy storage power station; carrying out median filtering denoising, graying and histogram equalization preprocessing on the original image; accurate identification and positioning of the equipment are realized by using a YOLO series algorithm; segmenting an equipment image by adopting a U-Net algorithm, screening out an overheat point candidate region through a gray average value and a variance threshold value, and performing accurate judgment by combining a multispectral image fusion technology and an SVM classification algorithm; and for the confirmed overheat point, the system uses a Kalman filtering algorithm to realize dynamic tracking, and predicts the future temperature change trend based on an ARIM model. A closed-loop feedback mechanism is formed in the whole process, the system performance is continuously improved through parameter tuning and feedback optimization, and finally high-precision, real-time and intelligent monitoring of the overheat points of the energy storage power station is achieved.

Description

Technical Field

[0001] This invention belongs to the field of energy storage power station safety monitoring technology, and in particular relates to a method and device for dynamic monitoring of overheated spots in energy storage power stations based on high-precision machine vision. Background Technology

[0002] With the increasing global demand for clean energy, energy storage power stations, as key facilities for energy storage and flexible allocation, have been widely adopted. However, during operation, battery packs, electrical connection components, and other equipment in energy storage power stations may experience overheating due to aging, poor contact, or abnormal heat dissipation. If these overheating points are not detected and addressed in a timely manner, they may cause serious accidents such as fires and explosions, resulting in huge economic losses and casualties.

[0003] Currently, overheat monitoring methods in energy storage power stations mainly include contact and non-contact temperature measurement methods. Contact methods, such as thermocouple and resistance temperature detector (RTD) measurements, offer high accuracy but require direct contact with the object being measured, making installation and maintenance complex and limiting their ability to achieve large-area, real-time monitoring. Non-contact methods, such as infrared thermal imaging, enable non-contact measurement and quickly acquire large-area temperature distribution images, but suffer from limitations such as measurement accuracy being significantly affected by environmental factors, inability to accurately identify minute overheats, and difficulty in dynamic tracking. Furthermore, most existing monitoring methods can only perform static temperature detection and cannot effectively analyze and predict the development trend of overheats.

[0004] High-precision machine vision technology has demonstrated powerful capabilities in target detection and image analysis. By extracting and analyzing image features, it can achieve accurate identification and localization of target objects. Applying high-precision machine vision technology to the monitoring of overheated spots in energy storage power stations is expected to overcome the shortcomings of existing technologies and achieve high-precision, dynamic monitoring of overheated spots. Summary of the Invention

[0005] The purpose of this invention is to provide a method and device for dynamic monitoring of overheated hot spots in energy storage power stations based on high-precision machine vision. This invention solves the problems of low measurement accuracy and inability to achieve dynamic tracking and trend prediction in existing methods for monitoring overheated hot spots in energy storage power stations. It enables high-precision, real-time, and dynamic monitoring of overheated hot spots in energy storage power stations, thereby improving the safety and reliability of energy storage power station operation.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0007] This invention provides a method for dynamic monitoring of overheated spots in energy storage power stations based on high-precision machine vision.

[0008] The method includes:

[0009] Based on the layout of the energy storage power station equipment, an image acquisition unit is arranged, including a high-precision imaging device, an auxiliary lighting device, and a multispectral imaging device. The image acquisition unit synchronously acquires the original visual images and multispectral images of the target equipment in the energy storage power station through the high-precision imaging device and the multispectral imaging device at set time intervals or trigger conditions. The image acquisition unit is connected to the data processing server via gigabit Ethernet.

[0010] The original visual images are preprocessed to improve the clarity of device details in the images; a deep learning target detection algorithm is used to extract and locate device features in the preprocessed images, and output the device category information and position coordinates in the image; a deep learning image segmentation algorithm is used to divide the target device region in the preprocessed image into sub-regions based on the position coordinates, calculate the feature parameters of each sub-region, and mark over-hotspot candidate regions by threshold determination based on the threshold range of feature parameters of normal device regions.

[0011] The original visual image and the multispectral image are fused together. The multidimensional feature vectors of the candidate regions of the hotspots in the fused image are extracted. Based on the classification model, the result of whether the candidate regions are hotspots is output.

[0012] The Kalman filter target tracking algorithm is used to dynamically track the confirmed hot spots, and the time series analysis method is used to predict the trend of temperature change data of the hot spots.

[0013] Set different levels of overheating temperature warning thresholds, and issue a warning when the detected overheating temperature reaches or exceeds the warning threshold.

