Industrial control algorithm self-learning optimization system and method based on cloud edge collaboration

Abstract

The invention relates to the technical field of edge computing, in particular to an industrial control algorithm self-learning optimization system and method based on cloud-edge collaboration, and the system comprises a cloud platform at a cloud side, edge nodes at an edge side, and controlled industrial equipment at an equipment side. One cloud platform is used for receiving and managing a plurality of edge nodes; the cloud platform comprises an Internet of Things communication service module; an algorithm management service module; a distributed task scheduler; an edge node management service module; the edge node on the edge side comprises an industrial communication service module; and an algorithm service container. According to the invention, through a cloud-side cooperation mode, computing power resources of the cloud are utilized, autonomous learning training of a machine learning algorithm for industrial equipment optimization control is realized, and the machine learning algorithm can be automatically updated and deployed to the edge computing equipment, so that the industrial control algorithm can be autonomously optimized along with data accumulation, and the optimization efficiency is improved. And when the production working condition is greatly changed, timely training updating and deployment application of the algorithm can be realized.

Description

Technical Field

[0001] This invention relates to the field of edge computing technology, and in particular to a system and method for self-learning optimization of industrial control algorithms based on cloud-edge collaboration. Background Technology

[0002] In current technologies, machine learning algorithms are increasingly being applied to the optimization control of industrial equipment. The common practice is to train the machine learning algorithm model offline and then deploy the trained model to edge computing devices close to the equipment to achieve intelligent optimization control. However, edge computing devices, considering on-site installation methods, industrial operating environments, and hardware investment costs, typically use embedded minicomputers with limited computing and storage capabilities, making them unsuitable for large-scale machine learning algorithm training. Furthermore, to ensure the stability of the algorithm's optimization control, direct machine learning training on edge devices is not recommended, as complex system fluctuations can negatively impact the stability of the algorithm's control.

[0003] The above approach has drawbacks such as the inability of the machine learning algorithm to be updated in a timely manner, preventing the algorithm from autonomously optimizing and iterating with the accumulation of historical data. Furthermore, when production conditions change significantly, the algorithm results may show increased errors. In such cases, manual intervention is required to retrain and update the machine learning algorithm before it can be used normally. Summary of the Invention

[0004] This invention addresses the problems of existing technologies by providing a system and method for self-learning optimization of industrial control algorithms based on cloud-edge collaboration. Through cloud-edge collaboration, the computing resources of the cloud are utilized to enable machine learning algorithms for optimizing the control of industrial equipment to learn and train autonomously, and to automatically update and deploy to edge computing devices. This allows the industrial control algorithms to autonomously optimize as data accumulates, and enables timely training, updating, and deployment of the algorithms when production conditions change significantly.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0006] This invention provides a system for self-learning optimization of industrial control algorithms based on cloud-edge collaboration, comprising a cloud platform on the cloud side, edge nodes on the edge side, and controlled industrial equipment on the device side; the cloud platform is used to manage multiple edge nodes;

[0007] The cloud platform on the cloud side is used for training machine learning algorithms, exporting models, and managing container images. The cloud platform includes:

[0008] IoT communication service module: used to subscribe to device data topics in the edge node message bus and store the data in the data warehouse;

[0009] Algorithm management service module: used for supporting a user to create an algorithm based on a PyTorch or TensorFlow framework and storing to an algorithm library;

[0010] Distributed task scheduler: configure algorithm training period and parameters, trigger training task at regular intervals; edge node management service module: manage edge nodes and algorithm container version upgrade through K3S Server;

[0011] The edge node on the edge side cooperates with the cloud platform on the cloud side through K3S Agent, and the edge node on the edge side comprises:

[0012] Industrial communication service module: communicate with controlled equipment through Modbus, OPC, S7 or CoAP protocol, support MQTT protocol to publish / subscribe device data and control instructions;

[0013] Algorithm service container: subscribe to device data topics for sensing decision, output optimized control instructions to control command topics;

[0014] The controlled industrial equipment on the equipment side receives the optimized control instructions issued by the edge node and feeds back running data.

