Training and speculation integrated power transformation inspection visual analysis system

Abstract

The invention relates to the technical field of intelligent detection of power equipment, and provides a training and inference integrated power transformation inspection visual analysis system, which comprises a basic support module for realizing dynamic distribution of training, testing and inference tasks through a heterogeneous calculation framework; the data management module is used for storing and managing various types of training data sets; the data labeling module is used for labeling electrical equipment defects in pictures or video data in the training data set; the model training module is used for training each training reference model based on selected training data after selecting different training reference models and configuring related training parameters to obtain a plurality of training models; the model test module is used for selecting a test data set to test each training model and selecting a proper training model according to a test result; and the online reasoning module is used for detecting the video stream data through the selected training model and outputting a detection result. Deep integration of the whole process of training, reasoning and testing is provided, multi-platform switching is avoided, and the data migration cost is saved.

Description

Technical Field

[0001] This invention relates to the field of intelligent detection technology for power equipment, and in particular to a substation inspection visual analysis system that integrates training and inference. Background Technology

[0002] Traditional substation inspection relies on manual visual inspection or fixed sensors, resulting in low efficiency, high missed detection rates, and poor real-time performance. Existing deep learning-based visual analysis systems largely depend on foreign GPU hardware and closed algorithm frameworks, posing a technological dependency risk. Furthermore, the separation of training, inference, and testing processes leads to low data flow efficiency and long model iteration cycles. In addition, labeling data for power equipment (such as insulator cracks and transformer oil temperature anomalies) relies on manual annotation, which is costly and difficult to adapt to complex scenarios. With the rapid development of artificial intelligence technology, intelligent applications in the power industry are gradually becoming a trend. Past methods relying on manual inspection and maintenance can no longer meet the needs of efficient operation and maintenance of modern power systems. Therefore, "less manned" or even "unmanned monitoring" has become the future direction. Leveraging technologies such as artificial intelligence, the Internet of Things, and big data, we can achieve efficient and real-time monitoring of power equipment, thereby further improving the operation and maintenance efficiency and safety of power systems. Therefore, there is an urgent need for an autonomous and controllable substation inspection visual analysis system that supports efficient data closed-loop processing and integrates visual training and inference. Summary of the Invention

[0003] The purpose of this invention is to solve at least one technical problem in the background art and to provide a substation inspection visual analysis system that integrates training and prediction.

[0004] To achieve the above objectives, the present invention provides a substation inspection visual analysis system integrating training and prediction, comprising: The basic support module is compatible with domestic GPU clusters and enables dynamic allocation of training, testing, and inference tasks through a heterogeneous computing framework. The data management module uses distributed file storage technology to store and manage various types of training datasets; The data annotation module annotates defects in power equipment in the centralized image or video data of the training dataset sent by the data management module; In the model training module, select the training data that has been labeled in the data labeling module, then select different training benchmark models and configure the relevant training parameters, and train each training benchmark model based on the selected training data to obtain multiple training models. The model testing module selects the test dataset sent by the data management module, performs capability tests on each training model using the test dataset, and selects the training model suitable for inspecting defects in power equipment based on the test results. The online inference module receives video stream data in real time, performs detection on the video stream data using a selected training model, and then outputs the detection results.

[0005] According to one aspect of the present invention, it further includes: a user management module, which, based on the RBAC model, implements data access permission hierarchy and operation audit log traceability.

[0006] According to one aspect of the present invention, the basic support module is based on a domestically developed and controllable GPU hardware layer, and achieves cross-platform compatibility of operators through a hardware abstraction layer, supporting FP16 / INT8 mixed precision calculation and memory access optimization.

[0007] According to one aspect of the present invention, the data management module uses distributed file storage technology to store image and video data, and uses permission management to ensure that each user can only operate on the data they create.

[0008] According to one aspect of the present invention, the data annotation module employs a Yolov8-based data annotation algorithm to automatically annotate power equipment defects in centralized image or video data of the training dataset sent by the data management module; After the data annotation module completes automatic annotation, it supports manual verification of the annotation results.

[0009] According to one aspect of the present invention, the online inference module accesses video stream data in real time, and the access method is WebSocket or FTP; The online inference module detects single or batch images using a selected training model and then outputs the detection results in real time.

[0010] According to the present invention, a substation inspection visual analysis system based on deep learning and domestically developed and controllable GPUs is provided, supporting an integrated training and inference platform for power equipment defect detection, condition monitoring, and full lifecycle data management. According to the solution of the present invention, compared with the prior art, the present invention can achieve the following beneficial effects: This integrated substation inspection visual analysis system, through its proprietary architecture, provides deep integration of the entire process of training, inference, and testing, avoiding multi-platform switching and saving data migration costs.

