CN114821448A - Intelligent analysis system and method for multi-channel videos of transformer substation - Google Patents

Abstract

The present application discloses an intelligent analysis system and method for multi-channel video in substations, mainly relates to the technical field of substation analysis, and is used to solve the problems of poor timeliness, low image recognition and processing efficiency, high cost, high resource occupation, and economical advantages of existing methods. gender issues, etc. Including: an interactive control unit for collecting video streams through an industrial computer processing module, extracting frames from the video stream, and determining the inference acceleration unit corresponding to the video stream; sending the video stream to the corresponding inference acceleration unit through the PCIE communication interface; inference acceleration A unit for processing video streams through several AI processing chips. In the present application, the identification efficiency and timeliness are improved by the above method, the operation and maintenance efficiency is greatly improved, and the effect of intelligent management of substations is further improved.

Description

An intelligent analysis system and method for multi-channel video in substation

technical field

The present application relates to the technical field of substation analysis, and in particular, to an intelligent analysis system and method for multi-channel video in substations.

Background technique

Substation is an important link in the field of power grid transmission, transformation, distribution, and power consumption. Its stable and reliable operation is an important guarantee for power grid security and power quality. In recent years, with the rapid development of advanced technologies such as artificial intelligence and edge computing, the operation and maintenance of substations has put forward higher requirements for intelligent equipment control and lean management. Unattended smart substations have become an important means to solve the shortage of personnel and intensive management.

The video data of the substation equipment of the existing invention is usually uniformly transmitted to the back-end server through a channel such as an optical fiber for analysis and processing. However, the existing methods increase the cost pressure of communication bandwidth and centralized data management, and reduce the timeliness and immediacy of data analysis. In particular, the number of video acquisition terminals in substations is large and continues to increase, and the quality and resolution of front-end video imaging continue to improve. , the performance requirements of communication channels and background servers continue to increase, and the efficiency of image recognition and processing drops significantly. Or the current solution uses a single channel of video to independently configure a single edge recognition device, which is costly, occupies a lot of resources, and is not economical.

SUMMARY OF THE INVENTION

In view of the above shortcomings of the prior art, the present invention provides an intelligent analysis system and method for multi-channel video in a substation to solve the above technical problems.

In the first aspect, the present application provides an intelligent analysis system for multi-channel video in substations. The system includes: an interactive control unit for collecting video streams through an industrial computer processing module, extracting frames from the video streams, and determining corresponding video streams. Inference acceleration unit; through the PCIE communication interface, the video stream is sent to the corresponding inference acceleration unit; the inference acceleration unit is used to process video streams through several AI processing chips.

Further, the interaction control unit includes a plurality of PCIE communication interfaces to communicate with different inference acceleration units respectively.

Further, the processing module of the industrial computer adopts an embedded Linux operating system and deploys a Docker container.

Further, the inference acceleration unit adopts an inference acceleration board based on AI processing chips, and each AI processing chip can process 3 video streams.

Further, the inference acceleration unit includes at least the Faster-RCNN framework algorithm identification model, the Yolo framework algorithm identification model, the TensorFlow framework algorithm identification model and the acceleration unit; the computing power provided by the acceleration unit is 105.6TOPsINT8 computing power or 6.6TOPs FP32 computing power, with Supports the solidification call and calculation acceleration of the Faster-RCNN algorithm recognition model, the Yolo algorithm recognition model, and the TensorFlow framework algorithm recognition model.

Further, the system further includes: a relay switch and a multi-channel camera; the multi-channel cameras are connected to the relay switch, and then send a video stream to the interactive control unit through the relay switch.

Further, the interactive control unit is also used to receive the video stream processed by the inference acceleration unit; to optimize and push the processed video stream to obtain the final video stream; the system further includes a main control module; the main control module, It is used to receive the final video stream and display the final video stream on the preset display interface.

In a second aspect, the present application provides an intelligent analysis method for multi-channel video in a substation. The method includes: collecting a video stream through an industrial computer processing module, extracting frames from the video stream, and determining an edge computing server corresponding to the video stream; communicating through PCIE interface to send the video stream to the corresponding edge computing server; process the video stream through several AI processing chips in the edge computing server to obtain the processed video stream; optimize and push the processed video stream to obtain the final video stream, and display the final video stream on the preset display interface.

It can be understood by those skilled in the art that the present invention at least has the following beneficial effects:

A real-time intelligent analysis method and device for multi-channel video streams applied to the edge side of substations is proposed, which can realize real-time collection of multi-channel video streams on the edge side of substations, online intelligent identification of defects, acceleration of edge computing inference, and video processing and streaming, etc. Function, the method and device can access multiple video streams at the same time, and complete the intelligent identification processing of multiple videos at the edge side at the same time, reduce the bandwidth pressure of network transmission from the edge side of the substation to the centralized control server, and improve the identification efficiency and efficiency. Timeliness, operation and maintenance efficiency have been greatly improved, and the effect of intelligent management of substations has been further improved.

