CN111582016A - Intelligent maintenance-free power grid monitoring method and system based on cloud-edge collaborative deep learning - Google Patents

Abstract

The invention discloses an intelligent maintenance-free power grid monitoring method and system based on cloud-edge collaborative deep learning. The method includes: sensing equipment collecting monitoring images, and alarming according to alarm information; edge computing nodes based on the monitoring images and cloud computing center download The power grid abnormality detection model developed by the system identifies the monitoring image, and when it is determined that there is a power grid abnormality, sends alarm information to the sensing device; transmits the identified monitoring image to the cloud computing center; the cloud computing center trains the depth of training images according to the stored training set The learning model generates a power grid abnormality detection model, trains and updates the power grid abnormality detection model according to the identified monitoring images, and sends the power grid abnormality detection model to an edge computing node; the invention also discloses a corresponding power grid monitoring system; The system can reduce the computing burden of the cloud computing center.

Description

Intelligent maintenance-free power grid monitoring method and system based on cloud-edge collaborative deep learning

technical field

The invention belongs to the field of power grid security protection, and in particular relates to an intelligent maintenance-free power grid monitoring method and system based on cloud-edge collaborative deep learning.

Background technique

The application of intelligent video surveillance in power systems is particularly important, while traditional video surveillance has shortcomings in timely processing and post-event forensics, and massive videos require a lot of manpower to identify and analyze abnormalities in video images. With the development of intelligence and deep learning, intelligent surveillance video has gradually become a development trend. Artificial intelligence algorithms are used to continuously analyze video images. Once an abnormality is found, an early warning message is immediately sent to the staff, which can reduce the workload of equipment maintenance for staff. , to improve the efficiency and effectiveness of the entire monitoring and management.

Cloud-based video surveillance technology reduces the cost of construction and maintenance for users, and the centralized computing and storage method of the cloud improves the security and reliability of video data; however, video data has two characteristics. One is that video surveillance generates The data of the cloud is mainly unstructured data, and the second is the explosive growth trend of video data. Therefore, the cloud-based video surveillance technology has the following problems:

The images obtained from massive video streams are transmitted to the cloud computing center, which will consume a lot of network bandwidth, which may cause service interruption, network delay and other problems, so the real-time performance cannot be guaranteed;

Image processing tasks are concentrated in the cloud computing center, which requires a lot of computing resources and increases the computing burden of the cloud computing center.

SUMMARY OF THE INVENTION

The purpose of the present invention is to solve one of the existing problems, provide an intelligent maintenance-free power grid monitoring method and system based on cloud-edge collaborative deep learning, and reduce the computing burden of the cloud computing center.

In order to achieve the purpose, an intelligent maintenance-free power grid monitoring method based on cloud-edge collaborative deep learning is used in a power grid monitoring system, where the power grid monitoring system includes a sensing device, an edge computing node and a cloud computing center, and the method includes:

Sensing equipment collects monitoring images;

The edge computing node obtains the monitoring images collected by the sensing equipment;

The edge computing node identifies the monitoring image to determine whether there is a power grid abnormality based on the monitoring image and the power grid abnormality detection model issued by the cloud computing center, and sends alarm information to the sensing device when there is a power grid abnormality;

The sensing device receives the alarm information of the edge computing node, and sends an alarm according to the alarm information;

The edge computing node transmits the identified surveillance images to the cloud computing center;

The cloud computing center trains a deep learning model according to the stored training set images to generate a power grid abnormality detection model, and trains and updates the power grid abnormality detection model according to the identified monitoring images transmitted by the edge computing nodes;

The cloud computing center sends the grid anomaly detection model to the edge computing nodes.

Optionally, the method further includes:

The edge computing node preprocesses the monitoring image;

The edge computing node deduplicates the monitoring images collected within the interval;

The monitoring image is identified based on the monitoring image and the power grid abnormality detection model issued by the cloud computing center to determine whether there is a power grid abnormality: according to the power grid abnormality detection model, the preprocessed and deduplicated monitoring images are identified to determine Whether there is grid abnormality in the monitoring image.

