Power equipment monitoring system and method

Abstract

The present disclosure provides a power equipment monitoring system and method, which relates to the field of power equipment monitoring. The system includes a large language model module, an edge device processing module, a device portrait module, and a result merging module. The large language model module generates block instructions based on the image size, processable sub-block size, and historical fault information recorded in the device portrait, and can output customized image processing instructions for different risk areas; the edge device processing module only needs to perform block-by-block cropping and multi-mode detection on the block-by-block image to reduce the burden of single processing; the result merging module fuses the overlapping areas based on the sub-block analysis results and address mapping information, and finally obtains the fault identification or feature distribution of the complete high-resolution power equipment image. Utilizing this system, efficient block analysis and personalized deep detection of power equipment images can be achieved, while reducing resource investment and improving the defect detection rate.

Description

Technical Field

[0001] The present disclosure relates to the field of power equipment monitoring, and in particular to a power equipment monitoring system and method. Background Art

[0002] In power systems, power equipment (such as transformers, circuit breakers, insulators, and switchgear) operates under high voltage, high temperature, and complex environments for extended periods. These equipment is susceptible to failure due to various factors, including insulation aging, partial discharge, contamination accumulation, and mechanical fatigue. If these failures are not promptly detected and addressed, they can lead to widespread power outages or serious safety incidents. Therefore, stable and accurate monitoring of power equipment and fault prediction have always been core requirements in the power operations and maintenance field.

[0003] With the advancement of digital and intelligent technologies, fault detection based on image analysis has become an increasingly important tool for monitoring power equipment. For example, by collecting visible light or infrared thermal images at the operation site, operators can identify surface cracks, discharge marks, and areas of abnormal temperature, helping operators detect potential hazards early. Compared to traditional manual inspections, image analysis is more efficient and enables remote or batch monitoring. However, existing technologies still face several limitations in practical applications. For example, images of power equipment generally have a high resolution. Especially in ultra-high voltage (EHV) and ultra-high voltage (UHV) scenarios, a single image may contain thousands to tens of thousands of pixels to clearly observe local details. Processing all of these images at once often places excessive demands on hardware computing power and memory, leading to long processing times and insufficient memory. Furthermore, the failure modes, operating environments, and historical conditions of different power equipment vary significantly. A single, universal image analysis process often fails to address these diverse scenarios, making it difficult to control missed detections or false positives.

[0004] While some research has introduced advanced algorithms like deep learning, these approaches lack refined configuration for factors like fault risk, edge region processing, and multimodal image fusion, and remain unable to automatically adapt to different devices or fault types within a unified system. Existing solutions often rely on manual configuration or fixed scripts for image analysis. Changes in device operating conditions or increases in acquisition resolution require manual updates to algorithms and parameters, making it difficult to implement timely and efficient dynamic adjustments. Summary of the Invention

[0005] In response to the deficiencies in the prior art, the present application provides a power equipment monitoring system and method.

[0006] In a first aspect, the present application provides a power equipment monitoring system, comprising:

[0007] A large language model module performs a block strategy analysis on the target-resolution power equipment image based on the image size of the power equipment image obtained from the device portrait and the sub-block size that can be processed by the edge device to obtain a block instruction; wherein the large language model module includes: a large model trained and / or fine-tuned based on the power equipment scenario data;

[0008] An edge device processing module, based on the block instruction, performs block processing on the power equipment image, and analyzes only the image blocks in the area limited by the processable sub-block size each time to obtain a sub-block analysis result;

[0009] An equipment portrait module, configured to store image metadata, historical fault records, and processable sub-block size information of power equipment, and provide the image size and processable sub-block size to the large language model module;

[0010] A result merging module, which performs result splicing and fusion based on the sub-block analysis results and the address mapping corresponding to the sub-blocks in the block instruction, to obtain fault identification and / or feature extraction results for the power equipment image with full target resolution;

[0011] Among them, the large language model module determines the way in which the edge device processing module performs block processing on the power equipment image based on the image size and processable sub-block size provided by the device portrait module, and the result merging module maps the sub-block analysis results after block processing back to the original image coordinates, and finally records the overall detection results in the device portrait module.

[0012] As an optional implementation, the large language model module calculates the number of blocks based on the image size and the processable sub-block size and pads or splices the edge parts that cannot be divided evenly to generate corresponding block instructions.

[0013] As an optional implementation, the device portrait module is used to store at least one of the sub-block size that can be processed by the edge device, the historical fault hot zone distribution, the image resolution, and the multimodal sensor data;

[0014] The large language model module determines whether to adopt overlapping blocks for edge areas based on the processable sub-block size and the historical fault hot zone distribution.

[0015] As an optional implementation manner, the large language model module generates the block instruction including:

[0016] In response to the image size being divided by an integer of a processable sub-block size, directly dividing the entire power equipment image into a plurality of image blocks of the same size;

[0017] In response to the image size not being divisible by an integer of the processable sub-block size, the edge portion that cannot be divided evenly is padded with zeros or concatenated with an adjacent block, and valid area information of the edge sub-block is recorded in the block instruction.

[0018] As an optional implementation, the edge device processing module includes:

[0019] An image processing unit, configured to crop or splice image blocks of a processable sub-block size according to the block division instruction;

[0020] An image analysis unit, configured to perform defect detection, texture analysis, and segmentation operations on each image block;

[0021] The communication unit is used to upload the output data of the image analysis unit to the result merging module.

[0022] As an optional implementation, the result merging module takes the maximum or average value of the output results of adjacent sub-blocks with overlapping areas according to the fusion method specified in the block instruction. When multiple sub-blocks have different detection results at the same target pixel coordinates, a preset weighted merging strategy is used to obtain the final result.

[0023] As an optional implementation, the large language model module is deployed on a cloud server or computing platform, and the edge device processing module is deployed in the edge hardware at the power site. The two communicate through the network to issue instructions and transmit sub-block analysis results.

[0024] As an optional implementation, the large language model module automatically adjusts the image segmentation strategy based on industry standards and historical failure cases provided by the device portrait module, and dynamically generates new segmentation instructions when the processable sub-block size or image size changes.

[0025] As an optional implementation, the large language model module is further configured to, based on personalized information related to historical fault records, operating environment, and sensor data of the power equipment obtained from the device profiling module, generate an image processing method instruction for each sub-block after determining a block processing method for the power equipment image, and send the image processing method instruction to the edge device processing module, wherein the image processing method instruction includes:

[0026] Target algorithm type, used to indicate the image processing, defect detection, and texture analysis algorithms used for each sub-block;

[0027] Parameter configuration, used to indicate whether additional filtering, enhancement or segmentation threshold is performed under the target spectrum or temperature distribution;

[0028] Priority information, used to instruct the edge device processing module to execute a preset second detection process for sub-blocks whose failure risk is greater than or equal to a preset risk value, and to use a preset first detection process for sub-blocks whose failure risk is less than the preset risk value;

[0029] A multimodal fusion strategy is used to respond to the mapping relationship between the visible light image and the infrared image recorded in the device portrait module, and the image processing method instruction instructs the edge device processing module to perform cross-spectral or multi-channel data fusion on the corresponding sub-blocks.

