CN113297415B - Edge video analysis intelligent service method and system for the power edge side - Google Patents

Abstract

The invention discloses an intelligent service method and system for edge video analysis facing to an electric power edge side, comprising the following steps: acquiring cross classification acquisition video data; performing video intelligent analysis model updating on historical video data and incremental updating video data in cross classification, and synchronously updating a video intelligent analysis model of the edge computing device according to a preset updating model issuing strategy; and monitoring the full life cycle running state of the video intelligent analysis model of the edge computing device according to the feedback information uploaded by the edge computing device of the associated index. According to the method, the full life cycle operation state of the edge computing device model is monitored according to the feedback information uploaded by the edge computing device of the associated index, interaction between the cloud edge video analysis intelligent service method and the edge computing device is achieved, further, on-demand deployment of the model of the edge computing device is achieved, and cloud edge coordination of the algorithm model is achieved.

Description

Edge video analysis intelligent service method and system for power edge side

Technical field

The present invention relates to the technical field of the Internet of Things, and specifically relates to an edge video analysis intelligent service method and system for the power edge side.

Background technique

In recent years, with the advent of the Internet of Everything era and the popularity of wireless networks, the number of devices at the edge of the network and the amount of data they generate have increased dramatically. As a people's livelihood enterprise, in order to ensure the stable operation of electricity, the State Grid Corporation of China has invested a large number of monitoring equipment such as surveillance cameras, sensing equipment, wearable devices, etc. in transmission lines, substations, distribution lines, safety surveillance and other scenarios. It needs to monitor the generated In this case, the centralized processing model represented by cloud computing will not be able to efficiently process the data generated by edge devices and will not be able to meet people's demand for service quality. Its disadvantages are mainly reflected in the following issues:

(1) The power intelligent video analysis algorithm model cannot be adapted to the edge side: The current cloud algorithm model requires a large amount of computing resources to run. It is limited by the insufficient computing resources on the edge side and cannot be run directly on the edge side. There are insufficient real-time performance and insufficient broadband resources. And other issues;

(2) Insufficient management and control of edge layer video analysis models in power scenarios: There is a lack of unified management and control mechanism for edge-side algorithm models, and there is confusion in edge-side algorithm model maintenance, views, and channels. There is an urgent need for unified edge-side model full life cycle management and control. ;

(3) The power cloud-edge collaboration mechanism has not yet been formed: At present, the power artificial intelligence short-term model is mainly deployed in the cloud, and the intelligence improvement on the edge and terminal sides is limited. There are technical barriers between the cloud and the terminal side, and there is an urgent need for edge intelligent services. Research and develop to improve the power cloud-edge-device collaboration mechanism.

Contents of the invention

In view of the problems existing in the above-mentioned existing technologies, the present invention provides an edge video analysis intelligent service method, system, electronic equipment, and storage medium for the power edge side, which can control the video analysis model of the edge computing device and implement cloud computing. Side collaboration. The technical solution is as follows:

The first aspect provides an edge video analysis intelligent service method for the power edge side, including the following steps:

According to the edge computing device of the associated index, the video surveillance data uploaded from the source data collection network and its source equipment are subjected to multi-source classification collection processing and cross-classification integration processing to obtain cross-classification collection video data;

Update the video intelligent analysis model on the historical video data and incremental update video data in the cross classification to obtain the first updated video analysis model, and synchronously update the video intelligent analysis model of the edge computing device according to the preset update model delivery strategy;

Based on the feedback information uploaded by the edge computing device of the associated index, the full life cycle operating status of the video intelligent analysis model of the edge computing device is monitored.

In a possible implementation, synchronously updating the video intelligent analysis model of the edge computing device according to a preset update model delivery strategy includes:

(21) Establish a model update service connection with the edge computing device;

(22) Based on the first parameter of the first updated video analysis model, determine whether the model is suitable for the edge computing device. If so, proceed to step (24). If not, perform model lightweight conversion processing;

(23) Based on the second parameters of the edge computing device, determine whether the edge computing device is adapted to the first updated video analysis model. If yes, proceed to step (24). If not, perform upgrade and optimization of the edge computing device;

(24) According to the updated model push information, the edge computing device deploys, updates and upgrades the video intelligent analysis model.

In a possible implementation manner, establishing a model update service connection between the edge computing device and the edge computing device includes one or a combination of the following methods:

Notify the edge computing device of the associated index to update in a predetermined manner;

Directly update the video intelligent analysis model of the associated indexed edge computing device in a predetermined manner;

According to the model update version request issued by the edge computing device, the real-time latest version of the video intelligent analysis model in the database is retrieved and sent to the edge computing device.