[0015] The cloud platform further comprises:

[0016] Data warehouse: used for storing device data synchronized from the edge node, for algorithm training calling; container image warehouse: used for storing algorithm service container images, supporting dynamic updating to the edge node.

[0017] The edge node comprises an edge node initialization process, and the edge node initialization process is that: an initialization script is executed through SSH connection, and K3S Agent program is downloaded and installed; K3S Agent registers a node to K3S Server, and completes state update of being managed.

[0018] The algorithm management service module comprises an algorithm service dynamic upgrade process, and the algorithm service dynamic upgrade process is that: a model file is exported after training; an algorithm service container image is built and pushed to the container image warehouse; the edge node container is updated through K3S Server, and algorithm self-optimization is realized.

[0019] The application further provides an industrial control algorithm self-learning optimization method based on cloud edge cooperation, which comprises the following steps:

[0020] Step S1, the cloud platform manages all edge nodes, subscribes to edge node device data topics through the Internet of Things communication service module, and stores to the data warehouse;

[0021] Step S2: The user creates a machine learning algorithm through the algorithm management service module, configures the training period and parameters;

[0022] Step S3: The distributed task scheduler triggers the training task at regular intervals, pulls data from the data warehouse, and iteratively trains the algorithm;

[0023] Step S4: After training is completed, the model file is exported, packaged into a container image, and pushed to the image repository;

[0024] Step S5: The edge node management service module triggers algorithm container rolling upgrade through K3S Server, completing dynamic optimization.

[0025] The timing steps of the cloud platform incorporating edge nodes are as follows:

[0026] Step 301: Add the edge node to be incorporated in the edge node management service, and enter the edge node access-related information, including: node code, node name, remote connection host, remote connection port, remote connection user, and password;

[0027] Step 302: Trigger the incorporation action in the edge node management service;

[0028] Step 303: The edge node management service module uses the edge node remote connection information entered in step 301 to establish a remote SSH connection with the edge node, and executes an initialization script on the edge node through the SSH connection channel;

[0029] In step 304, the edge node downloads the offline installation package of the edge node program from the edge node management service according to the offline installation package authorization download address in the initialization script;

[0030] In step 305, the edge node decompresses and installs the edge node program according to the installation command script in the initialization script, which includes the K3S Agent program;

[0031] In step 306, after installing the edge node program, start the K3S Agent through the K3S command line tool;

[0032] Step 307: The K3S Agent program starts and performs the initialization action of the node;

[0033] Step 308: K3S Agent registers with K3S Server;

[0034] Step 309: The edge node management service obtains the current incorporation state of the edge node from K3S Server through regular polling;

[0035] Step 310: The edge node management service updates the management status to "management successful" or "management failed" based on the edge node status reported by the K3S Server.

[0036] The initial deployment sequence of the algorithm management service module is as follows:

[0037] Step 401: Select the control algorithm to be deployed on the edge node from the algorithm library;

[0038] Step 402: Select the managed edge nodes where the algorithm service needs to be deployed;

[0039] Step 403: Orchestrate the algorithm service container according to the algorithm service Kubernetes Deployment template, and specify the edge node for algorithm deployment in the orchestration file through the node selector;

[0040] Step 404: Using the orchestration file output in step 403, trigger the container deployment action by calling the K3S Server API interface;

[0041] Step 405: K3S Server selects the corresponding edge nodes for deployment based on the orchestration file;

[0042] Step 406: K3SServer triggers the deployment of the algorithm service container to the specified edge node;

[0043] Step 407: The K3S Agent pulls the image from the container image repository according to the algorithm service container image version in the orchestration file and creates the algorithm service container;

[0044] Step 408: Create and start the algorithm service container to complete the initial deployment of the edge node algorithm.

[0045] The timing steps of the algorithm's self-learning dynamic optimization are as follows:

[0046] Step 501: The distributed task scheduler of the cloud platform triggers the algorithm training task periodically according to the configured algorithm training cycle.

[0047] Step 502: After the algorithm training task is started, the required device data is pulled from the data warehouse, and iterative training of the algorithm is performed.