[0030] In a second aspect, the present application provides a method for monitoring power equipment, comprising:

[0031] Using a large language model, based on the image size of the power equipment image obtained from the device portrait and the sub-block size that can be processed by the edge device, a block strategy analysis is performed on the target resolution power equipment image to obtain a block instruction; wherein the large language model includes: a large model trained and / or fine-tuned based on the power equipment scenario data;

[0032] Using the edge device, based on the block instruction, the power equipment image is processed in blocks, and each time only the image blocks in the area limited by the processable sub-block size are analyzed to obtain sub-block analysis results;

[0033] Using the device portrait, image metadata, historical fault records, and processable sub-block size information of the power equipment are stored, and the image size and processable sub-block size are provided to the large language model;

[0034] Based on the sub-block analysis results and the address mapping corresponding to the sub-blocks in the block instruction, performing result splicing and fusion to obtain fault identification and / or feature extraction results for the power equipment image with full target resolution;

[0035] Among them, the large language model determines the way in which the edge device divides the power equipment image into blocks based on the image size and processable sub-block size provided by the device portrait, maps the sub-block analysis results after division back to the original image coordinates, and records the overall detection results.

[0036] Compared with the existing technology, this application introduces the concepts of large language models and equipment profiling, and proposes an adaptive and scalable image analysis process for the generation of block strategies and fault risk assessment for high-resolution power equipment images. By combining image segmentation with personalized detection instructions, it not only reduces the dependence on hardware resources, but also enhances the processing capabilities of special fault modes and multimodal data. Based on existing image processing and power equipment operation and maintenance technologies, the present invention achieves more flexible and accurate fault monitoring and automated decision-making, which is of great value in ensuring the safe and stable operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A schematic diagram of a power equipment monitoring system provided in an embodiment of the present application;

[0038] Figure 2 A flowchart of a method for monitoring power equipment provided in an embodiment of the present application;

[0039] Figure 3 A schematic diagram of a maximum sub-block size in overall image analysis provided by an embodiment of the present application;

[0040] Figure 4 A schematic diagram of the location area of a historical fault in an overall image provided in an embodiment of the present application. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0042] See also Figure 1 As shown, the power equipment monitoring system provided in the embodiment of the present application includes:

[0043] The large language model module 10 performs a block strategy analysis on the target resolution power equipment image based on the image size of the power equipment image obtained from the device portrait and the sub-block size that can be processed by the edge device, and obtains a block instruction. The large language model module includes: a large model trained and / or fine-tuned based on the power equipment scenario data;

[0044] The edge device processing module 20 performs block processing on the power equipment image based on the block instruction, and analyzes only the image blocks in the area limited by the processable sub-block size each time to obtain sub-block analysis results;

[0045] The device portrait module 30 is used to store image metadata, historical fault records, and processable sub-block size information of the power equipment, and provide the image size and processable sub-block size to the large language model module;

[0046] A result merging module 40 performs result splicing and fusion based on the sub-block analysis results and the address mapping corresponding to the sub-blocks in the block instruction to obtain a fault identification and / or feature extraction result for the power equipment image with full target resolution;

[0047] Among them, the large language model module 10 determines the way in which the edge device processing module 20 performs block processing on the power equipment image based on the image size and processable sub-block size provided by the device portrait module 30, and the result merging module 40 maps the sub-block analysis results after block processing back to the original image coordinates, and finally records the overall detection results in the device portrait module 30.

[0048] As an optional implementation, the large language model module 10 calculates the number of blocks according to the image size and the processable sub-block size and pads or splices the edge parts that cannot be divided evenly to generate corresponding block instructions.

[0049] As an optional implementation, the device portrait module 30 is used to store at least one of the sub-block size that can be processed by the edge device, the historical fault hot zone distribution, the image resolution, and the multimodal sensor data;

[0050] The large language model module 10 determines whether to use overlapping blocks for edge areas based on the processable sub-block size and the historical fault hotspot distribution.

[0051] As an optional implementation manner, the large language model module 10 generates the block instruction including:

[0052] In response to the image size being divisible by an integer of a processable sub-block size, directly dividing the entire power equipment image into a plurality of image blocks of the same size;

[0053] In response to the image size being unable to be divided by the processable sub-block size, the edge portion that cannot be divided is padded with zeros or concatenated with adjacent blocks, and valid area information of the edge sub-block is recorded in the block instruction.

[0054] For the above-mentioned large language model module 10:

[0055] In a specific implementation, the large language model module 10 can obtain the image size of the current power equipment image (such as the target resolution of 4096×4096) and the sub-block size that the edge device can process (such as 512×512) from the device portrait module 30 through API / database query and other methods. If the device portrait also provides data such as historical fault hotspots or industry specifications, they can also be read together. This module can include a large model based on the Transformer structure, which has been pre-trained and / or fine-tuned on general corpus and power industry failure cases in the early stage to make it familiar with common failure modes of power equipment, image processing requirements, and scenarios with limited edge computing power.

[0056] For example, during inference, prompts can be input to the large model, including information such as "image resolution = 4096×4096", "processable sub-block size = 512×512", and "historical faults are concentrated in the lower right corner". The model will comprehensively process these contexts and produce a segmentation strategy.

[0057] It is understandable that the above prompt can be automatically extracted and executed by a script or other computer program.

[0058] For example, see Figure 4 , Figure 4 A schematic diagram of the location area of a historical fault in an overall image provided in an embodiment of the present application.

[0059] The historical fault area is concentrated in the sub-block in the lower right corner and is represented by a solid circle. The maximum sub-block size is 512×512, and the overall image size is 1536×1536.

[0060] In a specific implementation, the large language model module 10 is configured to decide whether to pad with zeros or splice the image based on whether the image can be divided evenly by processable sub-blocks; when the historical fault area is found to be located in the lower right corner, the model may give priority to recommending a more refined block division of the area or setting a higher detection priority; through prompts or internal rules, the model generates information on the coordinates, size, and edge processing methods of several sub-blocks.

[0061] The model outputs the above analysis results in a structured format (such as JSON), for example, if the edge pixel rows and columns are not divisible, they will be marked with "edge_handling": "padding" or "merge_with_block": "X".

[0062] For example, in an embodiment of the present application, when the large language model module 10 generates a block instruction, if it recognizes that the image size cannot be divided evenly by the processable sub-block size, it will attach an "edge_handling" field to each sub-block that cannot be divided evenly in the block instruction sent to the edge device processing module 20 to indicate the specific processing method. Among them, if "edge_handling": "padding" is written, it means that when the sub-block has insufficient edge pixels compared to the standard size (such as 512×512), the remaining area of the sub-block needs to be automatically filled with zero values to unify the sub-block size. For example, if the image width is exactly missing 4 columns of pixels to make up 512 columns, the 4 columns will be filled with 0 pixels (or invalid pixels) to ensure that the downstream image processing algorithm can be analyzed at a fixed size.

[0063] If "edge_handling": "merge_with_block": "X" is written, it means that the large language model module 10 determines that the edge portion is in close contact or adjacent to another already divided sub-block (BlockID is identified as X), and therefore instructs the edge device processing module 20 to directly merge the remaining edge area with the sub-block numbered X into a larger sub-block when cropping and loading the image, so as to avoid generating data blocks that are too small or zero-filled. At this time, "X" is a block_id value defined in the block instruction, which is used to uniquely identify the target sub-block to be merged. For example, when there is only a narrow 4-pixel wide area in the lower right corner of the image, which is in close contact with the adjacent sub-block 7, the instruction will specify "edge_handling": "merge_with_block": "7", indicating that this 4-pixel wide image band should be cropped, loaded and processed together with sub-block 7 in the same sub-block analysis process.