In a possible implementation, the model lightweight conversion process includes pruning, compressing and quantizing the first updated video analysis model, and pruning the first updated video analysis model includes the following steps:

The number of channels to be pruned in each layer is randomly generated based on the first updated video analysis model, and the first-generation second updated video analysis model after channel pruning is obtained;

Based on multiple random selections of the number of channels to be pruned in each layer, multiple first-generation and second-updated video analysis models are obtained;

Obtain the weight parameters of each first-generation second updated video analysis model;

According to the weight parameters of each first-generation and second-updated video analysis model, the fitness value of each first-generation and second-updated video analysis model is obtained, and combined with the intelligent optimization algorithm, the first-generation and second-updated video with the best fitness value is determined. The analysis model is recorded as the second updated video analysis model.

In a possible implementation, the weight parameters of each first-generation second updated video analysis model are obtained, and the trained weight prediction network is used to obtain the prediction results. The training method of the weight prediction network is:

(51) Based on the first updated video analysis model, the number of channels pruned out of each layer is randomly selected to obtain the remaining number of channels in each layer. Based on the remaining number of channels in each layer, the first generation second updated video analysis model is constructed, and after training, the first The second generation updates the weight parameters of the video analysis model;

(52) Repeat the preset step (51) to obtain multiple sets of first training samples composed of the remaining number of channels in each layer and the corresponding weight parameters of the first-generation second updated video analysis model, pre-constructed based on the first training sample pair The neural network model is trained to obtain the weight prediction network.

In one possible implementation, combining the intelligent optimization algorithm to determine the first-generation second updated video analysis model with the optimal fitness value includes: taking the remaining number of channels in each layer of the model as an individual, and corresponding to the obtained The loss function value of the first-generation and second-updated video analysis model determines the fitness function value, and the optimal first-generation and second-updated video analysis model is obtained through iterative search of the genetic algorithm.

In a possible implementation, the step of receiving feedback information uploaded by the edge computing device and monitoring the full life cycle operating status of the video intelligent analysis model of the edge computing device includes:

Call the corresponding model running status monitoring method according to the business logic type to which the edge computing device feedback information belongs;

Predict the model operating status of the edge computing device through the preset business logic operating model based on the business logic type of the feedback information and the current feedback information, and analyze abnormal conditions of the model operating state based on the prediction results and actual operating status;

Monitor according to the model running status monitoring method, obtain monitoring data, and generate running status monitoring record data through summary analysis;

Comparative analysis is performed based on the monitoring data and the model full life cycle operation model of the edge computing device to obtain abnormal conditions of the model operation status.

In the second aspect, an edge video analysis intelligent service system for the power edge side is provided, including:

The multi-source data collection and integration unit is used to perform multi-source classification collection processing and cross-classification integration processing on the video surveillance data uploaded from the source data collection network and its source equipment based on the edge computing device of the associated index, to obtain cross-classification collection video data;

The model update unit is used to update the video intelligent analysis model for historical video data and incrementally updated video data in cross-classification, and to synchronously update the video intelligent analysis model of the edge computing device according to the preset update model delivery strategy;

The model life cycle monitoring unit is used to monitor the full life cycle operating status of the video intelligent analysis model of the edge computing device based on the feedback information uploaded by the edge computing device of the associated index.

In a third aspect, an electronic device is provided, and the electronic device includes:

Memory, used to store executable instructions;

The processor is configured to implement the above-mentioned edge video analysis intelligent service method for the power edge side when running the executable instructions stored in the memory.

A fourth aspect provides a computer-readable storage medium storing executable instructions, which is characterized in that when the executable instructions are executed by a processor, the above-mentioned edge video analysis intelligent service method for the power edge side is implemented.

The edge video analysis intelligent service method and system for the power edge side of the present invention has the following beneficial effects:

1. The present invention monitors the full life cycle operation status of the edge computing device model based on the feedback information uploaded to the edge computing device of the associated index, realizes the interaction between the edge video analysis intelligent service method in the cloud and the edge computing device, and then realizes the monitoring of the edge computing device. Models can be deployed on demand to achieve cloud-edge collaboration of algorithm models.

2. The present invention synchronously updates the video intelligent analysis model of the edge computing device according to the preset update model delivery strategy, determines whether the parameters of the first updated video analysis model and the edge computing device are compatible with each other, and performs the first updated video analysis on When the first parameter of the model is not suitable for the edge computing device, the model is lightweight and converted to improve the intelligent level of video analysis on the edge side.