Description of drawings

Some embodiments of the present disclosure are described below with reference to the accompanying drawings, in which:

FIG. 1 is a schematic diagram of the internal structure of an intelligent analysis system for multi-channel video in a substation provided by an embodiment of the present application.

FIG. 2 is a flowchart of an intelligent analysis method for multi-channel video in a substation provided by an embodiment of the present application.

Detailed ways

It should be understood by those skilled in the art that the embodiments described below are only preferred embodiments of the present disclosure, which do not mean that the present disclosure can only be implemented by the preferred embodiments, and the preferred embodiments are only used to explain the present disclosure. The technical principle is not used to limit the protection scope of the present disclosure. Based on the preferred embodiments provided by the present disclosure, all other embodiments obtained by those of ordinary skill in the art without creative efforts should still fall within the protection scope of the present disclosure.

It should also be noted that the terms "comprising", "comprising" or any other variation thereof are intended to encompass a non-exclusive inclusion such that a process, method, article or device comprising a series of elements includes not only those elements, but also Other elements not expressly listed, or which are inherent to such a process, method, article of manufacture, or apparatus are also included. Without further limitation, an element qualified by the phrase "comprising a..." does not preclude the presence of additional identical elements in the process, method, article of manufacture, or device that includes the element.

The technical solutions proposed by the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

FIG. 1 is an intelligent analysis system for multi-channel video in a substation according to an embodiment of the present application. As shown in FIG. 1 , the system provided by the embodiment of the present application mainly includes: an interaction control unit 110 and an inference acceleration unit 120 .

The interactive control unit 110 is used to collect the video stream through the industrial computer processing module 111, frame the video stream, and determine the inference acceleration unit 120 corresponding to the video stream; send the video stream to the corresponding inference acceleration unit 120 through the PCIE communication interface 112 It should be noted that the interaction control unit 110 is connected with several inference acceleration units 120. As an example, the interaction control unit 110 includes several PCIE communication interfaces 112 to communicate with different inference acceleration units 120 respectively. Among them, the industrial computer processing module 111 mainly completes functions such as video stream collection, frame extraction, network communication, process scheduling, video stream push, human-computer interaction, etc. As an example, the industrial computer processing module 111 adopts an embedded Linux operating system and deploys Docker container.

To sum up, the interactive control unit 110 may specifically be: based on the industrial computer processing module 111, using an embedded Linux operating system, and deploying a Docker container, mainly to complete video stream collection, frame extraction, network communication, process scheduling, video stream push, human-machine Interaction and other functions are provided, and the network and PCIE communication interface 112 is provided to realize the scheduling control of multiple video streams in different inference acceleration units 120 for edge computing.

In addition, before the interaction control unit 110 obtains the video stream, the system further includes a relay switch and multi-channel cameras; the multi-channel cameras are connected to the relay switch, and then send the video stream to the interaction control unit 110 through the relay switch.

The inference acceleration unit 120 is used for processing video streams through several AI processing chips 121 .

As an example, the inference acceleration unit 120 adopts an inference acceleration board based on the AI processing chip 121 , and each AI processing chip 121 can process three video streams.

It should be noted that the inference acceleration unit 120 can import the video stream into different algorithm models for processing. As an example, the inference acceleration unit 120 at least includes the Faster-RCNN framework algorithm identification model, the Yolo framework algorithm identification model, and the TensorFlow framework algorithm identification model. and acceleration unit; the computing power provided by the acceleration unit is 105.6TOPs INT8 computing power or 6.6TOPs FP32 computing power to support the solidification call and calculation acceleration of the Faster-RCNN algorithm recognition model, the Yolo algorithm recognition model, and the TensorFlow framework algorithm recognition model.

To sum up, the inference acceleration unit 120 may specifically be: communicate with the interactive control unit 110 through the PCIE interface, adopt an inference acceleration board based on the AI processing chip 121, and each inference acceleration unit 120 is configured with three AI processing chips 121, each processing The chip can process 3 video streams, and each device can be configured with multiple inference acceleration units 120 according to the number of PCIE interfaces of the interactive control unit 110 . It supports the solidification call and calculation acceleration of recognition models of various framework algorithms such as Faster-RCNN, Yolo, and TensorFlow. A single acceleration unit can provide up to 105.6TOPs INT8 computing power and 6.6TOPs FP32 computing power, supporting high-precision computing.

In addition, the interaction control unit 110 is also used to receive the video stream processed by the inference acceleration unit 120; to optimize and push the processed video stream to obtain the final video stream; the system further includes a main control module; the main control module , which is used to receive the final video stream and display the final video stream on the preset display interface.

As an example, the present application may specifically be as follows: accessing multiple cameras through a relay switch, the interaction control unit 110 completes the multi-channel camera video data collection and access and frame extraction operations, and then based on memory interaction, the multiple video streams are respectively The frame-drawing picture is transferred into the memory of the inference acceleration unit 120, and the artificial intelligence algorithm model is called by the inference acceleration unit 120 to analyze and identify the picture, and then the picture and the video data are hard compressed, and the result is forwarded to the interactive control unit 110 memory, Finally, the interactive control unit 110 optimizes and pushes the video stream, and displays it in the main control module in real time.