[0014] In one embodiment, the high-precision imaging device is a high-precision industrial camera with a resolution of not less than 4K and a frame rate of not less than 30fps, and is equipped with an optical lens with adjustable focal length; the auxiliary lighting device is an LED lighting device with adjustable light intensity and angle, and has adaptive adjustment function for light intensity and angle; the spectral imaging device acquires data in the visible light and near-infrared bands.

[0015] In one embodiment, the preprocessing operation on the acquired raw visual image to improve the clarity of device details in the image includes:

[0016] The median filtering algorithm is used to denoise the original visual image, converting the color image into a grayscale image, and histogram equalization is used to enhance the image contrast, thereby improving the clarity of device details in the image.

[0017] In one implementation, the deep learning object detection algorithm is a YOLO series algorithm. The model is trained using a training set labeled with device category and location information, enabling the model to identify the device used for image acquisition.

[0018] In one implementation, the deep learning image segmentation algorithm is the U-Net algorithm, and the threshold is determined by statistical analysis of the gray-scale mean and variance of a large number of normal regions and overheated regions.

[0019] In one implementation, a weighted average-based image fusion algorithm is used to fuse the original visual image with a multispectral image. The multidimensional feature vector includes grayscale features, texture features, and spectral features, and the classification model is a support vector machine model.

[0020] In one implementation, the Kalman filter target tracking algorithm can combine the time interval between image frames and predict the position of the hot spot in the next frame image based on the position and motion state of the hot spot in the current frame image. By continuously updating and correcting the predicted value, accurate tracking of the hot spot can be achieved. The time series analysis method is an autoregressive integral moving average model, which can establish an autoregressive integral moving average model based on historical temperature data to predict the future temperature change trend of the hot spot.

[0021] In one embodiment, the step of setting different levels of overheating temperature warning thresholds, and performing warning processing when the detected overheating temperature reaches or exceeds the warning threshold, includes:

[0022] Two levels of warning thresholds are set: when the temperature exceeds 80℃, the first level of warning is activated and an alarm is issued through an audible and visual alarm device; when the temperature exceeds 90℃, the second level of warning is activated and staff are notified through audible and visual alarms, SMS, and email.

[0023] In one embodiment, the method further includes: generating corresponding treatment measures based on the location, type, and temperature change trend information of the overheated spot, combined with the operating status and emergency plan of the energy storage power station.

[0024] The present invention also provides a dynamic monitoring device for overheating in energy storage power stations based on high-precision machine vision, the device comprising:

[0025] The image acquisition unit is based on the distributed layout of energy storage power station equipment and includes a high-precision industrial camera, adaptive auxiliary lighting equipment and multispectral imaging equipment.

[0026] The data communication module uses the gigabit Ethernet protocol to achieve high-speed data transmission between the image acquisition module and the data processing server, and supports real-time image streaming.

[0027] The image preprocessing module is used to preprocess the acquired raw visual images to improve the clarity of device details in the images;

[0028] The device identification and positioning module integrates a deep learning target detection algorithm to extract and locate device features from preprocessed images, and outputs the device category information and its position coordinates in the image.

[0029] The hotspot candidate identification module is used to use a deep learning image segmentation algorithm to divide the target device region in the preprocessed image into sub-regions based on the location coordinates, calculate the feature parameters of each sub-region, and mark the hotspot candidate regions by threshold determination.

[0030] The multispectral fusion judgment module is used to fuse the original visual image and the multispectral image, extract the grayscale, texture and spectral feature vectors of the candidate region, and output the hot spot judgment result based on the support vector machine classification model.

[0031] The dynamic tracking and prediction module integrates a Kalman filter tracking algorithm and a time series analysis model to track the movement trajectory of confirmed hotspots and predict temperature change trends.

[0032] The intelligent early warning decision module is used to set different levels of overheating temperature warning thresholds. When the detected overheating temperature reaches or exceeds the warning threshold, an early warning is issued.

[0033] Beneficial effects of this invention:

[0034] 1. Employing high-precision industrial cameras and advanced image processing and analysis algorithms, it can achieve high-precision detection of overheated hot spots in energy storage power stations, identify tiny overheated hot spot areas, and improve monitoring accuracy.

[0035] 2. By using the Kalman filter algorithm and time series analysis, we can dynamically track hot spots and predict future temperature change trends, identify potential safety hazards in advance, and provide staff with sufficient time to handle them.