3. During the lightweight conversion process of the first updated video analysis model, the present invention uses a random generation method to obtain multiple lightweight network models, avoiding the defects of manual design. At the same time, it is based on multiple randomly generated lightweight network models. Model weight parameter prediction and selection based on multiple lightweight network models with weight parameters improve the efficiency of lightweight processing and the video analysis performance of lightweight models.

Description of the drawings

Figure 1 is a flow chart of an edge video analysis intelligent service method for the power edge side according to an embodiment of the present invention;

Figure 2 is a flow chart of pruning the first updated video analysis model according to an embodiment of the present invention;

Figure 3 is a flow chart of a method for monitoring the full life cycle operating status of the video intelligent analysis model of an edge computing device in an embodiment of the present invention;

Figure 4 is a structural block diagram of an edge video analysis intelligent service system for the power edge side according to an embodiment of the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be described in further detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present invention. Those of ordinary skill in the art will not make any All other embodiments obtained under the premise of creative work belong to the scope of protection of the present invention.

Figure 1 is a schematic flowchart of an edge video analysis intelligent service method for the power edge side according to an embodiment of the present disclosure. The method can be executed by an edge video analysis intelligent system for the power edge side. The system can be implemented through software and/or Or implemented in hardware and integrated into electronic equipment. Specifically, referring to Figure 1, the method includes the following steps:

According to the edge computing device of the associated index, the video surveillance data uploaded from the source data collection network and its source equipment are subjected to multi-source classification collection processing and cross-classification integration processing to obtain cross-classification collection video data;

The video intelligent analysis model is updated on the historical video data and incremental update video data in the cross classification to obtain the first updated video analysis model, and the video intelligent analysis model of the edge computing device is synchronously updated according to the preset update model delivery strategy, where The video intelligent analysis model is updated to obtain the first updated video analysis model. An update can be performed according to a predetermined time period, or whether to perform an update can be determined based on the amount of incrementally updated video data;

Based on the feedback information uploaded by the edge computing device of the associated index, the full life cycle operating status of the video intelligent analysis model of the edge computing device is monitored.

Embodiments of the present invention monitor the operating status of the edge computing device model throughout its life cycle based on the feedback information uploaded to the edge computing device associated with the index, thereby realizing the interaction between the cloud's edge video analysis intelligent service method and the edge computing device, and thereby realizing the monitoring of the edge computing device. Models can be deployed on demand to achieve cloud-edge collaboration of algorithm models.

The above-mentioned synchronous update of the video intelligent analysis model of the edge computing device according to the preset update model delivery strategy includes:

(21) Establish a model update service connection with the edge computing device;

(22) Based on the first parameter of the first updated video analysis model, which includes the model parameters of the video intelligent analysis model and the computing resources required for model operation, determine whether the model is suitable for the edge computing device. If so, proceed to step (24), if not, perform model lightweight conversion processing;

(23) Based on the second parameters of the edge computing device, which include operating parameters of the software hosting system of the edge computing device, such as firmware versions, etc., determine whether the edge computing device is adapted to the first updated video analysis model. If so, enter Step (24), if not, upgrade and optimize the edge computing device;

(24) According to the updated model push information, the edge computing device deploys, updates and upgrades the video intelligent analysis model.

The above-mentioned establishment of a model update service connection between the edge computing device and the edge computing device includes one or a combination of the following methods:

Notify the edge computing device of the associated index to update in a predetermined manner;

Directly update the video intelligent analysis model of the associated indexed edge computing device in a predetermined manner;

According to the model update version request issued by the edge computing device, the real-time latest version of the video intelligent analysis model in the database is retrieved and sent to the edge computing device.

The predetermined method includes a variety of ways to push request information and formats and forms of notification information. In this embodiment, the edge computing device intelligent server establishes a model update service connection with the edge computing device. According to the update parameter settings, The edge computing device intelligent server determines to directly issue an update request or directly update the model on the edge computing device or based on the update request information of the edge computing device.