[0064] Through the flexible setting of the "edge_handling" field mentioned above, the system can clearly indicate the two types of edge area processing methods: "padding with zeros" or "merging with adjacent sub-blocks" when the number cannot be divided evenly. After parsing the corresponding field, the edge device processing module 20 will automatically perform the corresponding operation: when it is "padding", the missing pixel segments are filled with zero values; when it is "merge_with_block": "X", the adjacent pixels are first extracted from the image area of the sub-block and the sub-block number X and merged, and then loaded into the memory for image analysis. This ensures that the edge area can also be included in the normal detection process, which not only reduces the interference with computing power and algorithm adaptability, but also enables all potential fault points in the high-resolution image to be covered and analyzed.

[0065] The finally generated block instruction is sent to the edge device processing module 20, which loads and analyzes the image in blocks according to the instruction.

[0066] In another embodiment, the model is fine-tuned using a specialized dataset containing thousands of images of faults such as cracks, partial discharges, and overheating in power equipment, along with their textual descriptions.

[0067] In specific implementation, through repeated reinforcement learning or supervised learning on issues such as "whether to cut blocks" and "how to set high-priority partitions", the model gradually masters the decision-making thinking of taking into account detection accuracy under limited computing power.

[0068] When performing block strategy analysis, the system constructs a Prompt example, such as: the model generates block instructions based on this and can add additional instructions: "The seventh sub-block requires padding" and "The lower right sub-block is set to a high detection priority" and other details.

[0069] In this way, the present application can implement efficient block strategy analysis and output block instructions based on image size and sub-block size.

[0070] Regarding the above-mentioned edge device processing module 20:

[0071] In order to address the problem of insufficient computing power or memory for high-resolution power equipment images on edge devices, the present invention designs an edge device processing module 20 based on the block instructions issued by the large language model module 10 to perform block-by-block analysis on the sub-block size that can be processed each time.

[0072] In a feasible embodiment of the present application, the edge device processing module 20 includes an image processing unit, an image analysis unit and a communication unit to complete the entire process of cropping and loading sub-block data, image processing and result output.

[0073] In a specific implementation, the edge device processing module 20 receives the block instructions generated by the large language model module 10 through a network interface or local shared memory. The block instructions can be in JSON or other structured formats, where key fields may include:

[0074] block_i: used to uniquely identify a sub-block;

[0075] x_start, y_start: indicates the starting row and column position of the sub-block in the original image coordinate system;

[0076] width, height: block size, such as 512×512;

[0077] edge_handling: When the image is not divisible, whether to fill it with zeros or splice it with adjacent sub-blocks;

[0078] detection_level or priority_merge: Indicates the detection level and subsequent fusion priority that should be adopted by this sub-block.

[0079] After parsing the block instruction, the edge device processing module 20 internally generates a set of sub-block queues to be processed. Each item in the queue contains the above fields and possible fault risk prompts (if the large language model determines that the probability of failure in a certain area is higher based on the device portrait).

[0080] The image processing unit in the edge device processing module 20 first reads the entire power equipment image (or, in actual implementation, loads it in segments from a high-resolution image file or stream) and crops the corresponding sub-block image data based on the coordinate data (x_start, y_start, width, height) specified in the instruction. If the edge_handling field indicates that zero padding is required, the system pads the lower right edge or the specified area with zero pixels to achieve a size compatible with the algorithm. If "join with block ID = XX" is specified, the two sub-blocks are loaded and merged at the boundary before being sent to the image analysis unit.

[0081] Through this block-based loading, the edge device only needs to retain image data of a processable size (such as 512×512) in memory at a time, avoiding excessive memory usage caused by loading the entire image.

[0082] The image analysis unit can perform various processing on the loaded sub-block image, mainly including:

[0083] Defect detection: Detect surface cracks, partial discharges, or foreign matter contamination using thresholding-based or deep learning-based (e.g., CNN / Transformer) detection algorithms.

[0084] Texture analysis: If crack risk is detected or the surface material of the equipment is special, texture features can be extracted using Gabor filtering or GLCM method;

[0085] Segmentation: Use models such as GrabCut and U-Net to segment infrared thermal images or visible light images to locate hot spots or possible fault areas;

[0086] Fusion processing: If the device image or block instructions prompt that the infrared band needs to be aligned with the same sub-block of visible light, image registration and multi-channel fusion can be performed at this stage.

[0087] If detection_level in the block instruction indicates that this block is in a high fault risk area, the image analysis unit switches to a higher-precision detection model (such as a deeper CNN) and outputs a more detailed segmentation mask or probability distribution map.

[0088] In an optional embodiment of the present invention, in addition to outputting block segmentation instructions, the large language model module may also output image processing instructions at the same time, which will be explained in detail below.

[0089] After completing the analysis of the sub-block image, the edge device processing module 20 generates a "sub-block analysis result", which includes, for example:

[0090] block_id: corresponds to the block instruction;

[0091] defect_coordinates or defect_mask: If cracks or hot spots are detected, record their specific positions or segmentation contours in the sub-block coordinate system;

[0092] confidence_scores: confidence scores for the detection algorithm output;

[0093] valid_area: If zero padding or splicing is used, it indicates the actual valid pixel range;

[0094] Other algorithm outputs (such as texture feature vectors, temperature distribution data, etc.).

[0095] If cloud / local collaboration is set, the communication unit packages the "sub-block analysis results" and transmits them back to the result merging module 40 or the cloud server for subsequent global image stitching and comprehensive fault judgment.

[0096] In addition, when the sub-block size cannot meet the algorithm requirements (for example, it must be a multiple of 2), the image processing unit can automatically adjust or throw an exception during the instruction parsing stage; in a parallel environment, multiple sub-blocks can be loaded at a time, and multi-threading or multi-GPU can be called for acceleration; if the device portrait shows that hot spot anomalies are of particular concern in specific seasons or working conditions, the edge device processing module 20 can also combine the prompts in the instruction to prioritize the analysis of relevant sub-blocks and push real-time alarms.

[0097] In this way, through the above-mentioned block-based image analysis method, the edge device processing module 20 only needs to meet the memory and computing power requirements within the block size range, and does not need to process the entire high-resolution image at one time, thereby effectively reducing hardware investment and ensuring detection accuracy.

[0098] Regarding the above-mentioned device portrait module 30:

[0099] In order to enable the large language model module 10 to accurately obtain the image size, processable sub-block size and more historical fault information of the power equipment, the present invention stores the relevant information in the equipment portrait module 30.

[0100] In one feasible implementation of the present application, the equipment portrait module 30 can be regarded as a set of "digital equipment archives" or "digital twin" databases, recording historical data, real-time monitoring information and operation and maintenance status from the installation of the equipment to the present.

[0101] The content stored in the device portrait module 30 may include:

[0102] Image metadata: records the image resolution, shooting time, shooting device ID, shooting angle, etc. of each shot;

[0103] Historical fault records: corresponding to the occurrence time, type (such as crack, overheating, dirt), fault location (possibly in the form of coordinates or text description), repair results, etc. of each fault event;

[0104] Processable sub-block size information: reflects the maximum sub-block size (e.g., 512×512) that the edge device can load for each image analysis; it can also be expanded to record specific limitations such as available video memory and bandwidth;

[0105] For example, see Figure 3 , is a schematic diagram of a maximum sub-block size in overall image analysis provided in an embodiment of the present application, wherein the maximum sub-block size is 512×512 and the overall image size is 1536×1536.

[0106] Multimodal sensor data: If there is an infrared heat map or vibration signal, its key features or file links can be stored;

[0107] Basic device attributes: such as device model, service life, manufacturer, installation location, and operating conditions such as ambient temperature and humidity in certain scenarios.