[0036] 3. By utilizing multispectral image fusion technology and combining information from visible light images and multispectral images, the characteristics of hot spots can be reflected more comprehensively, thereby improving the accuracy and reliability of hot spot identification.

[0037] 4. Based on the detection and analysis results of overheating points, provide intelligent decision support for staff, generate corresponding handling suggestions, improve the efficiency and accuracy of problem handling, and ensure the safe and stable operation of energy storage power stations.

[0038] 5. The system can adapt to different lighting conditions and complex environments. Through auxiliary lighting equipment and image preprocessing technology, it ensures that clear images can be obtained under various conditions, thus achieving reliable hot spot monitoring. Attached Figure Description

[0039] The accompanying drawings, as part of this invention, are provided to further illustrate the invention. The illustrative embodiments and descriptions of the invention are used to explain the invention, but do not constitute an undue limitation thereof. Clearly, the drawings described below are merely some embodiments, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0040] Figure 1 This invention provides a method flow for dynamic monitoring of overheated hot spots in an energy storage power station based on high-precision machine vision, as an embodiment of the present invention.

[0041] Figure 2 This is a schematic diagram of the structure of a dynamic monitoring device for overheated hot spots in an energy storage power station based on high-precision machine vision, provided in an embodiment of the present invention.

[0042] Figure 3 This is an architecture diagram of a dynamic monitoring system for overheated energy storage power stations based on high-precision machine vision, provided in one embodiment of the present invention.

[0043] Figure 4 This is a precision-confidence curve of the YOLOv5 model in one embodiment of the present invention;

[0044] Figure 5 This is a recall-confidence curve of the YOLOv5 model in one embodiment of the present invention;

[0045] Figure 6 This is a graph showing the loss and accuracy changes during the model training process in one embodiment of the present invention.

[0046] It should be noted that these accompanying drawings and textual descriptions are not intended to limit the scope of the invention in any way, but rather to illustrate the concept of the invention to those skilled in the art by referring to specific embodiments. Detailed Implementation

[0047] The embodiments of the present invention will now be described in further detail with reference to the accompanying drawings. Through this description, the features and advantages of the present invention will become clearer and more apparent.

[0048] like Figure 1 As shown in the figure, an embodiment of the present invention illustrates a method for dynamic monitoring of overheated spots in an energy storage power station based on high-precision machine vision. The method includes the following steps:

[0049] Step S100: Based on the layout of the energy storage power station equipment, arrange an image acquisition unit including high-precision imaging equipment, auxiliary lighting equipment and multispectral imaging equipment.

[0050] The image acquisition unit synchronously acquires raw visual images and multispectral images of target equipment within the energy storage power station using high-precision imaging equipment and spectral imaging equipment at set time intervals or trigger conditions; the image acquisition unit is connected to the data processing server via gigabit Ethernet.

[0051] Furthermore, the high-precision imaging device is a high-precision industrial camera with a resolution of no less than 4K and a frame rate of no less than 30fps, and is equipped with an optical lens with adjustable focal length; the auxiliary lighting device is an LED lighting device with adjustable light intensity and angle, and has adaptive adjustment function for light intensity and angle; the spectral imaging device has a collection range including visible light and near-infrared bands.

[0052] Specifically, multiple high-precision industrial cameras are strategically deployed within the energy storage power station. These cameras should have a resolution of at least 4K and a frame rate of at least 30fps to ensure clear and rapid acquisition of images of the equipment within the power station. Simultaneously, suitable optical lenses are equipped for the industrial cameras, with different focal lengths selected based on the size and distance of the monitoring area to ensure coverage of the entire monitoring region.

[0053] Install auxiliary lighting equipment, such as high-brightness LED lights, next to the industrial camera to ensure clear images are obtained under various lighting conditions. The intensity and angle of the lighting equipment can be adjusted according to the actual situation to avoid reflections and shadows that affect image quality.

[0054] Industrial cameras are connected to a data processing server via gigabit Ethernet to enable rapid transmission of image data. The data processing server is equipped with a high-performance CPU, GPU, and large-capacity memory, providing powerful image processing and analysis capabilities.

[0055] The system controls industrial cameras to capture images of equipment within the energy storage power station according to set time intervals or trigger conditions. The captured images include key equipment such as battery packs, electrical connection components, and cables.

[0056] Step S200: Perform preprocessing operations on the acquired raw visual images to improve the clarity of device details in the images.

[0057] Furthermore, a median filtering algorithm is used to denoise the original visual image, converting the color image into a grayscale image, and histogram equalization is used to enhance the image contrast, thereby improving the clarity of device details in the image.