Optionally, transmitting the identified surveillance images to the cloud computing center includes:

Temporarily store the identified surveillance images in edge computing nodes;

The temporarily stored monitoring images are uniformly transmitted to the cloud computing center in the time period when the network bandwidth flow is less than the preset value.

Optionally, training a deep learning model to generate a power grid anomaly detection model according to the stored training set images includes:

Adjust the training set image to 320*320;

Build a lightweight target detection network ThunderNet as a deep learning model;

The deep learning model is trained according to the adjusted training set images to generate a power grid anomaly detection model.

Optionally, the method further includes:

Cloud computing center builds and trains abnormal traffic monitoring model;

The cloud computing center sends the trained abnormal traffic monitoring model to the edge computing nodes;

The edge computing node monitors the traffic at the edge according to the abnormal traffic monitoring model, and sends an abnormal traffic signal to the cloud computing center if abnormal traffic is found at the edge;

When the cloud computing center receives the abnormal traffic signal, it will block the abnormal traffic.

In another aspect of the present disclosure, an intelligent maintenance-free power grid monitoring system based on cloud-edge collaborative deep learning includes sensing devices, edge computing nodes, and cloud computing centers;

The sensing device includes:

The image acquisition component is used to collect monitoring images;

an alarm device, used for receiving alarm information of the edge computing node, and alarming according to the alarm information;

The edge computing node includes:

The image processing module is used to obtain the monitoring image collected by the sensing device;

an image detection module for identifying the monitoring image and determining whether there is a power grid abnormality based on the monitoring image and the power grid abnormality detection model issued by the cloud computing center;

The alarm module is used to send alarm information to the sensing device when there is an abnormality in the power grid;

a first transmission module, used for transmitting the identified monitoring images to the cloud computing center;

The cloud computing center includes:

a power grid abnormality detection model training module, which trains a deep learning model according to the stored training set images to generate a power grid abnormality detection model, and trains and updates the power grid abnormality detection model according to the identified monitoring images transmitted by the edge computing nodes;

The second transmission module is used for delivering the grid abnormality detection model to the edge computing node.

Optionally, the image acquisition component is connected to a camera that shoots power monitoring videos, and is used to acquire monitoring images from video streams captured by the camera.

Optionally, the edge computing node further includes a repetition filtering module;

The image processing module is also used for preprocessing the monitoring image;

The repeated filtering module is used to perform a deduplication operation on the monitoring images collected within the interval;

The said monitoring image is identified based on the power grid abnormality detection model issued by the cloud computing center to determine whether there is a power grid abnormality: according to the power grid abnormality detection model, the preprocessed and deduplicated monitoring images are identified to determine the Check whether there is an abnormality in the power grid in the monitoring image.

Optionally, transmitting the identified surveillance images to the cloud computing center includes: temporarily storing the identified surveillance images in the edge computing node, and uniformly transmitting the temporarily stored surveillance images to the cloud during the time period when the network bandwidth and traffic is less than a preset value. computing center.

Optionally, training a deep learning model to generate a power grid anomaly detection model according to the stored training set images includes:

Adjust the training set image to 320*320;

Build a lightweight target detection network ThunderNet as a deep learning model;

The deep learning model is trained according to the adjusted training set images to generate a power grid anomaly detection model.

The following beneficial technical effects can be achieved by implementing the technical solutions of the present disclosure:

The cloud computing center is used to train and update the power grid abnormality detection model; the power grid abnormality detection model trained and updated by the edge computing nodes is used to judge whether there is a power grid abnormality detection model based on the monitoring images, the power grid abnormality detection model issued by the cloud computing center, and the monitoring images collected by the sensing equipment. The power grid is abnormal, and when the power grid is abnormal, alarm information is sent to the corresponding sensing equipment, and the sensing equipment will give an alarm, so that the staff can know the power grid abnormality in time, and take effective measures in time for the power grid abnormality.