[0108] In terms of technical implementation, relational databases (MySQL / PostgreSQL) or NoSQL document libraries (MongoDB, ElasticSearch) can be used to manage these fields, or key information can be saved in the form of JSON files on a cloud server.

[0109] In addition, the device portrait module 30 is not a static configuration library, but can be continuously updated based on the results of subsequent fault detection or operation and maintenance inspections. For example, when the edge device processing module 20 detects a new crack or hot spot, it will package the detection time, coordinate area and severity and write it back to the device portrait. This process can be collected uniformly by the result merging module 40, and then written to the database through the API, so that the "device portrait" maintains the latest portrayal of the current power equipment status. At the same time, the large language model module 10 can query this module again before the next inference to obtain the latest historical data and sub-block processing limits, thereby forming a closed loop.

[0110] When the large language model module 10 needs to analyze the segmentation strategy or decide on the image processing method, it first sends a query request based on the device ID to the device profiling module 30. The latter queries the record corresponding to the device in the table (or document) and returns the required fields.

[0111] The large language model module 10 can parse the specified fields, make a block strategy for the current image, and determine whether there are specific areas that require high-precision detection based on special fields such as historical maintenance records.

[0112] In actual deployment, the device profiling module 30 can be combined with a cloud server to provide data query and update operations through a RESTful or RPC interface. If the system is large, a cache layer can be added to speed up frequent queries.

[0113] For power equipment, a substation or line often contains multiple devices. Each device retains independent data records in the device profile module, distinguished by device_id or other unique identifiers. This can meet the management needs of multiple devices, multiple images, and high-concurrency scenarios.

[0114] Through this module's unified storage and dynamic maintenance, the large language model module 10 no longer needs to manually or temporarily obtain device information, greatly reducing the configuration burden. Furthermore, as the number of inspections increases, device profiles continuously enrich fault mode and repair information, providing data support for the personalized generation of segmentation and processing strategies, enabling the system to continuously improve fault monitoring accuracy and efficiency over the long term.

[0115] For the result merging module 40:

[0116] To integrate the analysis results of each sub-block into a fault detection or feature extraction output consistent with the original high-resolution image coordinate system, the present invention provides a result merging module 40. This module receives multiple "sub-block analysis results" from the edge device processing module 20, references the address mapping information in the block instructions generated by the large language model module 10, completes the splicing and fusion of the complete image, and finally writes the overall detection results to the device profiling module 30.

[0117] After the edge device processing module 20 completes the detection of each sub-block, it will generate a data structure (such as JSON) containing:

[0118] block_id: identifies the sub-block;

[0119] local_defect_mask or local_features: the detection results or features recorded in the local coordinate system of this sub-block (0width-1,0height-1);

[0120] confidence_scores: the confidence of each pixel or detected target;

[0121] valid_area: If the sub-block is padded with zeros or spliced, it indicates the actual valid pixel area.

[0122] At the same time, the block instruction generated by the large language model module 10 records the global starting coordinates (x_start, y_start) corresponding to the block_id, as well as width, height, etc. Based on this, the result merging module 40 can map the local coordinates back to the original image coordinates.

[0123] When adjacent sub-blocks are continuous in the horizontal or vertical direction and do not overlap, the result merging module 40 can directly place the local_defect_mask or local_features at the corresponding global coordinate position, such as placing sub-block ID=1 at (0511,0511) and sub-block ID=2 at (5121023,0511), thereby splicing out the entire high-resolution detection result image.

[0124] Among them, if the coordinate range of sub-block ID=1 is x=0 to 511, y=0 to 511 (denoted as (0511, 0511)), then the range of sub-block ID=2 can be immediately followed by x=512 to 1023, y=0 to 511 (denoted as (5121023, 0511)), thereby stitching together a partial image with a full width of 1024 pixels in the horizontal direction.

[0125] It is understandable that the above marks are only examples, and character segmentation can actually be performed, which will not be described in detail here.

[0126] If a sub-block has a "zero padding" or "edge_handling=padding" field, only the valid_area part of it will be spliced; the zero-padding area can be automatically ignored or overwritten.

[0127] If the model instruction specifies that there is an N-pixel-wide overlap between sub-blocks (or the edges of adjacent sub-blocks are partially overlapped to improve detection accuracy), the result merging module 40 needs to fuse the overlapped area. The following strategies can be used:

[0128] Maximum fusion: Take the maximum value of the detection value of the same pixel in two sub-blocks;

[0129] Mean fusion: average the same pixel;

[0130] Confidence weighting: If the detection confidence of a sub-block is higher, its result will be used first;

[0131] Priority strategy: If the priority_merge field of a sub-block is high, the result of another sub-block is overwritten.

[0132] When overlapping pixels are detected, the result merging module 40 will consult the confidence or priority fields of the sub-block analysis results and then generate the final pixel value (or final defect mark) accordingly. This process can be flexibly configured according to operation and maintenance needs.

[0133] After coordinate mapping and fusion, the result merging module 40 ultimately produces a defect detection / segmentation map or other feature map with the same resolution as the original image (e.g., 4096×4096). If the detection algorithm outputs a segmentation mask, multiple sub-block masks can be concatenated into the same two-dimensional matrix. If the output is a coordinate list, it can be converted into a global coordinate list and marked in the device image.

[0134] After the merging is complete, the result merging module 40 writes the entire image detection results (or the extracted fault coordinates and fault types) back to the device profiling module 30. When image analysis or manual maintenance is performed again, this global result can serve as a historical reference to help the large language model module 10 fine-tune the next round of segmentation strategy or conduct risk assessment.

[0135] Thus, through the result merging module 40, the system can accurately map the detection information scattered across each sub-block back to the original image coordinates, achieving overall fault visualization or feature representation for the high-resolution image. Compared to the global splicing difficulties or manual duplication that single-block algorithms often cause, the fusion mechanism of the present invention takes into account sub-block characteristics, overlap processing, and priority configuration, ensuring the accuracy and consistency of the detection results for the entire image.

[0136] In another optional embodiment of the present application, when performing block strategy analysis, the large language model module 10 focuses on edge portions that cannot be divided evenly and decides to adopt "zero padding" or "splicing with adjacent blocks".

[0137] For example, when the image size provided by the device portrait module 30 is , and can handle sub-block size , the model can calculate 8 complete sub-blocks with 4 remaining edge rows and columns. At this point, the large language model module 10 will generate the following in the block instruction:

[0138] If zero padding is required, specify edge_handling: "padding" in the instruction and indicate in the valid_area field that there are actually only 4 rows of available pixels;

[0139] If it is to be spliced with an adjacent block, merge_with_block=BlockID(n) is indicated in the instruction, instructing the edge device processing module 20 to merge the remainder area with the sub-block corresponding to BlockID(n) before performing analysis.

[0140] In this way, the block instruction accurately describes the processing scheme of the edge area, which facilitates the edge device pre-processing unit 20 to perform zero-padding or merging operations when loading the image block without manual intervention.

[0141] After the device profiling module 30 records the sub-block size that the edge device can process, image resolution, and "historical fault hotspot distribution," the present invention can determine whether to use overlapping blocks in the edge area based on whether there are high-fault areas. Assuming that historical records show a high incidence of cracks or overheating at the lower edge of the image (row coordinates > 3500), the large language model module 10 can specify "overlap_size:20" pixels in the block instruction, allowing adjacent sub-blocks to have a 20-pixel vertical overlap, thereby collecting more feature information in this high-risk area.