[0011] This training and prediction integrated substation inspection visual analysis system achieves cross-platform compatibility of operators based on the hardware abstraction layer (HAL) of domestically produced GPUs, and is compatible with mainstream domestic chips such as Cambricon and Ascend, realizing the security, autonomy and controllability of core hardware.

[0012] This integrated substation inspection visual analysis system, which combines training and inference, supports online inference functions and aims to provide efficient and accurate inference services to meet users' needs for real-time data processing. Attached Figure Description

[0013] Figure 1 The diagram illustrates the structure of a training and prediction integrated substation inspection visual analysis system according to one embodiment of the present invention. Detailed Implementation

[0014] The invention will now be discussed with reference to exemplary embodiments. It should be understood that the described embodiments are merely intended to enable those skilled in the art to better understand and thus implement the invention, and are not intended to imply any limitation on the scope of the invention.

[0015] As used herein, the term "comprising" and its variations are to be interpreted as open-ended terms meaning "including but not limited to". The term "based on" is to be interpreted as "at least partially based on". The terms "one embodiment" and "an embodiment" are to be interpreted as "at least one embodiment".

[0016] Figure 1 This schematic diagram illustrates the structural block diagram of a substation inspection visual analysis system integrating training and prediction according to one embodiment of the present invention. Figure 1 As shown, in this embodiment, the integrated training and prediction substation inspection visual analysis system includes: The basic support module is compatible with domestic GPU clusters and enables dynamic allocation of training, testing, and inference tasks through a heterogeneous computing framework. The data management module uses distributed file storage technology to store and manage various types of training datasets; The data annotation module annotates defects in power equipment in the centralized image or video data of the training dataset sent by the data management module; In the model training module, select the training data that has been labeled in the data labeling module, then select different training benchmark models and configure the relevant training parameters, and train each training benchmark model based on the selected training data to obtain multiple training models. The model testing module selects the test dataset sent by the data management module, performs capability tests on each training model using the test dataset, and selects the training model suitable for inspecting defects in power equipment based on the test results. The online inference module receives video stream data in real time, performs detection on the video stream data using a selected training model, and then outputs the detection results. The user management module, based on the RBAC model, implements hierarchical data access permissions and operation audit log traceability.

[0017] like Figure 1 As shown in this embodiment, the integrated training and inference substation inspection visual analysis system of the present invention is actually divided into two parts: hardware and software. The hardware environment constitutes the aforementioned basic support module, which includes a domestically produced GPU hardware layer and a storage system, providing a basic platform support for upper-layer data access, storage, model training, testing, and inference. The software environment adopts a B / S architecture, in which the front-end is a web-based annotation and visualization interface that supports image / video annotation of power equipment (such as rectangles, polygons, and key point annotations) and real-time monitoring data display.

[0018] The backend adopts a microservice architecture, including: The data management module is used to manage training datasets of different users and different types. The data annotation module is used to annotate the data in the original dataset; The model training module is used to train models based on a specified training dataset and evaluate the training results. The model testing module is used to implement model testing and evaluate test results based on a specified test dataset; The online inference module is used to perform model inference based on real-time data and a specified model, and to obtain data inference results; The user management module is used to implement identity authentication, permission management, data access control, and security auditing for the system and application. Furthermore, according to one embodiment of the present invention, the basic support module is based on a domestically developed and controllable GPU hardware layer. It achieves cross-platform compatibility of operators through a hardware abstraction layer, supporting FP16 / INT8 mixed-precision computation and memory access optimization. In this embodiment, the basic support module adapts to domestic GPU clusters, supports mixed-precision training, reduces memory usage, and accelerates model convergence. By dynamically allocating training, testing, and inference tasks through a heterogeneous computing framework, the efficiency of model training, testing, and inference is improved, providing a flexible, secure, and controllable basic support platform.

[0019] Furthermore, according to one embodiment of the present invention, the data management module uses distributed file storage technology to store image and video data. Through permission management, it ensures that each user can only operate on the data they create, guaranteeing data security and storage reliability. In this embodiment, the data management module supports the management of multiple data types (images, videos), has a built-in metadata tagging system specifically for power equipment, and supports automatic detection and manual correction of defects in power equipment (such as rust and cracks).