The cloud computing center is mainly used to train and update the power grid anomaly detection model, and the identification work is allocated to each edge computing node, which greatly reduces the operation volume of the cloud computing center and reduces the computing burden of the cloud computing center.

Massive video and image data are directly uploaded to the cloud computing center. The network bandwidth pressure is large, which often results in network delay. Edge computing nodes process the image data near the data source, make real-time judgments, effectively solve problems such as interaction delay, and reduce bandwidth. cost

Description of drawings

1 is a flowchart of an intelligent maintenance-free power grid monitoring method based on cloud-edge collaborative deep learning in one embodiment of the present disclosure;

2 is a block diagram of an intelligent maintenance-free power grid monitoring system based on cloud-edge collaborative deep learning in an embodiment of the present disclosure;

3 is another block diagram of an intelligent maintenance-free power grid monitoring system based on cloud-edge collaborative deep learning in an embodiment of the present disclosure;

4 is another block diagram of an intelligent maintenance-free grid monitoring system based on cloud-edge collaborative deep learning in an embodiment of the present disclosure;

FIG. 5 is another block diagram of an intelligent maintenance-free power grid monitoring system based on cloud-edge collaborative deep learning in an embodiment of the present disclosure.

Detailed ways

In order to facilitate the understanding of those skilled in the art, the present invention will be further described below in conjunction with specific embodiments:

Referring to Figures 1 and 2, an intelligent maintenance-free power grid monitoring method based on cloud-edge collaborative deep learning is used in a power grid monitoring system. The power grid monitoring system includes a sensing device 1, an edge computing node 2 and a cloud computing center 3, and the method includes:

S1: Sensing equipment collects monitoring images;

S2: The edge computing node obtains the monitoring image collected by the sensing device;

S3: Based on the monitoring image and the grid abnormality detection model issued by the cloud computing center, the edge computing node identifies the monitoring image to determine whether there is grid abnormality, and sends alarm information to the sensing device when grid abnormality exists;

S4: The sensing device receives the alarm information of the edge computing node, and generates an alarm according to the alarm information;

S5: The edge computing node transmits the identified surveillance image to the cloud computing center;

S6: The cloud computing center trains a deep learning model according to the stored training set images to generate a power grid abnormality detection model, and trains and updates the power grid abnormality detection model according to the identified monitoring images transmitted by the edge computing nodes;

S7: The cloud computing center delivers the grid anomaly detection model to the edge computing nodes.

It can be known that the above-mentioned edge computing node 2 can communicate with multiple sensing devices 1, and the cloud computing center 3 can communicate with multiple sensing devices.

It should be noted that the above numbers S1 to S6 do not represent that the execution sequence of the steps is from S1 to S6.

In the intelligent maintenance-free power grid monitoring method based on cloud-edge collaborative deep learning of the present disclosure, the cloud computing center is used to train and update the power grid abnormality detection model; the power grid abnormality detection model trained and updated by the edge computing nodes is based on the monitoring images and the cloud computing center. The issued power grid abnormality detection model and the monitoring images collected by the sensing equipment determine whether there is a power grid abnormality, and when the power grid is abnormal, send alarm information to the corresponding sensing equipment, and the sensing equipment will give an alarm, so that the staff can know in time. If the power grid is abnormal, take effective measures in time for the abnormal power grid.

Massive video and image data are directly uploaded to the cloud computing center. The network bandwidth pressure is large, which often results in network delay. Edge computing nodes process the image data near the data source, make real-time judgments, effectively solve problems such as interaction delay, and reduce bandwidth. cost;

The deep learning calculation is completed by the cloud computing center and the edge side. The cloud computing center is responsible for the basic model training of the large amount of historical data, and based on the computing and storage capabilities, the model is optimized and sent to the edge side to complete the inference process of the training model. , that is, the detection of real-time images, which realizes the intelligent collaboration between the cloud and the edge;

Since the cloud computing center is mainly used to train and update the grid anomaly detection model, the identification work is allocated to each edge computing node, which greatly reduces the operation volume of the cloud computing center and reduces the computing burden of the cloud computing center.