[0142] For example, the model can be written in the instruction:

[0143] {"block_id":8, "x_start":0, "y_start":3500,"width":512, "height":512,"overlap_with_block":7,"overlap_size":20}

[0144] "overlap_with_block":7 indicates vertical overlap with sub-block number 7, corresponding to the block_id of another sub-block defined in the block splitting instruction. "overlap_size":20 uniformly indicates the number of pixels in the overlapping area, meaning that 20 rows of pixels are repeatedly scanned or inspected between the two blocks. This approach allows the result merging module 40 to more accurately merge inspection results during subsequent fusion, reducing the chance of overlooked cracks or hot spots caused by boundary segmentation.

[0145] In another optional embodiment, when the large language model module 10 generates a block instruction, if the image size is exactly divisible by the processable sub-block size, , can be directly divided into Each sub-block has no extra edge_handling flag. , the model will add a "valid_area" field to the edge sub-block in the output instruction to describe the available pixel range.

[0146] Here, K represents the side length of the entire image (e.g., the number of pixels of width or height); P represents the side length of the processable sub-block; K mod P = 0, indicating that the remainder after the "modulus" operation is 0; K mod P ≠ 0, indicating that there is a remainder when the image side length K is divided by P.

[0147] For example:

[0148] {"block_id":10,"x_start":3584, "y_start":3584,"width":512, "height":512,"valid_area": {"width":420, "height":420},"edge_handling": "padding"}

[0149] This allows subsequent edge devices to analyze only the 420×420 valid pixel area; the remaining area can be treated as a padding area or ignored. This "valid area information" ensures that the system can seamlessly process the last few rows and columns of pixels in a high-resolution image, regardless of resolution.

[0150] As an optional implementation, the edge device processing module 20 includes:

[0151] An image processing unit, configured to crop or splice image blocks of a processable sub-block size according to the block division instruction;

[0152] An image analysis unit, configured to perform defect detection, texture analysis, and segmentation operations on each image block;

[0153] The communication unit is used to upload the output data of the image analysis unit to the result merging module 40.

[0154] In an optional embodiment of the present invention, the edge device processing module 20 may be composed of the following three units or coupled to implement them, so as to better complete the processing and analysis of the sub-block image:

[0155] The image processing unit is mainly responsible for receiving the block instructions issued by the large language model module 10, and performing cropping or splicing operations on the original high-resolution image according to the fields such as x_start, y_start, width, height, edge_handling, etc. in the instructions.

[0156] Cropping scenario: When normal sub-blocks are in a non-overlapping or neatly divided area, the image processing unit can directly extract the corresponding pixel block (e.g., 512×512) from the source image file or stream and cache it into memory;

[0157] Stitching scenario: If edge_handling="merge_with_block":X is specified in the block instruction, the image processing unit will perform edge merging with the image block corresponding to BlockID=X after loading the current edge area, and then cache the merged data to the image analysis unit;

[0158] Zero-padding scenario: If edge_handling="padding" and valid_area < width / height, the image processing unit fills the pixels outside the valid area with zero values to ensure that the downstream analysis algorithm can receive input of uniform size;

[0159] Through this block-based loading, the image processing unit only needs to retain data of a processable size in memory once, avoiding large images from occupying a large amount of resources and ensuring that the subsequent image analysis unit can correctly identify the zero-filled or spliced areas.

[0160] After acquiring the sub-block images cropped / joined by the image processing unit, the image analysis unit performs operations such as defect detection, texture analysis, and segmentation:

[0161] Defect detection: Thresholding-based or deep learning (CNN / Transformer) algorithms can be used to identify surface cracks, contamination, or partial discharges.

[0162] Texture analysis: If a sub-block is marked as a high-risk area or requires fine-grained detection, texture extraction such as Gabor filtering and GLCM can be performed;

[0163] Segmentation: GrabCut, U-Net, or other segmentation methods can be used to detect hot spot areas in infrared images, or to segment device surfaces in visible light images.

[0164] Before execution, fields such as detection_level and priority_merge are checked. If their values indicate a high risk of failure, the advanced detection model is enabled and more detailed probability distribution / mask information is output.

[0165] After the analysis is completed, the image analysis unit generates a "sub-block analysis result", which includes:

[0166] { "block_id": ...,"defect_mask": ..., "confidence_scores": ...,"valid_area": ...} and other fields.

[0167] After the image analysis unit completes sub-block level detection or segmentation, the communication unit uploads the "sub-block analysis results" to the result merging module 40, or uploads them to the remote server for further processing in the cloud-edge collaboration scenario.

[0168] Transmission can be accomplished through local shared memory queues, RPC interfaces, or HTTP requests. If the system is configured for parallel or distributed processing, the communication unit packages the analysis results of multiple sub-blocks by block_id, and can include timestamps and algorithm version information for subsequent merging or auditing. In the event of network anomalies or hardware failures, the communication unit can also cache the results and wait for recovery before sending them, ensuring that locally processed data is not lost.

[0169] In this way, through the division of labor and cooperation of the above three units, the edge device processing module 20 can automatically adapt to the size of different sub-blocks, edge processing methods, and fault risk levels, which not only reduces the demand for high-end hardware, but also can flexibly switch at the image analysis algorithm level to meet the diversified fault detection needs of power equipment.

[0170] As an optional implementation, the result merging module 40 further performs maximum or average processing on the output results of adjacent sub-blocks with overlapping areas according to the fusion method specified in the block instruction. When multiple blocks have different detection results at the same target pixel coordinates, a preset weighted merging strategy is used to obtain the final result.

[0171] In an optional embodiment of the present invention, the result merging module 40 can read a more refined "merging mode" field in the block instruction to guide the merging of detection results in overlapping areas. When adjacent sub-blocks have an N-pixel-wide overlap, or the same target pixel coordinates are detected in multiple sub-blocks, the result merging module 40 implements the following strategy:

[0172] Maximum value merging: If the fusion method field fusion_method="max", the maximum value of the detection value at the same coordinate is taken. This is suitable for scenarios where it is better to "overreport" than "underreport" for high-risk faults (such as cracks and overheating).

[0173] Mean merging: If the fusion method field fusion_method="mean", the detection values ​​at the same coordinate are averaged; this helps smooth detection noise.

[0174] Weighted merging: When performing weighted merging, if the large language model module 10 specifies "fusion_method = weighted" in the fusion method field, the result merging module 40 will calculate the final pixel or defect mark value using the following formula when multiple detection values appear at the same pixel coordinate :

[0175]

[0176] Where n represents the total number of sub-blocks involved in the detection output within the overlapping area of the pixel coordinates; represents the detection value or defect probability of the i-th sub-block at the pixel coordinate; Represents the weight of the i-th sub-block, which can be determined by the confidence level, historical failure risk or priority strategy of the sub-block; the numerator Represents the sum of the detection values of each sub-block multiplied by its weight, and the denominator This is used to normalize the weighted results to ensure that the final output value is within a reasonable range. Using this weighted merging formula, the result merging module 40 can generate a final detection value that is more consistent with the actual situation when there are discrepancies in the detection results of multiple sub-blocks, based on the reliability and risk factors of different blocks, thereby reducing missed detections or misjudgments caused by edge segmentation.

[0177] When multiple sub-blocks produce distinct detection results at the same target pixel coordinate, if the "priority_merge" field indicates that a sub-block has a higher priority, its output is directly used, ignoring the other sub-blocks. This configuration allows this implementation to more flexibly balance the risk of false detection and missed detection during overlapping merging.

[0178] In another optional embodiment of the present invention, the large language model module 10 can be deployed on a cloud server or a high-performance computing platform with sufficient GPU / CPU resources for model reasoning and block decision-making; while the edge device processing module 20 is deployed in an industrial PC, embedded system or edge hardware at the power site, and collaborates with the cloud module through network communication.

