A camera control parameter automatic adjustment method for intelligent patrol of a substation
By integrating multi-source data and deep learning models, the parameters of substation cameras are dynamically adjusted, solving the problem of unstable monitoring of cameras in extreme environments and dynamic scenarios in existing technologies, and achieving high-precision equipment status capture and continuous monitoring.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- STATE GRID SHANGHAI MUNICIPAL ELECTRIC POWER CO
- Filing Date
- 2026-03-03
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies have failed to effectively address extreme weather and dynamic environmental changes in substation camera parameter adjustments, resulting in unstable image quality, difficulty in capturing equipment defects, poor monitoring continuity, and a lack of autonomous optimization capabilities due to reliance on a single sensor and shallow learning models.
By integrating multi-source data to construct cross-modal fusion feature vectors, using deep learning models to extract associated features, dynamically adjusting camera parameters through self-supervised learning, establishing spatial mapping relationships between cameras, updating model parameters in real time to adapt to environmental changes, and calling surrounding cameras to continue monitoring in case of failure.
It improves the accuracy and robustness of camera parameter adjustment, enables stable monitoring in complex environments, automatically adapts to equipment aging and environmental changes, ensures monitoring continuity and image quality, and solves the problems of image noise and equipment detail in extreme environments.
Smart Images
Figure CN122437983A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent operation and maintenance technology for power systems, and in particular to a method for automatically adjusting the control parameters of cameras used for intelligent inspection of substations. Background Technology
[0002] As the core hub of the smart grid, substations rely on efficient intelligent inspection technology for equipment status monitoring. Chinese patent CN118504082A discloses a substation camera configuration method and device based on digital twin technology. This method uses digital twin technology to construct a digital twin 3D scene by collecting substation data, setting the coordinate path group of camera inspection points, determining the camera type and quantity based on regional data, and configuring a deployment scheme based on preset parameters, including installation height and angle, and incorporating anti-interference measures into the deployment scheme. However, this method only implements the camera deployment scheme and does not consider camera parameter control. Chinese patent CN120201286A discloses a substation camera adjustment method based on a sharpness evaluation algorithm. This method matches the parameters of a simulated camera with those of a real camera and utilizes an automatic adjustment system, including an intelligent detection and tracking module, a behavior recognition and analysis module, and a light adaptation module, to automatically adjust the camera's focal length, angle, and exposure parameters. However, this method relies on a single visible light sensor and does not consider the impact of extreme weather, making it unable to cope with complex environments such as heavy rain, strong light, and dense fog. The images exhibit significant noise in low light conditions and overexposed details of the equipment in strong light. It employs a shallow machine learning model, lacking autonomous optimization capabilities and unable to adapt to dynamic scenarios such as equipment aging and environmental changes, requiring manual intervention to correct parameters. It does not effectively integrate multi-source data such as infrared, temperature, humidity, and wind speed, resulting in low precision in parameter adjustment and difficulty in capturing hidden defects such as high temperatures and cracks in the equipment. The multi-camera linkage is statically pre-configured, lacking dynamic redundancy compensation in case of failure, making it easy to interrupt the continuity of inspections. Summary of the Invention
[0003] The purpose of this invention is to overcome the shortcomings of the existing technology and provide an automatic adjustment method for the control parameters of cameras used in intelligent inspection of substations. This method integrates multi-source data to construct environmental feature vectors to achieve multi-dimensional dynamic adjustment of parameters, uses self-supervised learning to extract associated features to enhance robustness in complex environments, collects and updates model parameters in real time to avoid human delay, and establishes camera spatial mapping for collaborative parameter adjustment, effectively improving environmental adaptability and system reliability, and ensuring continuous monitoring.
[0004] The objective of this invention can be achieved through the following technical solutions: According to a first aspect of the present invention, a method for automatically adjusting control parameters of a camera used in intelligent inspection of a substation is provided, the method comprising the following steps: Multimodal data, including visible light images, infrared images, and sensor data, is collected, preprocessed, and fused across modes to obtain cross-modal fusion feature vectors. Based on these cross-modal fusion feature vectors, environmental and equipment status analysis is performed to obtain the environmental complexity level. Abnormal areas are identified based on infrared or visible light images. The environmental complexity level and abnormal areas are input into the three-dimensional digital twin model of the substation to generate candidate control parameter combinations. After screening based on the image simulation clarity, the combinations are input into the PPO model and iteratively optimized to generate the optimal control parameter set. Based on the optimal control parameter set, the camera parameters are adjusted and the camera is controlled to acquire images. The image quality is judged. If the image quality is unqualified, the optimal control parameter set is regenerated. Otherwise, the PPO model parameters are updated based on the adjusted relevant data.