Since the cloud computing center will train and update the grid abnormality detection model based on the identified monitoring images uploaded by the edge computing nodes, the grid abnormality detection model will be more perfect, and the recognition rate of the grid abnormality detection model will be improved. Further manual verification can be carried out by the staff, and the verified monitoring images will be trained and updated for the grid anomaly detection model.

In one embodiment, the intelligent maintenance-free power grid monitoring method based on cloud-edge collaborative deep learning further includes:

The edge computing node preprocesses the monitoring image;

The edge computing node deduplicates the monitoring images collected within the interval;

The said monitoring image is identified based on the power grid abnormality detection model issued by the cloud computing center to determine whether there is a power grid abnormality: according to the power grid abnormality detection model, the preprocessed and deduplicated monitoring images are identified to determine the The monitoring image is used to determine whether there is an abnormality in the power grid.

In the embodiment of the present disclosure, the main purpose of preprocessing is to eliminate irrelevant information in the image, which includes removing mixed images, adjusting the size of the image to 320*320 pixels, and performing a normalization operation on the image. Since the acquisition of images in the video stream will result in a large number of repeated pictures in the interval, the de-duplication operation can reduce unnecessary repeated recognition operations and greatly improve the efficiency of recognition.

In one embodiment, a grid anomaly detection model can identify and locate grid anomalies.

In one embodiment, when an abnormality of the power grid is identified, the corresponding label text is used as the alarm information; at the same time, if the continuous frame number images are the same alarm information within a preset interval, the alarm information is only sent to the alarm device once. Avoid repeated alarms frequently.

In one embodiment, transmitting the identified surveillance images to the cloud computing center includes:

Temporarily store the identified surveillance images in edge computing nodes;

The temporarily stored monitoring images are uniformly transmitted to the cloud computing center in the time period when the network bandwidth flow is less than the preset value.

By selecting a time period with low network proxy traffic and sending monitoring images, loan occupancy can be reduced.

In one embodiment, training a deep learning model to generate a power grid anomaly detection model according to the stored training set images includes:

Adjust the size of the training set image to 320*320;

Build a lightweight target detection network ThunderNet as a deep learning model;

The deep learning model is trained according to the adjusted training set images to generate a power grid anomaly detection model.

ThunderNet consists of two parts: the backbone network and the detection network. The backbone network is SNet, and the detection network uses the Light-Head R-CNN network architecture. In the process of training the deep learning model according to the adjusted training set images, AP can be used as a performance evaluation indicator. When the accuracy of the model on each class reaches a preset threshold, the training is stopped; the cloud computing center is based on the deployment of this The application node computing and storage capacity required by the model, optimize the trained model, and generate the ThunderNet grid anomaly detection model required in the edge computing node.

When adjusting the size of the images in the training set, cleaning the image data, removing redundant and mixed images, and normalizing the images. In order to be consistent with the preprocessing operations performed by edge computing nodes on monitoring images.

In one embodiment, the intelligent maintenance-free power grid monitoring method based on cloud-edge collaborative deep learning further includes:

The cloud computing center builds and trains the abnormal traffic monitoring model DAE-RNN, uses the historical traffic data stored in the cloud, trains the built model, and mines the inherent difference between normal traffic and abnormal traffic;

The cloud computing center sends the trained abnormal traffic monitoring model to the edge computing nodes;

The edge computing node monitors the traffic at the edge according to the abnormal traffic monitoring model, and sends an abnormal traffic signal to the cloud computing center if abnormal traffic is found at the edge;

When the cloud computing center receives the abnormal traffic signal, it will block the abnormal traffic.

The computing center provides security for the edge side. When the edge side detects abnormal traffic, it notifies the cloud computing center to block it, realizing the security collaboration between the cloud and the edge.