[0020] Furthermore, according to one embodiment of the present invention, the data annotation module adopts a data annotation algorithm based on Yolov8 to automatically annotate the defects of power equipment in the centralized image or video data of the training dataset sent by the data management module, thereby reducing the workload of manual annotation. After the data annotation module completes automatic annotation, it supports manual verification of the annotation results. This setting allows for manual adjustment of the automatic annotation results after the system completes automatic annotation, as well as manual annotation of the original dataset, improving the quality and completeness of the annotation results.

[0021] In this embodiment, the data annotation module integrates a semi-automatic annotation tool, combining a pre-trained model with active learning technology to prioritize the annotation of samples with high uncertainty.

[0022] Furthermore, according to one embodiment of the present invention, the model training module supports functions such as selecting labeled datasets, configuring model training, real-time viewing of model training progress, real-time monitoring of model training effects, and viewing of model training results, thus realizing the overall visualization of model training. Users can only select datasets uploaded by themselves for model training and can only operate on models trained by themselves, ensuring data security.

[0023] Furthermore, according to one embodiment of the present invention, the model testing module supports test dataset selection and test result query functions, realizing the testing of the capabilities of the trained model and facilitating the selection of a suitable model.

[0024] Furthermore, according to one embodiment of the present invention, the online inference module accesses video stream data in real time, and the access method is WebSocket or FTP; The online inference module detects single or batch images using a selected training model and then outputs the detection results in real time.

[0025] Furthermore, according to one embodiment of the present invention, the user management module is used to implement identity authentication, permission management, data access control and security auditing of the system and application to ensure data security and availability.

[0026] According to the above-described solution of the present invention, compared with the prior art, the present invention can achieve the following beneficial effects: This integrated substation inspection visual analysis system, through its proprietary architecture, provides deep integration of the entire process of training, inference, and testing, avoiding multi-platform switching and saving data migration costs.

[0027] This training and prediction integrated substation inspection visual analysis system achieves cross-platform compatibility of operators based on the hardware abstraction layer (HAL) of domestically produced GPUs, and is compatible with mainstream domestic chips such as Cambricon and Ascend, realizing the security, autonomy and controllability of core hardware.

[0028] This integrated substation inspection visual analysis system, which combines training and inference, supports online inference functions and aims to provide efficient and accurate inference services to meet users' needs for real-time data processing.

[0029] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention is not limited to the specific combination of the above-described technical features, but also includes other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in this invention.

[0030] It should be understood that the sequence number of each step in the invention and its embodiments does not absolutely imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

Claims

1. A substation inspection visual analysis system integrating training and inference, characterized in that, Comprise: A basic support module adapted to a domestic GPU cluster to realize dynamic distribution of training, testing and inference tasks through a heterogeneous computing framework; A data management module that uses distributed file storage technology to store and manage various types of training datasets; A data labeling module that labels power equipment defects in image or video data in the training dataset sent by the data management module; A model training module that selects training data labeled in the data labeling module, then selects different training benchmark models and configures related training parameters, trains each training benchmark model based on the selected training data, and obtains multiple training models; A model testing module that selects a test dataset sent by the data management module, tests the ability of each training model through the test dataset, and selects a training model suitable for inspecting power equipment defects according to the test results; An online inference module that accesses video stream data in real time, detects the video stream data through the selected training model, and then outputs the detection results.

2. The visual inspection system for power transformation and patrol according to claim 1, characterized in that, Also include: A user management module that realizes hierarchical data access permission and operation audit log tracing based on the RBAC model.

3. The visual inspection system for power transformation and patrol according to claim 1, wherein, The basic support module is based on a domestic self-controllable GPU hardware layer, and realizes cross-platform compatibility of operators through a hardware abstraction layer, supports FP16 / INT8 mixed precision calculation and video memory access optimization.

4. The visual inspection system for power transformation and patrol according to claim 1, characterized in that, The data management module uses distributed file storage technology to store image and video data, and through permission management, each user can only operate the data they establish.

5. The visual inspection system for power transformation and patrol according to claim 1, wherein, The data labeling module uses a Yolov8-based data labeling algorithm to automatically label power equipment defects in image or video data in the training dataset sent by the data management module; After the data labeling module completes automatic labeling, it supports manual verification of the labeling results.

6. The visual analysis system for power transformation inspection integrated with training and inference according to any one of claims 1-5, characterized in that, The online inference module accesses video stream data in real time, and the access method is WebScoket or FTP; The online inference module detects single images or batches of images through the selected training model, and then outputs the detection results in real time.

Images