[0005] The data preprocessing specifically involves: removing image noise from visible light and infrared images, removing outliers from sensor data, and achieving spatiotemporal synchronization between infrared and visible light images based on timestamp alignment.
[0006] The cross-modal fusion specifically involves: extracting image feature vectors from visible light and infrared images using a deep learning model, and strengthening the feature weights of key areas of the device using an attention mechanism; and standardizing the preprocessed sensor data to obtain sensor feature vectors. A weighted summation formula is used to perform cross-modal feature fusion on the image feature vectors of visible light images and infrared images, as well as the sensor feature vectors, and output the cross-modal fused feature vector.
[0007] The environmental and equipment status analysis based on the cross-modal fusion feature vector specifically refers to: The cross-modal fused feature vector is input into the CNN classification model, which outputs the environment type and environment complexity level. The formula for calculating the environment complexity level is as follows: ; in, For environmental characteristics; For the first Normalized probabilities of dimensional features; This represents the level of environmental complexity.
[0008] The specific steps for identifying abnormal regions based on infrared or visible light images are as follows: Based on the temperature information contained in the infrared image, it is determined whether the temperature exceeds a preset temperature anomaly threshold range. The formula for calculating the temperature anomaly threshold is as follows: ; in, For statistical confidence coefficients; The standard deviation of the equipment's historical operating temperature; This is the temperature anomaly threshold. The rated operating temperature of the equipment; Defects in visible light images are detected based on the YOLOv8 model, and the proportion of the defect area to the visible light image is calculated. If the temperature of a certain area exceeds the preset range of the temperature anomaly threshold, or if the proportion of the defect area to the visible light image exceeds the preset value, it will be marked as an abnormal area.
[0009] In the PPO model, the state space is defined as environment type, environment complexity level, and abnormal region; the action space is defined as focal length, exposure time, ISO, and image stabilization level; and the reward function is defined as a weighted sum of image sharpness index, device anomaly recognition accuracy, and parameter adjustment response speed. The parameter adjustment response speed is calculated as follows: ; in, Adjust the response speed for the parameters, with a value range of [value range missing]. ; The actual adjustment time for the parameters; This represents the ideal response time.
[0010] The specific steps of adjusting camera parameters and controlling the camera to acquire images based on the optimal control parameter set are as follows: The system receives the optimal control parameter set and sends it to the physical cameras via a communication protocol. After the camera MCU completes the hardware adjustment of the parameters, it feeds back an adjustment completion signal and hardware status, and acquires the adjusted image. If a camera reports a fault signal, the system calculates the optimal viewing angle of adjacent cameras using a multi-camera spatial mapping formula, adjusts the parameters of adjacent cameras to cover the monitoring blind zone, and outputs the adjusted image, adjustment completion signal, hardware status, and fault signal. The multi-camera spatial mapping formula is as follows: ; in, The coordinates of the monitoring point of the faulty camera; These are the corresponding coordinates of adjacent cameras; It is a 3×3 perspective transformation matrix.
[0011] The process involves judging image quality. If the image quality is unsatisfactory, the optimal control parameter set is regenerated; otherwise, the PPO model parameters are updated based on the adjusted relevant data. Specifically: The image quality acquired by the camera after parameter adjustment is determined based on the sharpness and SSIM value, and the YOLOv8 model is used to detect abnormal areas in the image to confirm the validity of the parameters. If the image quality is acceptable and the accuracy of abnormal area identification is up to standard, the correlation data of scene-control parameter-effect is written into the closed-loop feedback dataset. Based on the closed-loop feedback dataset and the current parameters of the PPO model, when the cumulative amount of newly added correlation data in the closed-loop feedback dataset reaches a preset number, the PPO model parameters are updated and the reward function weights are adjusted using mini-batch gradient descent. At the same time, the optimal control parameter set of the new scene and the corresponding environment type and device type are added to the historical strategy library. The KNN algorithm is used to achieve fast parameter matching and fine-tuning of parameters for similar scenes. If the image quality is unsatisfactory or the accuracy of abnormal area recognition is not up to standard, the control parameter set is regenerated using the PPO model, and the historical strategy library is called to fine-tune the generated control parameter set, update the optimal control parameter set, and control the camera to acquire images again.