Example 2:

Referring to Figures 2 to 5, an intelligent maintenance-free power grid monitoring system based on cloud-edge collaborative deep learning includes a sensing device 1, an edge computing node 2 and a cloud computing center 3;

Sensing device 1, including:

an image acquisition component 11 for acquiring monitoring images;

The alarm device 12 is used for receiving alarm information of the edge computing node, and alarming according to the alarm information;

Edge computing node 2, including:

The image processing module 21 is used for acquiring the monitoring image collected by the sensing device;

The image detection module 22 is configured to identify the monitoring image and determine whether there is a power grid abnormality based on the monitoring image and the power grid abnormality detection model issued by the cloud computing center;

an alarm module 23, configured to send alarm information to the sensing device when there is an abnormality in the power grid;

The first transmission module 24 is used for transmitting the identified monitoring image to the cloud computing center;

Cloud Computing Center 3, including:

The model training module 31 trains a deep learning model according to the stored training set images to generate a power grid abnormality detection model, and trains and updates the power grid abnormality detection model according to the identified monitoring images transmitted by the edge computing nodes;

The second transmission module 32 is configured to deliver the grid abnormality detection model to the edge computing node.

In the intelligent maintenance-free grid monitoring system based on cloud-edge collaborative deep learning of the present disclosure, the cloud computing center is used to train and update the grid abnormality detection model; the grid abnormality detection model trained and updated by the edge computing nodes is based on the monitoring images and the cloud computing center. The issued power grid abnormality detection model and the monitoring images collected by the sensing equipment determine whether there is a power grid abnormality, and when the power grid is abnormal, send alarm information to the corresponding sensing equipment, and the sensing equipment will give an alarm, so that the staff can know in time. If the power grid is abnormal, take effective measures in time for the abnormal power grid.

Massive video and image data are directly uploaded to the cloud computing center. The network bandwidth pressure is large, which often results in network delay. Edge computing nodes process the image data near the data source, make real-time judgments, effectively solve problems such as interaction delay, and reduce bandwidth. cost;

The deep learning calculation is completed by the cloud computing center and the edge side. The cloud computing center is responsible for the basic model training of the large amount of historical data, and based on the computing and storage capabilities, the model is optimized and sent to the edge side to complete the inference process of the training model. , that is, the detection of real-time images, which realizes the intelligent collaboration between the cloud and the edge;

Since the cloud computing center is mainly used to train and update the grid anomaly detection model, the identification work is allocated to each edge computing node, which greatly reduces the operation volume of the cloud computing center and reduces the computing burden of the cloud computing center.

Since the cloud computing center will train and update the grid abnormality detection model based on the identified monitoring images uploaded by the edge computing nodes, the grid abnormality detection model will be more perfect, and the recognition rate of the grid abnormality detection model will be improved. Further manual verification can be carried out by the staff, and the verified monitoring images will be trained and updated for the grid anomaly detection model.

In one embodiment, referring to FIG. 3 , the image acquisition component 11 is connected to a camera that captures power monitoring video, and is used to acquire monitoring images from the video stream captured by the camera.

In one embodiment, referring to FIG. 4 , the edge computing node further includes a repetition filtering module 25;

The image processing module 21 is further configured to perform a preprocessing operation on the monitoring image;

repeating the filtering module 25, for performing deduplication operations on the monitoring images collected within the interval;

Based on the monitoring image and the power grid abnormality detection model issued by the cloud computing center, identifying the monitoring image to determine whether there is a power grid abnormality includes: identifying the preprocessed and deduplicated monitoring image according to the power grid abnormality detection model to determine the power grid abnormality detection model. Monitor the image to determine whether there is an abnormality in the power grid.

In the embodiment of the present disclosure, the main purpose of preprocessing is to eliminate irrelevant information in the image, which includes removing mixed images, adjusting the size of the image to 320*320, and performing a normalization operation on the image. Since the acquisition of images in the video stream will result in a large number of repeated pictures in the interval, the de-duplication operation can reduce unnecessary repeated recognition operations and greatly improve the efficiency of recognition.