[0048] Step 503: After the algorithm training is completed, export the trained algorithm model. PyTorch supports exporting PyTorch native format and ONNX format model files, while TensorFlow supports exporting SavedModel format, HDF5 format, and TensorFlow Lite format model files. Then, use the algorithm service container Dockerfile template to build and package the algorithm service container image.

[0049] Step 504: Push the constructed algorithm service container image to the container image repository for storage;

[0050] Step 505: The cloud platform's distributed task scheduler automatically triggers rolling updates of the deployment algorithm service through the kubectl command of K3SServer based on the container information associated with the algorithm.

[0051] Step 506: Update the image version in the K3S Server algorithm service orchestration file;

[0052] Step 507: The K3S Server triggers the corresponding container to perform a rolling update through the communication interface with the K3S Agent;

[0053] Step 508: The K3S Agent pulls the corresponding version of the algorithm service container image from the container image repository;

[0054] Step 509: Perform rolling updates.

[0055] The rolling update step in step 509 is as follows: Step 510, create a new version algorithm service container;

[0056] Step 511: Start the algorithm service container; once the new version of the algorithm service container is ready, proceed to step 512.

[0057] Step 512: Perform container traffic switching;

[0058] Step 513: Start stopping the old version of the algorithm service container; after finally stopping the old version of the algorithm service container, the autonomous learning and dynamic upgrade of the edge node algorithm is completed.

[0059] The beneficial effects of this invention are:

[0060] This invention is ingeniously designed. By using cloud-edge collaboration, it leverages cloud computing resources to enable machine learning algorithms for optimizing the control of industrial equipment to learn and train autonomously. The algorithms can then be automatically updated and deployed to edge computing devices, allowing the industrial control algorithms to autonomously optimize as data accumulates. When production conditions change significantly, the algorithms can be trained, updated, and deployed in a timely manner. Attached Figure Description

[0061] Figure 1 This is a system framework diagram of a cloud-edge collaborative industrial control algorithm self-learning optimization system according to the present invention.

[0062] Figure 2 This is a schematic diagram illustrating the specific architecture of the cloud platform, edge nodes, and controlled devices of the present invention.

[0063] Figure 3 This is a flowchart of the timing steps for managing edge nodes in the cloud platform according to the present invention.

[0064] Figure 4 This is a flowchart of the timing steps for the initial deployment of the algorithm management service module of the present invention.

[0065] Figure 5 This is a flowchart of the timing steps for the self-learning dynamic optimization algorithm of the present invention. Detailed Implementation

[0066] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to embodiments and accompanying drawings. The content mentioned in the embodiments is not intended to limit the present invention. The present invention will be described in detail below with reference to the accompanying drawings.

[0067] Example 1

[0068] In Embodiment 1 of this application, as Figure 1 As shown Figure 2 As shown, a system for self-learning optimization of industrial control algorithms based on cloud-edge collaboration includes a cloud platform on the cloud side, edge nodes on the edge side, and controlled industrial equipment on the device side; the cloud platform is used to manage multiple edge nodes.

[0069] The cloud platform on the cloud side is used for training machine learning algorithms, exporting models, and managing container images. The cloud platform includes:

[0070] IoT communication service module: used to subscribe to device data topics in the edge node message bus and store the data in the data warehouse;

[0071] Algorithm Management Service Module: This module supports users in creating algorithms based on the PyTorch or TensorFlow framework and storing them in the algorithm library.

[0072] Distributed task scheduler: Configures algorithm training cycles and parameters, and triggers training tasks on a scheduled basis; Edge node management service module: Manages edge nodes and algorithm container version upgrades through K3S Server;

[0073] The edge nodes on the edge side collaborate with the cloud platform on the cloud side through the K3S Agent. The edge nodes on the edge side include:

[0074] Industrial communication service module: Communicates with controlled devices via Modbus, OPC, S7 or CoAP protocols, and supports MQTT protocol for publishing / subscribing to device data and control commands;

[0075] Algorithm service container: Subscribes to device data topics to perform perception and decision-making, and outputs optimized control instructions to the control command topic;

[0076] The controlled industrial equipment on the equipment side receives optimized control commands issued by the edge nodes and feeds back operating data.