[0012] The method further includes: Choose between full or incremental update of the strategy library based on the proportion of new scenarios, and perform lightweight model iterations periodically during device operation to output updated PPO model parameters and a new strategy library.
[0013] The method further includes: If the environmental complexity level meets the standard, the cross-modal enhancement strategy is activated and the new exposure time is calculated according to the extreme environment exposure compensation formula; the extreme environment exposure compensation formula is: ; in, The optimal exposure time under normal conditions; Ideal light intensity; This represents the current light intensity. This is an environmental correction factor; If the camera hardware fails, an emergency mechanism is triggered, and the backup camera takes over the monitoring task. The parameters are adjusted based on the spatial mapping of multiple cameras, and alarm information is sent to the operation and maintenance platform at the same time.
[0014] According to a second aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method described thereon.
[0015] According to a third aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.
[0016] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention integrates visible light, infrared, sensor and other data to construct cross-modal fusion feature vectors, realizes multi-dimensional dynamic adjustment of parameters, considers a variety of factors, and has high parameter adjustment accuracy, and can effectively capture hidden defects such as high temperature and cracks in equipment. (2) The present invention uses a deep learning model (self-supervised learning) to extract the correlation features of multi-sensor data, which can improve the robustness in complex environments; (3) The present invention collects camera feedback data (such as recognition results and user annotations) in real time and dynamically updates model parameters, which can automatically adapt to dynamic scenarios such as equipment aging and environmental changes, and avoid the lag of manual rules; (4) The present invention establishes a spatial mapping relationship between cameras and achieves collaborative parameter tuning through image registration and feature matching. When a camera fails, it can call on surrounding cameras to continue shooting, ensuring the continuity of monitoring. (5) The present invention initiates a cross-modal enhancement strategy in complex environments and calculates the new exposure time based on the exposure compensation formula for extreme environments. It can cope with complex environments such as rainstorms, strong light, and dense fog, and solve the problems of significant image noise under low light and overexposure of device details under strong light. Attached Figure Description
[0017] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation
[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0019] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms “a,” “an,” “an,” “the,” and similar words used in this application do not indicate quantity limitation and may indicate singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms “connected,” “linked,” “coupled,” and similar words used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following objects are in an "or" relationship. The terms "first," "second," and "third" used in this application are merely to distinguish similar objects and do not represent a specific ordering of the objects.
[0020] Example 1 This embodiment provides a method for automatically adjusting the control parameters of a camera used in intelligent substation inspections, such as... Figure 1 As shown, the method includes the following steps: S1: Collect multimodal data including visible light images, infrared images, and sensor data, and perform data preprocessing and cross-modal fusion to obtain cross-modal fusion feature vectors.
[0021] In this embodiment, a visible light camera, an infrared thermal imager, and an environmental sensor are selected as the data acquisition devices. The acquired visible light images, infrared images, and sensor data are stored in a preset format to obtain multimodal data. Specifically, the visible light camera has a resolution ≥4K, a frame rate of 25fps, and a focal length range of 4-12mm; the infrared thermal imager has a temperature measurement range of -20~300℃ and a temperature measurement accuracy of ±0.5℃; the environmental sensor has a temperature and humidity accuracy of ±0.1℃ / ±2%RH and a wind speed measurement range of 0-30m / s; a fixed camera acquires data in real-time at 1s / frame, while an inspection robot / drone acquires data every 50cm along a preset path; a digital twin system synchronizes physical data and updates virtual data every 100ms. Visible light images are stored in JPEG format, infrared images in TIFF format, and sensor data in CSV format. The CSV format fields include "acquisition timestamp - device ID - data type - value - location coordinates"; the output is the original multimodal data.
[0022] After removing image noise from visible light and infrared images and eliminating outliers from sensor data, spatiotemporal synchronization between infrared and visible light images is achieved based on timestamp alignment. In one embodiment, image noise removal uses a median filter with a window size of 3×3; the timestamp alignment error is less than 10ms.