[0179] After receiving the image resolution, processable sub-block size, and fault hotspot information provided by the device portrait module 30, the cloud-side large language model module 10 generates corresponding block instructions and sends them to the edge device processing module 20 via HTTP / HTTPS, RPC, or message queues. The latter loads and detects sub-blocks of the high-resolution image and packages the analysis results and uploads them back to the cloud. When the edge device network is unstable, the sub-block analysis results can be cached locally and then sent to the cloud after the link is restored. This can fully utilize the powerful computing power of the cloud while avoiding the need for overly expensive hardware on site.

[0180] As an optional implementation, the large language model module 10 is deployed on a cloud server or a high-performance computing platform, and the edge device processing module 20 is deployed in the edge hardware of the power site. The two communicate through the network to issue instructions and transmit sub-block analysis results.

[0181] In another optional embodiment of the present application, the large language model module 10 dynamically reads incremental data such as industry specifications and historical failure cases provided by the device profiling module 30 when generating block instructions. Once a change in the "processable sub-block size" or "image resolution" is detected (for example, the camera resolution is upgraded on-site, or the edge device hardware is replaced to support larger block processing), the model automatically recalculates the block strategy. For high-risk areas with frequent failures, additional detection efforts are increased based on new specifications or updated historical missed detection statistics.

[0182] Specifically, each time the system collects a new image, the cloud or local will push its resolution and fault risk record to the large language model module 10; if it is found that the edge_device_block_size is upgraded from 512×512 to 1024×1024, the model will reduce the number of blocks accordingly and retain moderate overlap for high-risk areas to reduce the impact of edge cutting on crack detection.

[0183] Through this automated adjustment mechanism, this implementation can ensure that the application can still perform high-efficiency and high-accuracy fault detection capabilities after equipment updates, changes in sensor resolution, or upgrades in industry requirements. There is no need for frequent manual modifications to block rules or processing procedures, and it has better scalability and self-learning potential.

[0184] As an optional implementation, the large language model module 10 automatically adjusts the image segmentation strategy based on industry standards and historical failure cases provided by the device portrait module 30, and dynamically generates new segmentation instructions when the processable sub-block size or image size changes.

[0185] In practice, the large language model module 10 automatically updates image segmentation strategies based on industry standards and historical fault case information recorded in the equipment profiling module 30. When conditions such as the processable sub-block size or image resolution change significantly, the large language model module 10 dynamically generates new segmentation instructions, ensuring high-precision detection of power equipment images despite hardware or environmental changes.

[0186] Regarding the triggering conditions of this embodiment:

[0187] For example, it can be a hardware upgrade / downgrade. When the device profiling module 30 records that the processable sub-block size is upgraded from 512×512 to 1024×1024, or the edge device is downgraded to 256×256 for some reason, the large language model module 10 detects the update and recalculates the number of blocks and the overlap strategy of each sub-block;

[0188] For example, when industry standards or failure cases change, for example, if a new entry is added to the database, such as "Insulator crack detection requires ≥95% coverage" or "Humid environments in a certain region lead to more overheating failures," the model can incorporate this standard / case into the prompt, increase overlapping areas, or prioritize inspections in high-risk areas.

[0189] For example, when the image size changes from 4096×4096 to 8192×8192, the model must again determine whether blocks need to be merged, zero-padded, or expanded.

[0190] Regarding further automated adjustment processes, the latest portrait information can be obtained. Whenever the edge device or the cloud finds an update to the sub-block size, image resolution, or industry specifications, the device portrait module 30 will record the corresponding field (e.g., edge_device_block_size = 1024×1024);

[0191] For model reasoning, the large language model module 10 adds new hardware parameters and historical fault cases to the prompt and then performs internal reasoning. For example, if the historical missed detection rate is high for a certain fault type, the block refinement of that area will be automatically increased;

[0192] For outputting new block instructions, in the generated JSON or XML instructions, the model updates the sub-block width, height, or "edge_handling" strategy and sets a "high detection level" for specific fault areas. If industry regulations require greater coverage, "overlap_size=XX" can be written in the instructions to increase detection redundancy.

[0193] For edge device execution, the edge device processing module 20 crops and loads the image block according to the new instruction, and the communication unit transmits the sub-block result back to the result merging module 40. The entire process does not require manual configuration of the block threshold or rule.

[0194] Through this dynamic adjustment mechanism, this optional implementation allows the large language model module 10 to provide the optimal or appropriate blocking strategy for high-resolution power equipment images in different stages or environmental changes, thereby maintaining the overall efficient fault detection capability of the system.

[0195] As an optional embodiment, the large language model module 10 is further configured to, based on the personalized information related to the historical fault records, operating environment, and sensor data of the power equipment obtained from the device profiling module 30, further generate an image processing method instruction for each sub-block after determining the block processing method of the power equipment image, and send the image processing method instruction to the edge device processing module 20, wherein the image processing method instruction includes:

[0196] Target algorithm type, used to indicate the image processing, defect detection, and texture analysis algorithms to be used for a sub-block;

[0197] Parameter configuration, used to indicate whether additional filtering, enhancement, or segmentation thresholds are required under specific spectral or temperature distributions;

[0198] Priority information, used to instruct the edge device processing module to execute a preset second detection process for sub-blocks whose failure risk is greater than or equal to a preset risk value, and to use a preset first detection process for sub-blocks whose failure risk is less than the preset risk value;

[0199] A multimodal fusion strategy is used to respond to the mapping relationship between the visible light image and the infrared image recorded in the device portrait module 30, and the image processing method instruction instructs the edge device processing module 20 to perform cross-spectral or multi-channel data fusion on the corresponding sub-blocks.

[0200] The first detection process may be a conventional image-based detection process corresponding to the power equipment recorded in the equipment portrait module 30. The second detection process may be a special image-based detection process corresponding to a special problem recorded in the equipment portrait module 30 of the target power equipment.

[0201] Among them, the large language model module 10 is not only responsible for determining the block strategy based on the power equipment image size and processable sub-block size obtained from the equipment portrait module 30, but can also further combine the power equipment historical fault records, operating environment and sensor data (hereinafter collectively referred to as "personalized information"). After completing the block processing of the image, it generates a more refined image processing method instruction for each sub-block, and sends the instruction to the edge device processing module 20, thereby realizing customized image processing, detection and analysis processes for sub-blocks with different risk levels or different spectral requirements.

[0202] Power equipment often faces a variety of fault types: surface cracks, partial discharge (PD), hot spot overheating, corrosion, or contamination accumulation. A single image processing algorithm often struggles to account for all scenarios, especially when each sub-block corresponds to a different equipment location, environmental conditions, or historical failure probability.

[0203] To this end, this optional implementation method proposes: after the block division scheme is determined, the large language model module 10 tailors image processing instructions for each sub-block based on the historical fault records, operating environment (such as temperature, humidity, salt spray), and multimodal sensor data (such as the correspondence between infrared / visible light) recorded in the equipment portrait, and sends them to the edge device processing module 20 for local execution.

[0204] Personalized information may include:

[0205] Historical fault records: refers to past fault cases, maintenance records, fault frequency and time distribution stored in the equipment portrait;

[0206] Operating environment: refers to the temperature, humidity, strong salt spray or dust in the environment where the equipment is located;

[0207] Sensor data: If the device supports the coexistence of infrared thermal images and visible light images, the device image may include their registration (alignment) information or spectral mapping relationship.