The above-mentioned model lightweight conversion processing of the first updated video analysis model includes pruning, compression and quantization processing of the first updated video analysis model. The pruning of the first updated video analysis model includes the following steps:

The number of channels to be pruned in each layer is randomly generated based on the first updated video analysis model, and the first-generation second updated video analysis model after channel pruning is obtained;

Based on multiple random selections of the number of channels to be pruned in each layer, multiple first-generation and second-updated video analysis models are obtained;

Obtain the weight parameters of each first-generation second updated video analysis model;

According to the weight parameters of each first-generation and second-updated video analysis model, the fitness value of each first-generation and second-updated video analysis model is obtained, and combined with the intelligent optimization algorithm, the first-generation and second-updated video with the best fitness value is determined. The analysis model is recorded as the second updated video analysis model. Specifically, the intelligent optimization algorithm uses a genetic algorithm, and uses the remaining number of channels in each layer of the model as an individual to correspond to the loss function value of the first generation second updated video analysis model. Determine the fitness function value and obtain the optimal first-generation and second-updated video analysis model through iterative search using a genetic algorithm.

Specifically, based on the population size parameter m of the intelligent optimization algorithm, the number of channels for each layer to be pruned m times is determined randomly. Then the population initialization method of the intelligent optimization algorithm is: pruning the first updated video analysis model with m channels After pruning the channels, the method obtains the remaining number of channels in each layer of m first-generation and second-updated video analysis models. Based on the initial population, the fitness function value is calculated and individual updates are performed to iterate the intelligent optimization algorithm.

This embodiment randomly generates the number of channels to be pruned in each layer to obtain the number of channels in each layer after pruning the channels, avoiding the limitations of manual design. The number of pruned channels in each layer is randomly generated, and by comparing multiple first-generation channels after pruning the channels. The second updated video analysis model is compared to select the optimal first-generation second updated video analysis model, which improves the goodness of fit of the second updated video analysis model obtained after model pruning;

By predicting the weight parameters of the first-generation and second-updated video analysis models, it is avoided to train multiple first-generation and second-updating video analysis models through input video data to obtain the first-generation and second-updating output video analysis results. The model weight parameters of the video analysis model avoid the iterative training process of multiple video analysis models and improve the efficiency of model pruning.

The weight parameters of each first-generation second updated video analysis model are obtained as above, and the trained weight prediction network is used to obtain the prediction results. The training method of the weight prediction network is:

(51) Based on the first updated video analysis model, the number of channels pruned out of each layer is randomly selected to obtain the remaining number of channels in each layer. Based on the remaining number of channels in each layer, the first generation second updated video analysis model is constructed, and after training, the first Generate the weight parameters of the second updated video analysis model. This training step is the fine-tuning training of the model after pruning a random number of channels of the first updated video analysis model;

(52) Repeat the preset step (51) to obtain multiple sets of first training samples composed of the remaining number of channels in each layer and the corresponding weight parameters of the first-generation second updated video analysis model, pre-constructed based on the first training sample pair The weight prediction network is obtained by training the neural network model. The pre-built neural network model can adopt a fully connected network structure.

The above-mentioned method receives feedback information uploaded by the edge computing device and monitors the full life cycle operation status of the video intelligent analysis model of the edge computing device, including:

Call the corresponding model running status monitoring method according to the business logic type to which the edge computing device feedback information belongs;

Predict the model operating status of the edge computing device through the preset business logic operating model based on the business logic type of the feedback information and the current feedback information, and analyze abnormal conditions of the model operating state based on the prediction results and actual operating status;

Monitor according to the model running status monitoring method, obtain monitoring data, and generate running status monitoring record data through summary analysis;

Comparative analysis is performed based on the monitoring data and the model full life cycle operation model of the edge computing device to obtain abnormal conditions of the model operation status.

Among them, business logic types include uploading, downloading, deployment, updating, version management, etc. of video intelligent analysis models. Different business logic types correspond to different running status monitoring methods. For example, in the model update business, a kind of the entire business logic The implementation method generally includes: the edge computing device requests to establish an update service connection with the edge computing device intelligent server, and at the same time sends the edge computing device's local video intelligent analysis model version serial number, and the edge computing device intelligent server based on the edge computing device's local video intelligent analysis model version The serial number is compared with the latest current version serial number in the database to determine whether an update is required. When an update is required, an updateable prompt message is sent to the edge computing device. When the edge computing device determines to update, video intelligent analysis is provided to the edge computing device. Based on the business logic of the download business of the updated version of the model, when receiving feedback information from the edge computing device, the corresponding business logic operation model can be called according to the feedback information to judge the operating status of the edge computing device and predict the next edge. Calculate the operating status of the device and analyze whether the operating status is abnormal based on the predicted operating status and the actual operating status. For example, the edge computing device currently sends the local video intelligent analysis model version number of the edge computing device. Next, it is predicted that the edge computing device may Make a download request for an updated version; at the same time, the model operating status monitoring method corresponding to the business logic can be determined based on the feedback information, and the corresponding model operating parameters, such as the response time during interaction, can be monitored based on the operating status monitoring data and preset video intelligent analysis Compare and analyze the corresponding operating parameter settings in the model's full life cycle operation model, and analyze whether the operating status is abnormal.