[0023] Image feature vectors are extracted from visible light and infrared images using deep learning models, and an attention mechanism is employed to enhance the feature weights of key areas of the device. In this embodiment, a ResNet50 model is used to extract 2048-dimensional texture features from the visible light image, and a MobileNet model is used to extract 1024-dimensional temperature distribution features from the infrared image. An attention mechanism is then used to enhance the feature weights of key areas of the device (such as contacts and oil level gauges). The preprocessed sensor data is standardized and mapped to the [0,1] interval to eliminate the influence of dimensions, resulting in sensor feature vectors.
[0024] A weighted summation formula is used to perform cross-modal feature fusion on the image feature vectors of visible light images and infrared images, as well as the sensor feature vectors, to output a cross-modal fused feature vector: ; in, This is a cross-modal fusion feature vector; The visible light image feature vector; For infrared image feature vectors; The sensor feature vector; , , The weight is denoted as , and its value is . , , .
[0025] S2 performs environmental and equipment status analysis based on cross-modal fusion feature vectors to obtain the environmental complexity level, and identifies abnormal areas based on infrared or visible light images.
[0026] The cross-modal fused feature vector is input into the CNN classification model, which outputs the environment type and environment complexity level. The formula for calculating the environment complexity level is as follows: ; in, Environmental characteristics (such as light, humidity, and wind speed); For the first Normalized probabilities of dimensional features; In this embodiment, the environmental complexity level is defined as follows: .
[0027] Based on the temperature information contained in the infrared image, it is determined whether the temperature exceeds a preset temperature anomaly threshold range. The formula for calculating the temperature anomaly threshold is as follows: ; in, In this embodiment, to calculate the confidence coefficient, ; The standard deviation of the equipment's historical operating temperature; This is the temperature anomaly threshold. The rated operating temperature of the equipment.
[0028] Meanwhile, based on the YOLOv8 model, defects in the visible light image are detected, and the proportion of the defect area to the visible light image is calculated. In this embodiment, the detection confidence of the YOLOv8 model is ≥0.7.
[0029] If the temperature of a certain area exceeds a preset temperature anomaly threshold, or if the proportion of the defect area to the visible light image exceeds a preset value, it is marked as an abnormal area. In one embodiment, if the temperature of a certain area... Or the percentage of defect area in a visible light image. If so, it is marked as an abnormal region. .
[0030] S3 inputs the environmental complexity level and abnormal areas into the three-dimensional digital twin model of the substation to generate candidate control parameter combinations.
[0031] In this embodiment, there are specifically 5 sets of candidate control parameter combinations, among which focal length The exposure times are: 4mm (wide-angle coverage), 6mm (mid-range monitoring), 8mm (device detail observation), 10mm (long-distance close-up), and 12mm (ultra-close-up detail capture). The timeframes are: 10ms (strong light scene), 20ms (normal light), 30ms (cloudy day), 40ms (dusk), and 50ms (low light).
[0032] S4: After screening candidate control parameter combinations based on image simulation clarity, input them into the PPO model to iteratively generate a preliminary optimal parameter set; call the strategy library to fine-tune the preliminary optimal parameter set and output the optimal control parameter set.
[0033] The method for calculating the simulated sharpness of an image is as follows: ; in, To simulate the structural similarity between the image and a standard clear image; This is to simulate the edge intensity of an image.
[0034] Calculate the simulated image sharpness for each set of parameters. ,filter The candidate set is input into the PPO model.
[0035] In the PPO model, the state space is set as the environment type. Environmental complexity level and abnormal areas ,Right now The motion space is set to the focal length. Exposure time ISO And image stabilization level, i.e. The reward function is set as a weighted sum of image sharpness index, device anomaly detection accuracy, and parameter adjustment response speed: ; in, , , As weight; Image sharpness index; To improve the accuracy of equipment malfunction identification. Adjust the response speed for the parameters.
[0036] The formula for calculating sharpness is as follows: ; in, Image sharpness index; Image pixel size; For image The second Laplace derivative of a pixel.
[0037] The formula for calculating the SSIM value is: ; in, The image after adjustment; For standard clear images; , This represents the mean of the corresponding image; , The standard deviation of the corresponding image; Let be the covariance of the two images; , It is a constant.
[0038] The formula for calculating the response speed of parameter adjustment is: ; in, The range of values is ; The actual adjustment time for the parameters; This represents the ideal response time.
[0039] In this embodiment, the constant .