[0208] This information enables the large language model module 10 to accurately assess the failure risk or algorithm requirements of a sub-block, thereby specifying a more suitable algorithm category, parameters and fusion strategy in the instructions.

[0209] In specific implementations, the large language model module 10 first determines how to divide the entire high-resolution image into several sub-blocks based on basic parameters such as image size and processable sub-block size. If there are parts with edges that cannot be divided evenly, they will also be marked with zero padding or splicing methods. After determining the block division, the model again queries the equipment profiling module 30 for fields such as "historical fault records of power equipment, operating environment, and sensor data." Based on the query results, the model will identify, for example:

[0210] The area corresponding to a sub-block has frequently experienced cracks in the past (high failure risk); the device's current temperature exceeds the standard or is in a hot and humid environment; the image has both infrared and visible light channels available for fusion.

[0211] Furthermore, based on the sub-block's fault risk level, spectral requirements, and available hardware computing power, the model outputs an image processing method instruction for each sub-block, indicating which image processing or detection algorithm should be used for the sub-block.

[0212] For example, when the spectrum or temperature distribution is special, it is necessary to configure the filter kernel size, enhancement parameters, or segmentation threshold in the instruction. Through the fields in the instruction, the edge device can automatically load the corresponding model weights or preprocessing parameters in the image analysis unit.

[0213] When the device image records the mapping relationship between visible light and infrared, and the current sub-block is in a position that requires cross-spectral alignment, the command can be marked:

[0214] "fusion": "ir_visible",

[0215] "fusion_params": { "offset_x": 5, "offset_y": 2}

[0216] This indicates that the edge device should perform detection after the image processing unit aligns the visible light and infrared data to improve the accuracy of defect recognition. The large language model module sends the above instructions and the "block coordinate instructions" to the edge device processing module 20 together or in two batches.

[0217] After the edge device parses the instruction, it calls the specified algorithm process in the image analysis unit according to the field content; it invests more computing power in sub-blocks with high priority or requiring infrared fusion.

[0218] After the processing is completed, the edge device communication unit returns the detection results of each sub-block to the result merging module 40 or the cloud;

[0219] The large language model module 10 and the device portrait module 30 can update historical failures or algorithm effects again for subsequent further fine-tuning and iteration.

[0220] In an optional embodiment of the present invention, to determine whether a sub-block of a power equipment image has a high risk of failure, the large language model module 10 integrates information from the equipment profiling module 30 and real-time monitoring data to form a "failure risk value" (also known as a "failure risk score" or "risk_score"). This score can be determined by the following factors:

[0221] First, the historical fault records. If the device portrait records that a certain area of the device (such as the x, y coordinates or physical location) has been detected with cracks or overheating multiple times in the past six months, the large language model module 10 will assign a higher basic risk score to the area in the prompt or internal logic.

[0222] For example, if there are more than 5 historical cracks, the risk_score can be initially set to 0.7 (70%); if there has never been a crack, it can be only 0.2.

[0223] Secondly, the operating environment or real-time sensor data. If the equipment is in a high-temperature, high-humidity, or salt-fog corrosive environment, the probability of failure (such as insulation aging and corrosion) tends to increase.

[0224] If real-time sensing (temperature, vibration, etc.) shows that the current operating conditions are close to the rated limits of the equipment, the failure rate may also increase.

[0225] The large language model module 10 can input something like: "Current temperature = 45°C, humidity = 80%, the device is under high load" in the prompt, thereby increasing the risk_score.

[0226] The equipment portrait can also record certain industry standards. For example, oil-immersed transformers are prone to cracking in high-altitude and low-temperature environments.

[0227] Based on these specifications, the large language model increases the risk score of the corresponding area by one level, such as from 0.5 to 0.7.

[0228] In another optional embodiment of the present invention, the large language model module 10 can also automatically select between different image detection processes recorded in the device profiling module 30 and generate image processing instructions for each sub-block based on personalized information such as the power equipment's historical fault records, operating environment, and sensor data. In this embodiment, the device profiling module 30 pre-stores at least two sets of detection processes: the first detection process can be a conventional image-based detection process for the power equipment, used in common or low-risk scenarios; the second detection process is a high-precision detection process designed specifically for specific fault issues, used for more in-depth analysis when the fault risk of a certain area is higher.

[0229] In a specific implementation, after the large language model module 10 preliminarily determines the block segmentation strategy based on the image size and the processable sub-block size, it further evaluates the fault risk value corresponding to the power equipment. This risk value is derived by the large language model module 10 based on a comprehensive analysis of historical failure statistics recorded in the equipment image (e.g., multiple cracks or hot spots in the area over the past year), the operating environment (e.g., high temperature, high humidity, salt spray corrosion, etc.), and sensor data (e.g., infrared temperature distribution). If the assessment results indicate that the risk value of a sub-block is greater than or equal to a pre-set threshold, the large language model module 10 designates the sub-block as the "second detection process" in the image processing method instruction; otherwise, the "first detection process" is used by default.

[0230] In the device profiling module 30, the first detection process typically documents the image detection methods typically performed on the power equipment. These methods may include simpler algorithms such as threshold-based segmentation, Canny edge detection, or lightweight convolutional neural networks, which are fast and resource-efficient, making them suitable for low-risk scenarios. The second detection process, stored in the device profiling module 30 as a specific identifier for a specific fault, often involves a deep learning multi-layer model or multi-channel fusion algorithm. These algorithms require higher computing power and execution time, but can provide more accurate fault location in high-risk scenarios.

[0231] For example, the relevant records in the device portrait may be such as "process_id='std_detect_v1'" indicating the first detection process, "process_id='special_detect_v2'" indicating the second detection process, and the respective algorithm composition and applicable scope are explained.

[0232] When the large language model module 10 detects that the risk score of a certain area exceeds a threshold (for example, 0.7), it will specify "target algorithm type = 'special_detect_v2'" and indicate higher priority and stricter processing parameters when generating image processing instructions; if the assessment result is only low risk, "target algorithm type = 'std_detect_v1'" will be written into the instruction. At the same time, if the equipment portrait module 30 also records the visible light and infrared image registration information of the power equipment, the large language model module 10 will also attach a "multimodal fusion strategy" to the instruction, requiring the edge device processing module 20 to perform cross-spectral or multi-channel data alignment when analyzing the sub-block. For certain scenarios with specific spectral or temperature distributions, details such as filtering method, threshold size or model version can be further marked in the "parameter configuration".

[0233] When the edge device processing module 20 receives the "image processing method instruction" for each sub-block, it reads the algorithm type, detection process priority, and other configuration information in the instruction field and loads the corresponding script or model based on the specific implementation of the "first detection process" or "second detection process" recorded in the device profile. If the instruction indicates a high priority, it will perform advanced operations such as depth detection or cross-spectral fusion; if it is a normal case, it will use a lightweight process to save computing power and time.

[0234] This model significantly improves the balance between detection accuracy and efficiency: for high-risk sub-blocks, higher-precision algorithms can be applied to avoid missed detections; for low-risk sub-blocks, rapid detection methods can be used to increase overall processing speed, thereby enabling refined and personalized analysis of power equipment images. Because these detection processes and their applicable scopes are pre-documented in the equipment profiling module 30, the automated reasoning process of the large language model module 10 fully utilizes equipment profiling information without additional human intervention, maintaining dynamic adaptability to system environments and fault propensities during real-time or periodic analysis.