This embodiment also provides an edge video analysis intelligent service system for the power edge side, including:

The multi-source data collection and integration unit is used to perform multi-source classification collection processing and cross-classification integration processing on the video surveillance data uploaded from the source data collection network and its source equipment based on the edge computing device of the associated index, to obtain cross-classification collection video data;

The model update unit is used to update the intelligent video analysis model of the historical video data and incrementally updated video data in cross-classification to obtain the first updated video analysis model, and to synchronously update the edge computing device according to the preset update model delivery strategy. Video intelligent analysis model;

The model life cycle monitoring unit is used to monitor the full life cycle operating status of the video intelligent analysis model of the edge computing device based on the feedback information uploaded by the edge computing device of the associated index.

For specific limitations on the edge video analysis intelligent service system for the power edge side, please refer to the limitations on the edge video analysis intelligent service method for the power edge side mentioned above, which will not be described again here. Each unit in the above-mentioned edge video analysis intelligent service system for the power edge side can be implemented in whole or in part through software, hardware, and combinations thereof. Each of the above units may be embedded in or independent of the processor of the computer device in the form of hardware, or may be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above units.

This embodiment also provides an electronic device, which includes:

Memory, used to store executable instructions;

The processor provides computing and control capabilities, and is used to implement the above-mentioned edge video analysis intelligent service method for the power edge side when running executable instructions stored in the memory.

The memory of the electronic device includes non-volatile storage media and internal memory. The non-volatile storage medium stores an operating system, electronic programs and a database. The database is used to store video data uploaded by the edge computing device, etc. This internal memory provides an environment for the execution of operating systems and computer programs in non-volatile storage media. The electronic device is provided with a network interface for communicating with an external terminal through a network connection. When the computer program is executed by the processor, the edge video analysis intelligent service method for the power edge side is implemented.

This embodiment also provides a computer-readable storage medium that stores executable instructions. When the executable instructions are executed by a processor, the above-mentioned edge video analysis intelligent service method for the power edge side is implemented. The computer-readable storage medium stores executable instructions. The media can be read-only memory (ROM), random access memory (RAM), compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, optical data storage node, etc. .

The present invention is not limited to the above-mentioned specific embodiments. Various modifications made by those of ordinary skill in the art based on the above-mentioned concepts without creative efforts fall within the protection scope of the present invention.

Claims (9)