[0040] The system calls a preset strategy library, whose key-value pair format is device type-inspection task-optimal parameters (e.g., "transformer-oil level check" corresponds to a focal length of 8mm and an exposure time of 30ms). Based on the current inspection scenario, it matches the corresponding parameters. Make fine adjustments; for example, if the current parameter Then fine-tune to .
[0041] S5 adjusts camera parameters and controls the camera to acquire images based on the optimal set of control parameters.
[0042] The system receives the optimal control parameter set and sends it to the physical camera via the MQTT communication protocol (transmission delay ≤20ms). After the camera's MCU completes hardware adjustments to parameters such as focus and exposure, it sends back an adjustment completion signal. and hardware status Acquire and adjust the image If a camera reports a fault signal (For example, if there is no response for 3 consecutive times), the optimal viewing angle of adjacent cameras is calculated using the multi-camera spatial mapping formula. The parameters of the adjacent cameras are adjusted to cover the monitoring blind spot, and the adjusted image, adjustment completion signal, hardware status, and fault signal are output. The multi-camera spatial mapping formula is as follows: ; in, The coordinates of the monitoring point of the faulty camera; These are the corresponding coordinates of adjacent cameras; It is a 3×3 perspective transformation matrix.
[0043] S6. Determine the image quality. If the image quality is unacceptable, regenerate the optimal control parameter set. Otherwise, update the PPO model parameters based on the adjusted relevant data.
[0044] In this embodiment, the image quality captured by the camera after parameter adjustment is determined based on sharpness and SSIM value. and The image quality was deemed acceptable.
[0045] Simultaneously, the YOLOv8 model is used to detect abnormal regions in the image to confirm the validity of the parameters; if the accuracy of the YOLOv8 model in detecting abnormal regions is high... If the accuracy of abnormal area identification meets the standard, then the abnormal area identification accuracy is determined to be satisfactory. In one embodiment, further, if both the abnormal area identification accuracy and the positioning deviation (less than 2 pixels) meet the requirements, then the parameters are confirmed to be valid; otherwise, the reason for failure is recorded.
[0046] If the image quality is satisfactory and the accuracy of abnormal region identification meets the standard, then the correlation data of scene-control parameters-effect is written into the closed-loop feedback dataset. Based on the closed-loop feedback dataset When the cumulative amount of newly added associated data in the closed-loop feedback dataset reaches a preset number, the PPO model parameters are updated using mini-batch gradient descent and the reward function weights are adjusted. At the same time, the optimal control parameter set of the new scenario and the corresponding environment type and device type are added to the historical policy library. The KNN algorithm is used to achieve fast parameter matching and fine-tuning of similar scenarios.
[0047] For example, the initial parameters of the PPO model are For every 1000 associated data points accumulated, the neural network parameters of the PPO model are updated using mini-batch gradient descent, and the reward function weights are adjusted. The formula for updating the neural network parameters is: ; in, No. Wheel model parameters; The learning rate; Let be the objective function of the PPO algorithm.
[0048] Add the optimal control parameter set and corresponding environment and equipment types for new scenarios (such as new equipment or rare weather) to the historical strategy database. The KNN algorithm (k=5) is used to achieve fast parameter matching for similar scenes; the similarity matching formula of the policy library is: ; in, For current scene characteristics (such as environment type, device type); Features of historical scenarios in the strategy library; For Euclidean distance, choose The average of the five smallest historical scenarios was used as the initial candidate values.
[0049] If a new scene is added Deviation from the mean If the deviation is found, it will be retained directly; otherwise... Then fine-tune To ensure parameter compatibility, the value should be close to the mean.
[0050] When the number of new scenarios accounts for Total number of scenes Trigger a full update: Generate a new strategy library. , covering the original At the same time, back up the original .
[0051] During operation, the device collects and provides feedback data in real time, and performs a lightweight model iteration once per hour (iteration rounds = 20) to adapt to dynamic situations such as camera lens aging and long-term environmental changes.
[0052] If the image quality is unsatisfactory or the accuracy of abnormal area recognition is not up to standard, the control parameter set is regenerated using the PPO model, and the historical strategy library is called to fine-tune the generated control parameter set, update the optimal control parameter set, and control the camera to acquire images again.
[0053] Example 2 This embodiment provides a further improved solution for extreme environmental scenarios based on embodiment 1.
[0054] If the environmental complexity level meets the standard, the cross-modal enhancement strategy is activated and the new exposure time is calculated based on the extreme environment exposure compensation formula.