[0235] Based on the same inventive concept, an embodiment of the present disclosure also provides an electric power equipment monitoring method corresponding to the electric power equipment monitoring system. Since the principle of solving the problem by the method in the embodiment of the present disclosure is similar to that of the above-mentioned electric power equipment monitoring system in the embodiment of the present disclosure, the implementation of the method can refer to the implementation of the method, and the repeated parts will not be repeated.

[0236] Reference Figure 2 FIG. 1 is a flow chart of a method for monitoring electric power equipment according to an embodiment of the present application, comprising steps S101 to S104, wherein:

[0237] S101: Using a large language model, based on the image size of the power equipment image obtained from the device portrait and the sub-block size that can be processed by the edge device, perform a block strategy analysis on the target resolution power equipment image to obtain a block instruction; wherein the large language model includes: a large model trained and / or fine-tuned based on power equipment scenario data;

[0238] S102: Using the edge device, based on the block instruction, the power equipment image is processed in blocks, and each time only the image blocks in the area limited by the processable sub-block size are analyzed to obtain sub-block analysis results;

[0239] S103: Using the device portrait, storing image metadata, historical fault records, and processable sub-block size information of the power equipment, and providing the image size and processable sub-block size to the large language model;

[0240] S104: Based on the sub-block analysis results and the address mapping corresponding to the sub-blocks in the block instruction, performing result splicing and fusion to obtain a fault identification and / or feature extraction result for the power equipment image with a complete target resolution;

[0241] Among them, the large language model determines the way in which the edge device performs block processing on the power equipment image based on the image size and processable sub-block size provided by the equipment portrait module 30, and maps the sub-block analysis results after block processing back to the original image coordinates, and finally records the overall detection results in the equipment portrait module 30.

[0242] Those skilled in the art will understand that in the above-described methods of specific embodiments, the order in which the steps are presented does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible inherent logic. It should be understood that determining B based on A does not mean determining B based solely on A; B can also be determined based on A and / or other information.

[0243] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0244] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that the specific features, structures, materials, or characteristics described in conjunction with that embodiment or example are included in at least one embodiment or example of the present invention. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

Claims

1. Power equipment monitoring system, characterized in that, include: A large language model module performs a block strategy analysis on the target-resolution power equipment image based on the image size of the power equipment image obtained from the device portrait and the sub-block size that can be processed by the edge device to obtain a block instruction; wherein the large language model module includes: a large model trained and / or fine-tuned based on the power equipment scenario data; An edge device processing module, based on the block instruction, performs block processing on the power equipment image, and analyzes only the image blocks in the area limited by the processable sub-block size each time to obtain a sub-block analysis result; The device portrait module is used to store image metadata, historical fault records, and information about the sub-block size that can be processed by the edge device of the power equipment, and provide the image size and the sub-block size that can be processed by the edge device to the large language model module; A result merging module, which performs result splicing and fusion based on the sub-block analysis results and the address mapping corresponding to the sub-blocks in the block instruction, to obtain fault identification and / or feature extraction results for the power equipment image with full target resolution; Among them, the large language model module determines the way in which the edge device processing module divides the power equipment image into blocks based on the image size provided by the device portrait module and the sub-block size that can be processed by the edge device, and the result merging module maps the sub-block analysis results after division back to the original image coordinates, and finally records the overall detection results in the device portrait module.

2. The power equipment monitoring system according to claim 1, characterized in that: The large language model module calculates the number of blocks based on the image size and the processable sub-block size, and pads or splices the edge parts that cannot be divided evenly, and generates corresponding block instructions.

3. The power equipment monitoring system according to claim 1, characterized in that: The device portrait module is used to store at least one of the sub-block size that can be processed by the edge device, the historical fault hot zone distribution, the image resolution and the multimodal sensor data; The large language model module determines whether to adopt overlapping blocks for edge areas based on the processable sub-block size and the historical fault hot zone distribution.

4. The power equipment monitoring system according to claim 2, characterized in that: The large language model module generates the block instruction including: In response to the image size being divided by an integer of a processable sub-block size, directly dividing the entire power equipment image into a plurality of image blocks of the same size; In response to the image size not being divisible by an integer of the processable sub-block size, the edge portion that cannot be divided evenly is padded with zeros or concatenated with an adjacent block, and valid area information of the edge sub-block is recorded in the block instruction.

5. The power equipment monitoring system according to claim 1, characterized in that: The edge device processing module includes: An image processing unit, configured to crop or splice image blocks of a processable sub-block size according to the block division instruction; An image analysis unit, configured to perform defect detection, texture analysis, and segmentation operations on each image block; The communication unit is used to upload the output data of the image analysis unit to the result merging module.

6. The power equipment monitoring system according to claim 1, characterized in that: The result merging module takes the maximum or average value of the output results of adjacent sub-blocks with overlapping areas according to the fusion method specified in the block instruction. When multiple sub-blocks have different detection results at the same target pixel coordinate, a preset weighted merging strategy is used to obtain the final result.

7. The power equipment monitoring system according to claim 1, characterized in that: The large language model module is deployed on a cloud server or computing platform, and the edge device processing module is deployed in the edge hardware at the power site. The two communicate through the network to issue instructions and transmit sub-block analysis results.

8. The power equipment monitoring system according to any one of claims 1 to 7, characterized in that: The large language model module automatically adjusts the image segmentation strategy based on industry standards and historical failure cases provided by the device portrait module, and dynamically generates new segmentation instructions when the processable sub-block size or image size changes.

9. The power equipment monitoring system according to claim 1, characterized in that: The large language model module is further configured to, based on personalized information related to historical fault records, operating environment, and sensor data of the power equipment obtained from the device profiling module, generate an image processing method instruction for each sub-block after determining a block processing method for the power equipment image, and send the image processing method instruction to the edge device processing module, wherein the image processing method instruction includes: Target algorithm type, used to indicate the image processing, defect detection, and texture analysis algorithms used for each sub-block; Parameter configuration, used to indicate whether additional filtering, enhancement or segmentation threshold is performed under the target spectrum or temperature distribution; Priority information, used to instruct the edge device processing module to execute a preset second detection process for sub-blocks whose failure risk is greater than or equal to a preset risk value, and to use a preset first detection process for sub-blocks whose failure risk is less than the preset risk value; A multimodal fusion strategy is used to respond to the mapping relationship between the visible light image and the infrared image recorded in the device portrait module, and the image processing method instruction instructs the edge device processing module to perform cross-spectral or multi-channel data fusion on the corresponding sub-blocks.

10. A method for monitoring electric power equipment, implemented based on the electric power equipment monitoring system according to any one of claims 1 to 9, characterized in that: include: Using a large language model, based on the image size of the power equipment image obtained from the device portrait and the sub-block size that can be processed by the edge device, a block strategy analysis is performed on the target resolution power equipment image to obtain a block instruction; wherein the large language model includes: a large model trained and / or fine-tuned based on the power equipment scenario data; Using the edge device, based on the block instruction, the power equipment image is processed in blocks, and each time only the image blocks in the area limited by the processable sub-block size are analyzed to obtain sub-block analysis results; Using device portraits, storing image metadata, historical fault records, and information about sub-block sizes that can be processed by edge devices for power equipment, and providing the image size and sub-block sizes that can be processed by edge devices to the large language model; Based on the sub-block analysis results and the address mapping corresponding to the sub-blocks in the block instruction, performing result splicing and fusion to obtain fault identification and / or feature extraction results for the power equipment image with full target resolution; Among them, the large language model determines the way in which the edge device divides the image of the power equipment into blocks based on the image size provided by the device portrait and the sub-block size that can be processed by the edge device, maps the sub-block analysis results after division back to the original image coordinates, and records the overall detection results.

Images