It can be understood by those skilled in the art that the present application realizes the simultaneous real-time online intelligent analysis and identification of multiple video streams, and solves the problem of poor real-time performance caused by alternate identification of multiple video streams under the condition of single-channel video stream analysis and identification. It solves the problems of unified transmission of multiple video streams to the server, resulting in large transmission bandwidth pressure, long calculation time, and large resource occupation, and provides a method for intelligent management and control of substations. By deploying multiple inference acceleration units 120 to analyze and identify multiple video streams in different inference acceleration units 120, computing resources and inference capabilities are ensured on the edge side, the real-time performance and reliability of analysis and identification are improved, and the solution to the problem is solved. On the terminal side, the analysis and identification cause the shortage of computing resources, poor identification effect, and low reliability. It solves the problems of high resource occupation, high cost and low real-time performance caused by analysis and identification on the server side. On the basis of ensuring the accuracy, real-time performance and reliability of intelligent analysis and identification, it reduces costs and improves economy. In addition, a plurality of inference acceleration units 120 can be expanded in one device, which reduces the cost, improves the computing efficiency, and meets the real-time requirements of the intelligent substation inspection service.

In addition, the embodiment of the present application also provides an intelligent analysis method for multi-channel video in a substation. As shown in FIG. 2 , the method provided by the embodiment of the present application mainly includes the following steps:

Step 210: Collect the video stream through the industrial computer processing module, extract frames from the video stream, and determine the edge computing server corresponding to the video stream; send the video stream to the corresponding edge computing server through the PCIE communication interface.

Step 220: Process the video stream through several AI processing chips in the edge computing server to obtain the processed video stream.

Step 230: Optimize and push the processed video stream to obtain a final video stream, and display the final video stream on a preset display interface.

So far, the technical solutions of the present disclosure have been described in conjunction with the foregoing embodiments, but those skilled in the art will readily understand that the protection scope of the present disclosure is not limited to these specific embodiments. Without departing from the technical principles of the present disclosure, those skilled in the art can split and combine the technical solutions in the above-mentioned embodiments, and can also make equivalent changes or substitutions to the relevant technical features. Any modification, equivalent replacement, improvement, etc. made within the concept and/or technical principle will fall within the protection scope of the present disclosure.

Claims (8)

1. an intelligent analysis system for substation multi-channel video, is characterized in that, described system comprises: An interactive control unit, configured to collect a video stream through an industrial computer processing module, extract frames from the video stream, and determine an inference acceleration unit corresponding to the video stream; send the video stream to the corresponding inference acceleration unit through the PCIE communication interface; The inference acceleration unit is used to process video streams through several AI processing chips. 2 . The intelligent analysis system for multi-channel video in a substation according to claim 1 , wherein the interactive control unit comprises a plurality of PCIE communication interfaces to communicate with different inference acceleration units respectively. 3 . 3. The intelligent analysis system for multi-channel video in substation according to claim 1, is characterized in that, described industrial computer processing module adopts embedded Linux operating system and arranges Docker container. 4. The intelligent analysis system for multi-channel video in a substation according to claim 1, wherein the inference acceleration unit adopts an inference acceleration board based on an AI processing chip, and each AI processing chip can process 3 channels video stream. 5. the intelligent analysis system for substation multi-channel video according to claim 1, is characterized in that, described inference acceleration unit at least comprises Faster-RCNN framework algorithm identification model, Yolo framework algorithm identification model, TensorFlow framework algorithm identification model and acceleration unit; the computing power provided by the acceleration unit is 105.6TOPs INT8 computing power or 6.6TOPsFP32 computing power to support the solidification call and calculation acceleration of the Faster-RCNN algorithm recognition model, the Yolo algorithm recognition model, and the TensorFlow framework algorithm recognition model. 6. The intelligent analysis system for substation multi-channel video according to claim 1, wherein the system further comprises: a relay switch and a multi-channel camera; The multi-channel cameras are connected to the relay switch, and then the video stream is sent to the interaction control unit through the relay switch. 7. The intelligent analysis system for substation multi-channel video according to claim 1, wherein the interactive control unit is also used to receive the video stream processed by the inference acceleration unit; The video stream is optimized and pushed to obtain the final video stream; The system also includes a main control module; The main control module is configured to receive the final video stream and display the final video stream on a preset display interface. 8. An intelligent analysis method for multi-channel video in a substation, wherein the method comprises: Collect the video stream through the processing module of the industrial computer, extract frames from the video stream, and determine the edge computing server corresponding to the video stream; send the video stream to the corresponding edge computing server through the PCIE communication interface; Process the video stream through several AI processing chips in the edge computing server to obtain the processed video stream; The processed video stream is optimized and pushed to obtain a final video stream, and the final video stream is displayed on a preset display interface.

Images