[0055] In highly complex environments ( The system automatically initiates a cross-modal enhancement strategy, fusing infrared and visible light images (using a Generative Adversarial Network (GAN)). The GAN then fuses the infrared and visible light images and calculates the new exposure time based on an exposure compensation formula for extreme environments. This ensures that image details are discernible. The environmental exposure compensation formula is as follows: ; in, The optimal exposure time under normal conditions; Ideal light intensity; This represents the current light intensity. This is an environmental correction factor for heavy rain. Dense fog Strong light .
[0056] Example 3 This embodiment provides an improved solution for camera malfunction scenarios, based on Embodiment 1.
[0057] If the camera hardware fails (such as lens jamming or communication interruption), an emergency mechanism is triggered within 10ms, and the backup camera takes over the monitoring task. The parameters (focal length and gimbal angle) are adjusted based on the spatial mapping of multiple cameras, and alarm information is sent to the operation and maintenance platform at the same time.
[0058] Example 4 In one embodiment, the electronic device includes a computing unit that can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) or a computer program loaded from a storage unit into a random access memory (RAM). The RAM may also store various programs and data required for device operation. The computing unit, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.
[0059] Multiple components in an electronic device are connected to an I / O interface, including: input units such as keyboards and mice; output units such as various types of displays and speakers; storage units such as disks and optical discs; and communication units such as network interface cards (NICs), modems, and wireless transceivers. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.
[0060] The computing unit can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of computing units include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit performs the various methods and processes described above, such as methods S1-S6. For example, in some embodiments, methods S1-S6 may be implemented as computer software programs tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via ROM and / or a communication unit. When the computer program is loaded into RAM and executed by the computing unit, one or more steps of methods S1-S6 described above may be performed. Alternatively, in other embodiments, the computing unit may be configured to perform methods S1-S6 by any other suitable means (e.g., by means of firmware).
[0061] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.
[0062] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.
[0063] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.
[0064] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A method for automatically adjusting control parameters of a camera used in intelligent substation inspection, characterized in that, The method includes the following steps: Multimodal data, including visible light images, infrared images, and sensor data, is collected, preprocessed, and fused across modes to obtain cross-modal fusion feature vectors. Based on these cross-modal fusion feature vectors, environmental and equipment status analysis is performed to obtain the environmental complexity level. Abnormal areas are identified based on infrared or visible light images. The environmental complexity level and abnormal areas are input into the three-dimensional digital twin model of the substation to generate candidate control parameter combinations. After screening based on the image simulation clarity, the combinations are input into the PPO model and iteratively optimized to generate the optimal control parameter set. Based on the optimal control parameter set, the camera parameters are adjusted and the camera is controlled to acquire images. The image quality is judged. If the image quality is unqualified, the optimal control parameter set is regenerated. Otherwise, the PPO model parameters are updated based on the adjusted relevant data.
2. The method for automatically adjusting control parameters of a camera used for intelligent inspection of substations according to claim 1, characterized in that, The data preprocessing specifically involves: removing image noise from visible light and infrared images, removing outliers from sensor data, and achieving spatiotemporal synchronization between infrared and visible light images based on timestamp alignment.
3. The method for automatically adjusting control parameters of a camera used for intelligent inspection of substations according to claim 1, characterized in that, The cross-modal fusion specifically involves: extracting image feature vectors from visible light and infrared images using a deep learning model, and strengthening the feature weights of key areas of the device using an attention mechanism; and standardizing the preprocessed sensor data to obtain sensor feature vectors. A weighted summation formula is used to perform cross-modal feature fusion on the image feature vectors of visible light images and infrared images, as well as the sensor feature vectors, and output the cross-modal fused feature vector.
4. The method for automatically adjusting control parameters of a camera used for intelligent inspection of substations according to claim 1, characterized in that, The environmental and equipment status analysis based on the cross-modal fusion feature vector specifically refers to: The cross-modal fused feature vector is input into the CNN classification model, which outputs the environment type and environment complexity level. The formula for calculating the environment complexity level is as follows: ； in, For environmental characteristics; For the first Normalized probabilities of dimensional features; This represents the level of environmental complexity.