In one embodiment, when an abnormality of the power grid is identified, the corresponding label text is used as the alarm information; at the same time, if the continuous frame number images are the same alarm information within a preset interval, the alarm information is only sent to the alarm device once. Avoid repeated alarms frequently.

In one embodiment, transmitting the identified monitoring images to the cloud computing center includes: temporarily storing the identified monitoring images on edge computing nodes, and uniformly transmitting the temporarily stored monitoring images during a time period when the network bandwidth and traffic is less than a preset value to the cloud computing center. By selecting a time period with low network proxy traffic and sending monitoring images, loan occupancy can be reduced.

In one embodiment, training a deep learning model to generate a power grid anomaly detection model according to the stored training set images includes:

Adjust the training set image to 320*320;

Build a lightweight target detection network ThunderNet as a deep learning model;

The deep learning model is trained according to the adjusted training set images to generate a power grid anomaly detection model.

ThunderNet consists of two parts: the backbone network and the detection network. The backbone network is SNet, and the detection network uses the Light-Head R-CNN network architecture. In the process of training the deep learning model according to the adjusted training set images, AP can be used as a performance evaluation indicator. When the accuracy of the model on each class reaches a preset threshold, the training is stopped; the cloud computing center is based on the deployment of this The application node computing and storage capacity required by the model, optimize the trained model, and generate the ThunderNet grid anomaly detection model required in the edge computing node.

When adjusting the size of the images in the training set, cleaning the image data, removing redundant and mixed images, and normalizing the images. In order to be consistent with the preprocessing operations performed by edge computing nodes on monitoring images.

The intelligent maintenance-free grid monitoring system based on cloud-edge collaborative deep learning also includes a flow monitoring module, which is used for:

Build and train the abnormal traffic monitoring model DAE-RNN, use the historical traffic data stored in the cloud to train the built model, and mine the inherent differences between normal traffic and abnormal traffic;

Send the trained abnormal traffic monitoring model to edge computing nodes;

When an abnormal flow signal is received, the abnormal flow is blocked.

The edge computing node is further configured to monitor the traffic at the edge according to the abnormal traffic monitoring model, and if abnormal traffic is found at the edge, send a traffic abnormal signal to the cloud computing center.

The computing center provides security for the edge side. When the edge side detects abnormal traffic, it notifies the cloud computing center to block it, realizing the security collaboration between the cloud and the edge.

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 above are only specific embodiments of the present invention, but the technical features of the present invention are not limited thereto. Any changes or modifications made by those skilled in the art in the field of the present invention are all covered by the patent scope of the present invention. among.

Claims (10)