[0058] Specifically, the acquired raw images undergo preprocessing, including image denoising, grayscale conversion, and contrast enhancement. Median filtering is used to remove noise and improve image clarity; color images are converted to grayscale to reduce data processing; and histogram equalization is used to enhance image contrast and highlight details.

[0059] Step S300: Use a deep learning object detection algorithm to extract and locate device features in the preprocessed image, and output the device category information and its position coordinates in the image.

[0060] Furthermore, the deep learning object detection algorithm is the YOLO series algorithm. The model is trained using a training set labeled with device category and location information, enabling the model to identify the device used for image acquisition.

[0061] Specifically, deep learning-based object detection algorithms, such as the YOLO (You Only Look Once) series, are used to extract and locate device features in the preprocessed images. During the training of the object detection model, a large amount of image data of energy storage power station equipment is collected, and the location and category information of the equipment are labeled. The model is then trained and optimized to accurately identify equipment such as battery packs, electrical connection components, and cables. When the model detects equipment in the input image, it outputs the equipment's category and its location coordinates in the image, providing a foundation for subsequent hotspot detection.

[0062] Step S400: Using a deep learning image segmentation algorithm, the target device region in the preprocessed image is divided into sub-regions based on the location coordinates. The feature parameters of each sub-region are calculated. The hotspot candidate region is marked by threshold determination based on the threshold range of the feature parameters of the normal device region.

[0063] Furthermore, the deep learning image segmentation algorithm is the U-Net algorithm, and the threshold is determined by statistical analysis of the gray-level mean and variance of a large number of normal regions and overheated regions.

[0064] Specifically, for the detected device image regions, a deep learning-based image segmentation algorithm, such as the U-Net algorithm, is used to segment the device image into different sub-regions. By annotating and training a large number of device images containing both hot and normal areas, the segmentation model can accurately distinguish different parts of the device, such as battery cells and electrical connectors.

[0065] Calculate the grayscale mean and variance, among other characteristic parameters, for each sub-region. Set an appropriate threshold, and mark sub-regions whose grayscale mean and variance exceed the threshold range as candidate hotspot regions. These candidate regions may be areas of image grayscale change due to temperature increases, requiring further analysis and confirmation.

[0066] Step S500: The original visual image and the multispectral image are fused together. The multidimensional feature vectors of the candidate regions for over-hot spots in the fused image are extracted. Based on the classification model, the determination result of whether the candidate regions are over-hot spots is output.

[0067] Furthermore, a weighted average-based image fusion algorithm is used to fuse the original visual image with the multispectral image. The multidimensional feature vector includes grayscale features, texture features, and spectral features, and the classification model is a support vector machine model.

[0068] Specifically, for the marked hotspot candidate regions, a method based on multispectral image fusion is used for accurate identification. A multispectral imaging device is added to a high-precision industrial camera to acquire multispectral images of the same area. Multispectral images contain spectral information at different wavelengths, enabling a more comprehensive reflection of the object's characteristics.

[0069] Visible light images and multispectral images are fused using a weighted average-based image fusion algorithm. Different weights are assigned based on the characteristics and importance of the different spectral images to obtain the fused image. In the fused image, the features of overheated areas are more pronounced, making them easier to distinguish from normal areas.

[0070] Classification algorithms from machine learning, such as Support Vector Machines (SVM), are used to classify hotspot candidate regions in a fused image. During model training, various features of the hotspot candidate regions are extracted, including grayscale features, texture features, and spectral features. These features are used as training data to train the model. The trained model can accurately determine whether a region is a true hotspot based on the input candidate region features.

[0071] Step S600: The Kalman filter target tracking algorithm is used to dynamically track the confirmed hot spots, and the time series analysis method is used to predict the trend of temperature change data of the hot spots.

[0072] Furthermore, the Kalman filter target tracking algorithm can combine the time interval between image frames and predict the position of the hot spot in the next frame image based on the position and motion state of the hot spot in the current frame image. By continuously updating and correcting the predicted value, it can achieve accurate tracking of the hot spot. The time series analysis method is an autoregressive integral moving average model, which can establish an autoregressive integral moving average model based on historical temperature data to predict the future temperature change trend of the hot spot.

[0073] Step S700: Set different levels of overheating temperature warning thresholds, and issue a warning when the detected overheating temperature reaches or exceeds the warning threshold.