5. The method for automatically adjusting control parameters of a camera used for intelligent inspection of substations according to claim 1, characterized in that, The specific steps for identifying abnormal regions based on infrared or visible light images are as follows: Based on the temperature information contained in the infrared image, it is determined whether the temperature exceeds a preset temperature anomaly threshold range. The formula for calculating the temperature anomaly threshold is as follows: ； in, For statistical confidence coefficients; The standard deviation of the equipment's historical operating temperature; This is the temperature anomaly threshold. The rated operating temperature of the equipment; Defects in visible light images are detected based on the YOLOv8 model, and the proportion of the defect area to the visible light image is calculated. If the temperature of a certain area exceeds the preset range of the temperature anomaly threshold, or if the proportion of the defect area to the visible light image exceeds the preset value, it will be marked as an abnormal area.
6. The method for automatically adjusting control parameters of a camera used for intelligent inspection of substations according to claim 1, characterized in that, In the PPO model, the state space is defined as environment type, environment complexity level, and abnormal region; the action space is defined as focal length, exposure time, ISO, and image stabilization level; and the reward function is defined as a weighted sum of image sharpness index, device anomaly recognition accuracy, and parameter adjustment response speed. The parameter adjustment response speed is calculated as follows: ； in, Adjust the response speed for the parameters, with a value range of [value range missing]. ; The actual adjustment time for the parameters; This represents the ideal response time.
7. The method for automatically adjusting control parameters of a camera used for intelligent inspection of substations according to claim 1, characterized in that, The specific steps of adjusting camera parameters and controlling the camera to acquire images based on the optimal control parameter set are as follows: The system receives the optimal control parameter set and sends it to the physical camera via a communication protocol. After the camera MCU completes the hardware adjustment of the parameters, it feeds back the adjustment completion signal and hardware status, and then acquires the adjusted image. If a camera reports a fault signal, the optimal viewing angle of adjacent cameras is calculated using a multi-camera spatial mapping formula. The parameters of these adjacent cameras are then adjusted to cover the monitoring blind spot. The adjusted image, adjustment completion signal, hardware status, and fault signal are output. The multi-camera spatial mapping formula is as follows: ； in, The coordinates of the monitoring point of the faulty camera; These are the corresponding coordinates of adjacent cameras; It is a 3×3 perspective transformation matrix.
8. The method for automatically adjusting control parameters of a camera used for intelligent inspection of substations according to claim 1, characterized in that, The process involves judging image quality. If the image quality is unsatisfactory, the optimal control parameter set is regenerated; otherwise, the PPO model parameters are updated based on the adjusted relevant data. Specifically: The image quality acquired by the camera after parameter adjustment is determined based on the sharpness and SSIM value, and the YOLOv8 model is used to detect abnormal areas in the image to confirm the validity of the parameters. If the image quality is acceptable and the accuracy of abnormal area identification is up to standard, the correlation data of scene-control parameter-effect is written into the closed-loop feedback dataset. Based on the closed-loop feedback dataset and the current parameters of the PPO model, when the cumulative amount of newly added correlation data in the closed-loop feedback dataset reaches a preset number, the PPO model parameters are updated and the reward function weights are adjusted using mini-batch gradient descent. At the same time, the optimal control parameter set of the new scene and the corresponding environment type and device type are added to the historical strategy library. The KNN algorithm is used to achieve fast parameter matching and fine-tuning of parameters for similar scenes. If the image quality is unsatisfactory or the accuracy of abnormal area recognition is not up to standard, the control parameter set is regenerated using the PPO model, and the historical strategy library is called to fine-tune the generated control parameter set, update the optimal control parameter set, and control the camera to acquire images again.
9. The method for automatically adjusting control parameters of a camera used for intelligent inspection of substations according to claim 8, characterized in that, The method further includes: Choose between full or incremental update of the strategy library based on the proportion of new scenarios, and perform lightweight model iterations periodically during device operation to output updated PPO model parameters and a new strategy library.
10. The method for automatically adjusting control parameters of a camera used for intelligent inspection of a substation according to claim 7, characterized in that, The method further includes: If the environmental complexity level meets the standard, the cross-modal enhancement strategy is activated and the new exposure time is calculated according to the extreme environment exposure compensation formula; the extreme environment exposure compensation formula is: ； in, The optimal exposure time under normal conditions; Ideal light intensity; This represents the current light intensity. This is an environmental correction factor; If the camera hardware fails, an emergency mechanism is triggered, and the backup camera takes over the monitoring task. The parameters are adjusted based on the spatial mapping of multiple cameras, and alarm information is sent to the operation and maintenance platform at the same time.