1. The edge video analysis intelligent service method for the power edge side is characterized by including the following steps: According to the edge computing device of the associated index, the video surveillance data uploaded from the source data collection network and its source equipment are subjected to multi-source classification collection processing and cross-classification integration processing to obtain cross-classification collection video data; Update the video intelligent analysis model on the historical video data and incremental update video data in the cross classification to obtain the first updated video analysis model, and synchronously update the video intelligent analysis model of the edge computing device according to the preset update model delivery strategy; Based on the feedback information uploaded by the edge computing device of the associated index, monitor the full life cycle operating status of the video intelligent analysis model of the edge computing device; The feedback information uploaded by the edge computing device based on the association index monitors the full life cycle operating status of the video intelligent analysis model of the edge computing device, including: Call the corresponding model running status monitoring method according to the business logic type to which the edge computing device feedback information belongs; Predict the model operating status of the edge computing device through the preset business logic operating model based on the business logic type of the feedback information and the current feedback information, and analyze abnormal conditions of the model operating state based on the prediction results and actual operating status; Monitor according to the model running status monitoring method, obtain monitoring data, and generate running status monitoring record data through summary analysis; Comparative analysis is performed based on the monitoring data and the model full life cycle operation model of the edge computing device to obtain abnormal conditions of the model operation status. 2. The edge video analysis intelligent service method for the power edge side according to claim 1, characterized in that the video intelligent analysis model of the edge computing device is synchronously updated according to the preset update model delivery strategy, including: (21) Establish a model update service connection with the edge computing device; (22) Based on the first parameter of the first updated video analysis model, determine whether the model is suitable for the edge computing device. If so, proceed to step (24). If not, perform model lightweight conversion processing; (23) Based on the second parameter of the edge computing device, determine whether the edge computing device is adapted to the first updated video analysis model. If yes, proceed to step (24). If not, perform upgrade and optimization of the edge computing device; (24) According to the updated model push information, the edge computing device deploys, updates and upgrades the video intelligent analysis model. 3. The edge video analysis intelligent service method for the power edge side according to claim 2, characterized in that the establishment of the model update service connection between the edge computing device includes one or a combination of the following methods: Notify the edge computing device of the associated index to update in a predetermined manner; Directly update the video intelligent analysis model of the associated indexed edge computing device in a predetermined manner; According to the model update version request issued by the edge computing device, the real-time latest version of the video intelligent analysis model in the database is retrieved and sent to the edge computing device. 4. The edge video analysis intelligent service method for the power edge side according to claim 2, characterized in that the lightweight conversion process of the model includes pruning, compression and quantization of the first updated video analysis model, The pruning of the first updated video analysis model includes the following steps: The number of channels to be pruned in each layer is randomly generated based on the first updated video analysis model, and the first-generation second updated video analysis model after channel pruning is obtained; Based on multiple random selections of the number of channels to be pruned in each layer, multiple first-generation and second-updated video analysis models are obtained; Obtain the weight parameters of each first-generation second updated video analysis model; According to the weight parameters of each first-generation and second-updated video analysis model, the fitness value of each first-generation and second-updated video analysis model is obtained, and combined with the intelligent optimization algorithm, the first-generation and second-updated video with the best fitness value is determined. The analysis model is recorded as the second updated video analysis model. 5. The edge video analysis intelligent service method oriented to the power edge side according to claim 4, characterized in that, in obtaining the weight parameters of each first-generation second updated video analysis model, the weight parameters after training are used The prediction network obtains the prediction results, and the training method of the weight prediction network is: (51) Based on the first updated video analysis model, randomly select the number of channels to be pruned in each layer to obtain the remaining number of channels in each layer. Based on the remaining number of channels in each layer, the first generation second updated video analysis model is constructed, and after training, the first The second generation updates the weight parameters of the video analysis model; (52) Repeat the preset step (51) to obtain multiple sets of first training samples consisting of the remaining number of channels in each layer and the corresponding weight parameters of the first-generation second updated video analysis model, pre-constructed based on the first training sample pair The neural network model is trained to obtain the weight prediction network. 6. The edge video analysis intelligent service method for the power edge side according to claim 4, characterized in that the combination of intelligent optimization algorithms to determine the first-generation second updated video analysis model with the optimal fitness value includes: Taking the remaining number of channels in each layer of the model as an individual, the fitness function value is determined based on the loss function value corresponding to the first-generation second-update video analysis model, and the optimal first-generation second-update video is obtained through iterative search using a genetic algorithm. Analytical model. 7. The edge video analysis intelligent service system for the power edge side is characterized by: The multi-source data collection and integration unit is used to perform multi-source classification collection processing and cross-classification integration processing on the video surveillance data uploaded from the source data collection network and its source equipment based on the edge computing device of the associated index, to obtain cross-classification collection video data; The model update unit is used to update the video intelligent analysis model for historical video data and incrementally updated video data in cross-classification, and to synchronously update the video intelligent analysis model of the edge computing device according to the preset update model delivery strategy; A model life cycle monitoring unit, configured to monitor the full life cycle operating status of the video intelligent analysis model of the edge computing device based on the feedback information uploaded by the edge computing device based on the association index; The model life cycle monitoring unit includes: Call the corresponding model running status monitoring method according to the business logic type to which the edge computing device feedback information belongs; Predict the model operating status of the edge computing device through the preset business logic operating model based on the business logic type of the feedback information and the current feedback information, and analyze abnormal conditions of the model operating state based on the prediction results and actual operating status; Monitor according to the model running status monitoring method, obtain monitoring data, and generate running status monitoring record data through summary analysis; Comparative analysis is performed based on the monitoring data and the model full life cycle operation model of the edge computing device to obtain abnormal conditions of the model operation status. 8. An electronic device, characterized in that the electronic device includes: Memory, used to store executable instructions; A processor, configured to implement the edge video analysis intelligent service method for the power edge side described in any one of claims 1 to 6 when running executable instructions stored in the memory. 9. A computer-readable storage medium storing executable instructions, characterized in that when the executable instructions are executed by a processor, the edge video analysis for the power edge side according to any one of claims 1 to 6 is implemented. Smart service approach.