[0074] Furthermore, two levels of warning thresholds are set: when the temperature exceeds 80℃, a level one warning is activated, and an alarm is issued through an audible and visual alarm device; when the temperature exceeds 90℃, a level two warning is activated, and staff are notified through audible and visual alarms, SMS, and email.

[0075] In an optional embodiment, the method further includes: generating corresponding treatment measures based on the location, type, and temperature change trend information of the overheated spot, combined with the operating status and emergency plan of the energy storage power station.

[0076] Specifically, different warning thresholds are set according to the severity of the overheating. When the detected overheating temperature reaches or exceeds the warning threshold, warning information is sent to staff through audible and visual alarms, SMS notifications, email alerts, etc., while the location and relevant information of the overheating point are highlighted on the monitoring interface.

[0077] Provide decision support for staff by generating corresponding handling suggestions based on information such as the location, type, and temperature change trend of overheating points, combined with the operating status and emergency plan of the energy storage power station. These suggestions include adjusting equipment operating parameters and carrying out equipment maintenance, helping staff to make quick and accurate decisions and address overheating issues in a timely manner.

[0078] This invention uses a high-precision industrial camera to periodically acquire images of energy storage devices. After preprocessing operations such as median filtering for noise reduction, grayscale conversion, and histogram equalization, the YOLOv5 algorithm is used to achieve accurate identification and positioning of the devices (average accuracy of 95%). Subsequently, the U-Net algorithm is used to segment the device images, and candidate hotspot regions are screened by grayscale mean and variance thresholds. Then, multispectral image fusion technology and SVM classification algorithm are combined for accurate judgment (accuracy of 98%). For confirmed hotspots, the system uses the Kalman filter algorithm to achieve dynamic tracking (error controlled within 2 pixels) and predicts the temperature change trend for the next 30 minutes based on the ARIMA(1,1,1) model. The entire process forms a closed-loop feedback mechanism, continuously improving system performance through parameter tuning and feedback optimization, ultimately achieving high-precision, real-time, and intelligent monitoring of hotspots in energy storage power stations.

[0079] The following is an embodiment of a dynamic monitoring device for overheated hot spots in an energy storage power station based on high-precision machine vision, which can be used to execute an embodiment of a dynamic monitoring method for overheated hot spots in an energy storage power station based on high-precision machine vision, according to the present invention. For details not disclosed in the embodiment of the dynamic monitoring device for overheated hot spots in an energy storage power station based on high-precision machine vision, please refer to the embodiment of the dynamic monitoring method for overheated hot spots in an energy storage power station based on high-precision machine vision, according to the present invention.

[0080] Reference Figure 2 As shown, in one embodiment, a dynamic monitoring device for overheating in an energy storage power station based on high-precision machine vision is provided. The device includes:

[0081] The image acquisition unit is based on the distributed layout of energy storage power station equipment and includes a high-precision industrial camera, adaptive auxiliary lighting equipment and multispectral imaging equipment.

[0082] The data communication module uses the gigabit Ethernet protocol to achieve high-speed data transmission between the image acquisition module and the data processing server, and supports real-time image streaming.

[0083] The image preprocessing module is used to preprocess the acquired raw visual images to improve the clarity of device details in the images;

[0084] The device identification and positioning module integrates a deep learning target detection algorithm to extract and locate device features from preprocessed images, and outputs the device category information and its position coordinates in the image.

[0085] The hotspot candidate identification module is used to use a deep learning image segmentation algorithm to divide the target device region in the preprocessed image into sub-regions based on the location coordinates, calculate the feature parameters of each sub-region, and mark the hotspot candidate regions by threshold determination.

[0086] The multispectral fusion judgment module is used to fuse the original visual image and the multispectral image, extract the grayscale, texture and spectral feature vectors of the candidate region, and output the hot spot judgment result based on the support vector machine classification model.

[0087] The dynamic tracking and prediction module integrates a Kalman filter tracking algorithm and a time series analysis model to track the movement trajectory of confirmed hotspots and predict temperature change trends.

[0088] The intelligent early warning decision module is used to set different levels of overheating temperature warning thresholds. When the detected overheating temperature reaches or exceeds the warning threshold, an early warning is issued.

[0089] It should be noted that the various functional modules in the embodiments of the present invention can be integrated into one processing module, or each unit can exist as a separate physical entity, or two or more units can be integrated into one module. The integrated module can be implemented in hardware or as a software functional module. Specific Implementation

[0090] 1. Monitoring System Setup

[0091] Within a large-scale energy storage power station, ten high-precision industrial cameras with a resolution of 5K and a frame rate of 60fps were installed at suitable locations such as the roof and walls, based on the layout of the battery packs and electrical equipment. Each industrial camera was equipped with an adjustable focal length optical lens to ensure coverage of the critical equipment area throughout the entire energy storage power station.