1. The intelligent maintenance-free power grid monitoring method based on cloud-edge collaborative deep learning, is characterized in that, is used for power grid monitoring system, and described power grid monitoring system comprises sensing equipment, edge computing node and cloud computing center, and described method comprises: Sensing equipment collects monitoring images; The edge computing node obtains the monitoring images collected by the sensing equipment; The edge computing node identifies the monitoring image to determine whether there is a power grid abnormality based on the monitoring image and the power grid abnormality detection model issued by the cloud computing center, and sends alarm information to the sensing device when there is a power grid abnormality; The sensing device receives the alarm information of the edge computing node, and sends an alarm according to the alarm information; The edge computing node transmits the identified surveillance images to the cloud computing center; The cloud computing center trains a deep learning model according to the stored training set images to generate a power grid abnormality detection model, and trains and updates the power grid abnormality detection model according to the identified monitoring images transmitted by the edge computing nodes; The cloud computing center sends the grid anomaly detection model to the edge computing nodes. 2. The intelligent maintenance-free power grid monitoring method based on cloud-edge collaborative deep learning as claimed in claim 1, characterized in that, The method also includes: The edge computing node preprocesses the monitoring image; The edge computing node deduplicates the monitoring images collected within the interval; The said monitoring image is identified based on the power grid abnormality detection model issued by the cloud computing center to determine whether there is a power grid abnormality: according to the power grid abnormality detection model, the preprocessed and deduplicated monitoring images are identified to determine the Check whether there is an abnormality in the power grid in the monitoring image. 3. The intelligent maintenance-free power grid monitoring method based on cloud-edge collaborative deep learning as claimed in claim 2, wherein transmitting the identified monitoring image to the cloud computing center comprises: Temporarily store the identified surveillance images in edge computing nodes; The temporarily stored monitoring images are uniformly transmitted to the cloud computing center in the time period when the network bandwidth flow is less than the preset value. 4. The intelligent maintenance-free power grid monitoring method based on cloud-edge collaborative deep learning as claimed in claim 3, characterized in that, according to the stored training set image training deep learning model to generate power grid abnormality detection model comprising: Adjust the training set image to 320*320; Build a lightweight target detection network ThunderNet as a deep learning model; The deep learning model is trained according to the adjusted training set images to generate a power grid anomaly detection model. 5. The intelligent maintenance-free power grid monitoring method based on cloud-edge collaborative deep learning as claimed in claim 4, wherein the method further comprises: Cloud computing center builds and trains abnormal traffic monitoring model; The cloud computing center sends the trained abnormal traffic monitoring model to the edge computing nodes; The edge computing node monitors the traffic at the edge according to the abnormal traffic monitoring model, and sends an abnormal traffic signal to the cloud computing center if abnormal traffic is found at the edge; When the cloud computing center receives the abnormal traffic signal, it will block the abnormal traffic. 6. An intelligent maintenance-free power grid monitoring system based on cloud-edge collaborative deep learning, characterized in that it includes sensing equipment, edge computing nodes and cloud computing centers; The sensing device includes: The image acquisition component is used to collect monitoring images; an alarm device, used for receiving alarm information of the edge computing node, and alarming according to the alarm information; The edge computing node includes: The image processing module is used to obtain the monitoring image collected by the sensing device; an image detection module for identifying the monitoring image and determining whether there is a power grid abnormality based on the monitoring image and the power grid abnormality detection model issued by the cloud computing center; The alarm module is used to send alarm information to the sensing device when there is an abnormality in the power grid; a first transmission module, used for transmitting the identified monitoring images to the cloud computing center; The cloud computing center includes: a power grid abnormality detection model training module, which trains a deep learning model according to the stored training set images to generate a power grid abnormality detection model, and trains and updates the power grid abnormality detection model according to the identified monitoring images transmitted by the edge computing nodes; The second transmission module is used for delivering the grid abnormality detection model to the edge computing node. 7. The intelligent maintenance-free power grid monitoring system based on cloud-edge collaborative deep learning as claimed in claim 6, wherein the image acquisition component is connected with a camera that shoots power monitoring video, and is used in the video stream captured by the camera. Obtain surveillance images. 8. The intelligent maintenance-free power grid monitoring system based on cloud-edge collaborative deep learning according to claim 7, wherein the edge computing node further comprises a repetition filtering module; The image processing module is also used for preprocessing the monitoring image; The repeated filtering module is used to perform a deduplication operation on the monitoring images collected within the interval; The said monitoring image is identified based on the power grid abnormality detection model issued by the cloud computing center to determine whether there is a power grid abnormality: according to the power grid abnormality detection model, the preprocessed and deduplicated monitoring images are identified to determine the Check whether there is an abnormality in the power grid in the monitoring image. 9. The intelligent maintenance-free power grid monitoring system based on cloud-edge collaborative deep learning as claimed in claim 8, wherein transmitting the identified monitoring images to the cloud computing center comprises: temporarily storing the identified monitoring images in edge computing The node transmits the temporarily stored monitoring images to the cloud computing center uniformly in the time period when the network bandwidth flow is less than the preset value. 10. The intelligent maintenance-free power grid monitoring system based on cloud-edge collaborative deep learning as claimed in claim 9, wherein, training a deep learning model to generate a power grid abnormality detection model according to the stored training set image comprises: Adjust the training set image to 320*320; Build a lightweight target detection network ThunderNet as a deep learning model; The deep learning model is trained according to the adjusted training set images to generate a power grid anomaly detection model.

Images