[0092] High-brightness LED auxiliary lighting equipment was installed next to the industrial camera. Through on-site debugging, the light intensity and angle of the lighting equipment were adjusted to avoid reflections and shadows on the equipment surface, ensuring clear images can be obtained both day and night.

[0093] The industrial camera is connected to the data processing server via gigabit Ethernet. The data processing server uses a high-performance CPU (Intel Xeon Platinum 8380), GPU (NVIDIA A100) and 128GB of memory to meet the needs of rapid processing of large amounts of image data.

[0094] An industrial camera is set to capture an image of the equipment in the energy storage power station every 10 seconds, with the captured image resolution being 5120×2880 pixels.

[0095] 2. Image Acquisition and Preprocessing Implementation

[0096] An industrial camera is set to capture an image of the equipment in the energy storage power station every 10 seconds, with the captured image resolution being 5120×2880 pixels.

[0097] For the acquired raw images, the median filtering algorithm is first used to remove noise from the images, with the filter window size set to 3×3; then the color images are converted to grayscale images; finally, histogram equalization is used to enhance the contrast of the images and improve the clarity of device details in the images.

[0098] 3. Equipment Feature Extraction and Localization Implementation

[0099] A large amount of image data of equipment in the energy storage power station was collected, including battery packs, electrical connection components, cables, etc., totaling 5,000 images. The LabelImg tool was used to label the equipment in the images, indicating the category and location information of the equipment.

[0100] Figure 4 This is the precision-confidence curve of the YOLOv5 model on the validation set. The average precision across all device categories reaches 0.967, indicating that the model maintains good detection accuracy even at high confidence thresholds. Specifically, the precision of key device categories such as battery modules and PCS cabinets all exceed 0.95.

[0101] Figure 5 This is the recall-confidence curve for the model. The average recall for all categories is 0.63. Considering the high cost of missed detections in energy storage power station equipment monitoring scenarios, we balanced precision and recall by adjusting the model thresholds, ultimately achieving 96.7% mAP and 95.2% recall on the validation set.

[0102] The YOLOv5 algorithm was used to train the model on labeled image data. After 100 training epochs, the model achieved a mean average precision (mAP) of 95% on the validation set. Applying the trained model to the acquired images enabled accurate identification and location of devices within the images.

[0103] Figure 6 The graph illustrates the changes in the loss function and accuracy metrics during the training process of the YOLOv5 model. As can be seen from the graph: the bounding box loss (box_loss) of both the training and validation sets decreases significantly with increasing training epochs, stabilizing after 200 epochs; the classification loss (cls_loss) and distribution focus loss (dfl_loss) also show a stable decreasing trend; the precision and mean average precision (mAP50) increase with increasing training epochs, eventually reaching over 0.967 and 0.40 respectively; the training curves indicate that the model converges well and there is no overfitting.

[0104] 4. Implementation of hotspot candidate region identification

[0105] Two thousand device images, including both hot and normal regions, were selected. The LabelMe tool was used to label the device sub-regions in the images, and the U-Net image segmentation model was trained. After training, the model was able to accurately segment device images into different sub-regions, such as battery cells and electrical connectors.

[0106] The mean and variance of grayscale values ​​for each sub-region are calculated. Through statistical analysis of a large number of normal and abnormal regions, a grayscale mean threshold of 150 and a variance threshold of 20 are set. Sub-regions with a grayscale mean greater than 150 and a variance greater than 20 are marked as candidate overheated regions.

[0107] 5. Accurate hotspot assessment and implementation

[0108] Install multispectral imaging equipment next to an industrial camera to acquire multispectral images of the same area, with the spectral range including visible light (400 - 760 nm) and near-infrared (760 - 1100 nm) bands.

[0109] Visible light images and multispectral images were fused using a weighted average fusion algorithm, with the weight of the visible light image set to 0.6 and the weight of the multispectral image set to 0.4.

[0110] A 20-dimensional feature vector was extracted from candidate hotspot regions, including grayscale features, texture features (such as gray-level co-occurrence matrix features), and spectral features (such as reflectance in different bands). Feature data containing 1000 hotspot samples and 1000 normal samples was collected to train a support vector machine (SVM) classification model, using radial basis function (RBF) as the kernel function. The trained model achieved a 98% accuracy rate in classifying hotspots.

[0111] 6. Implementation of dynamic tracking and trend prediction of hotspots

[0112] For detected hot spots, a Kalman filter algorithm is used for dynamic tracking. Based on the pixel coordinates and movement speed of the hot spot in the image, its position in the next frame is predicted, and the prediction model is continuously updated to achieve accurate tracking of the hot spot, with the tracking error controlled within 2 pixels.

[0113] Temperature change data of the superheated hotspot over the past hour were collected, with temperature values ​​recorded every 10 seconds, resulting in a total of 360 data points. The ARIMA(1,1,1) model was used to analyze and predict the temperature change trend of the superheated hotspot over the next 30 minutes.

[0114] 7. Implementation of Early Warning and Decision Support

[0115] A temperature warning threshold for overheating is set. When the temperature exceeds 80℃, a Level 1 warning is issued, and an alarm is triggered in the monitoring room via an audible and visual alarm device. When the temperature exceeds 90℃, a Level 2 warning is issued, and in addition to audible and visual alarms, relevant staff are notified via SMS and email.

[0116] Based on the location, type, and temperature trend of the overheated spot, and in conjunction with the emergency plan of the energy storage power station, handling recommendations are generated. For example, when an overheated spot is detected in a single cell of the battery pack and the temperature continues to rise, it is recommended to immediately disconnect the power supply to that cell and arrange for professional personnel to carry out maintenance.

[0117] Figure 3The overall architecture of a dynamic monitoring system for hotspots in an energy storage power station based on high-precision machine vision was demonstrated. The system adopts a distributed deployment approach, strategically placing multiple high-precision industrial cameras (resolution no less than 4K, frame rate no less than 30fps) within the energy storage power station. These cameras are equipped with LED auxiliary lighting and multispectral imaging devices, and connected to a data processing server via gigabit Ethernet. The data processing server integrates a complete image processing and analysis algorithm chain, including image acquisition and preprocessing, YOLO-based device feature extraction, U-Net-based hotspot identification, SVM-based accurate judgment, Kalman filter-based dynamic tracking, and an ARIMA-based trend prediction module. The monitoring and early warning system provides a real-time monitoring interface, supports multi-level early warning mechanisms (Level 1 warning at 80℃, Level 2 warning at 90℃), and sends alarm notifications through various methods such as audible and visual alarms, SMS, and email. It also provides intelligent decision support for staff, generating equipment maintenance suggestions and emergency response plans.

[0118] Through the above specific implementation methods, the dynamic monitoring method for overheated hot spots in energy storage power stations based on high-precision machine vision of the present invention can effectively achieve high-precision, real-time, and dynamic monitoring of overheated hot spots in energy storage power stations, thereby improving the safe operation level of energy storage power stations.

[0119] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-described technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for dynamic monitoring of overheated spots in an energy storage power station based on high-precision machine vision, characterized in that, The method includes: Based on the layout of the energy storage power station equipment, an image acquisition unit is arranged, which includes a high-precision imaging device, an auxiliary lighting device, and a multispectral imaging device. The image acquisition unit synchronously acquires the original visual images and multispectral images of the target equipment in the energy storage power station through the high-precision imaging device and the multispectral imaging device at a set time interval or triggering condition. The image acquisition unit is connected to the data processing server via gigabit Ethernet; it performs preprocessing operations on the acquired raw visual images to improve the clarity of device details in the images; A deep learning object detection algorithm is used to extract and locate device features in the preprocessed image, and output the device category information and its position coordinates in the image. A deep learning image segmentation algorithm is used to divide the target device region in the preprocessed image into sub-regions based on the location coordinates, calculate the feature parameters of each sub-region, and mark the hotspot candidate regions by threshold determination based on the threshold range of the feature parameters of the normal device region. The original visual image and the multispectral image are fused together. The multidimensional feature vectors of the candidate regions of the hotspots in the fused image are extracted. Based on the classification model, the result of whether the candidate regions are hotspots is output. The Kalman filter target tracking algorithm is used to dynamically track the confirmed hot spots, and the time series analysis method is used to predict the trend of temperature change data of the hot spots. Set different levels of overheating temperature warning thresholds, and issue a warning when the detected overheating temperature reaches or exceeds the warning threshold.

2. The method for dynamic monitoring of overheated spots in an energy storage power station based on high-precision machine vision according to claim 1, characterized in that, The high-precision imaging device is a high-precision industrial camera with a resolution of no less than 4K and a frame rate of no less than 30fps, and is equipped with an adjustable focal length optical lens; the auxiliary lighting device is an LED lighting device with adjustable light intensity and angle, and has adaptive adjustment function for light intensity and angle; the spectral imaging device has a collection range including visible light and near-infrared bands.

3. The method for dynamic monitoring of overheated spots in an energy storage power station based on high-precision machine vision according to claim 1, characterized in that, The preprocessing operation on the acquired raw visual image to improve the clarity of device details in the image includes: The median filtering algorithm is used to denoise the original visual image, converting the color image into a grayscale image, and histogram equalization is used to enhance the image contrast, thereby improving the clarity of device details in the image.

4. The method for dynamic monitoring of overheated spots in an energy storage power station based on high-precision machine vision according to claim 1, characterized in that, The deep learning object detection algorithm is a YOLO series algorithm. The model is trained using a training set labeled with device category and location information, enabling the model to identify the device used for image acquisition.

5. The method for dynamic monitoring of overheated spots in an energy storage power station based on high-precision machine vision according to claim 1, characterized in that, The deep learning image segmentation algorithm is the U-Net algorithm, and the threshold is determined by statistical analysis of the gray-scale mean and variance of a large number of normal regions and overheated regions.

6. The method for dynamic monitoring of overheated spots in an energy storage power station based on high-precision machine vision according to claim 1, characterized in that, A weighted average-based image fusion algorithm is used to fuse the original visual image with a multispectral image. The multidimensional feature vector includes grayscale features, texture features, and spectral features, and the classification model is a support vector machine model.

7. The method for dynamic monitoring of overheated spots in an energy storage power station based on high-precision machine vision according to claim 1, characterized in that, The Kalman filter target tracking algorithm can combine the time interval between image frames and predict the position of the hot spot in the next frame image based on the position and motion state of the hot spot in the current frame image. By continuously updating and correcting the predicted value, it can achieve accurate tracking of the hot spot. The time series analysis method is an autoregressive integral moving average model, which can establish an autoregressive integral moving average model based on historical temperature data to predict the future temperature change trend of the hot spot.

8. The method for dynamic monitoring of overheated spots in an energy storage power station based on high-precision machine vision according to claim 1, characterized in that, The system sets different levels of overheating temperature warning thresholds, and performs warning processing when the detected overheating temperature reaches or exceeds the warning threshold, including: Two levels of warning thresholds are set: when the temperature exceeds 80℃, the first level of warning is activated and an alarm is issued through an audible and visual alarm device; when the temperature exceeds 90℃, the second level of warning is activated and staff are notified through audible and visual alarms, SMS, and email.

9. The method for dynamic monitoring of overheated spots in an energy storage power station based on high-precision machine vision according to claim 1, characterized in that, The method also includes generating corresponding treatment measures based on the location, type, and temperature change trend information of the overheated hotspots, combined with the operating status and emergency plan of the energy storage power station.

10. A dynamic monitoring device for overheating in an energy storage power station based on high-precision machine vision, characterized in that, The device includes: an image acquisition unit, which is distributed based on the layout of energy storage power station equipment, and includes a high-precision industrial camera, adaptive auxiliary lighting equipment and multispectral imaging equipment; The data communication module uses the gigabit Ethernet protocol to achieve high-speed data transmission between the image acquisition module and the data processing server, and supports real-time image streaming. The image preprocessing module is used to preprocess the acquired raw visual images to improve the clarity of device details in the images; The device identification and positioning module integrates a deep learning target detection algorithm to extract and locate device features from preprocessed images, and outputs the device category information and its position coordinates in the image. The hotspot candidate identification module is used to use a deep learning image segmentation algorithm to divide the target device region in the preprocessed image into sub-regions based on the location coordinates, calculate the feature parameters of each sub-region, and mark the hotspot candidate regions by threshold determination. The multispectral fusion judgment module is used to fuse the original visual image and the multispectral image, extract the grayscale, texture and spectral feature vectors of the candidate region, and output the hot spot judgment result based on the support vector machine classification model. The dynamic tracking and prediction module integrates a Kalman filter tracking algorithm and a time series analysis model to track the movement trajectory of confirmed hotspots and predict temperature change trends. The intelligent early warning decision module is used to set different levels of overheating temperature warning thresholds. When the detected overheating temperature reaches or exceeds the warning threshold, an early warning is issued.

Images