[0077] The cloud platform also includes:

[0078] Data warehouse: Used to store device data synchronized from edge nodes for algorithm training; Container image warehouse: Used to store container images of algorithm services, supporting dynamic updates to edge nodes.

[0079] The edge node includes an edge node initialization process, which is as follows: execute the initialization script via SSH connection, download and install the K3S Agent program; the K3S Agent registers the node with the K3S Server and completes the management status update.

[0080] The algorithm management service module includes a dynamic upgrade process for the algorithm service, which is as follows: after training is completed, export the model file; build an algorithm service container image and push it to the container image repository; trigger the edge node container rolling update through K3S Server to realize the algorithm's autonomous optimization.

[0081] In this embodiment, K3s is a fully compatible, lightweight Kubernetes distribution. PyTorch is an open-source deep learning framework developed by Meta, primarily used for machine learning and artificial intelligence research. TensorFlow is an open-source machine learning framework developed by Google, widely used in deep learning, neural network construction, and various AI application development. Apache Doris is a high-performance, real-time analytical database based on an MPP architecture, supporting not only high-concurrency query scenarios but also high-throughput complex analysis scenarios. gRPC is a modern, open-source, high-performance Remote Procedure Call (RPC) framework that can run in any environment. MQTT (Message Queuing Telemetry Transport) is a publish-subscribe messaging protocol based on the ISO standard (ISO / IEC PRF20922). It operates on top of the TCP / IP protocol and is a publish / subscribe messaging protocol designed for remote devices with low hardware performance and poor network conditions.

[0082] likeFigure 1 and Figure 2 As shown, this application embodiment consists of a three-layer structure: a cloud platform on the cloud side, edge nodes on the edge side, and controlled devices on the device side. One cloud platform manages multiple edge nodes simultaneously, and one edge node connects to multiple controlled industrial devices simultaneously. This application embodiment enables the entire process of autonomous learning and dynamic optimization of industrial control machine learning algorithms, which are trained on the cloud side and dynamically updated on the edge side. The cloud platform can be deployed on an enterprise's public or private cloud platform, or it can be directly deployed on the host computer of the industrial system.

[0083] like Figure 2 As shown, the industrial communication service of the edge node communicates with the controlled equipment via industrial protocols such as Modbus, OPC, S7, or CoAP, supporting both uplink and downlink communication. Data reported by the controlled equipment to the industrial communication service is published to the device data topic on the message bus via the MQTT protocol. The algorithm service container running on the edge node subscribes to the device data in the device data topic on the message bus, performs algorithmic perception and decision-making based on the device data, and outputs optimized control commands. These optimized control commands are published to the device control command topic on the message bus via the MQTT protocol. The industrial communication service subscribes to the optimized control command messages in the device control command topic on the message bus and distributes them to the corresponding controlled equipment via industrial protocols, thereby achieving optimized control of the industrial equipment.

[0084] like Figure 2 As shown, the cloud platform's IoT communication service subscribes to device data topics in the edge node message bus and stores the device data in the data warehouse, enabling the sharing of edge node device data with the cloud platform.

[0085] like Figure 2As shown, users create algorithms based on PyTorch or TensorFlow machine learning frameworks and store them in the algorithm library through the algorithm management service. The training cycle and parameters of the algorithm are defined in the distributed task scheduling service. The distributed task scheduler will trigger algorithm training periodically according to the corresponding training cycle. The algorithm obtains the required device data from the data warehouse through a standard data interface, performs iterative training of the machine learning algorithm, and automatically exports the corresponding model file after training. PyTorch supports exporting PyTorch native format and ONNX format model files, while TensorFlow supports exporting SavedModel format, HDF5 format, and TensorFlow Lite format model files. After the algorithm model file is exported, the algorithm file is compiled and packaged to generate an algorithm service container image and pushed to the container image repository. Then, the distributed task scheduler sends an algorithm container version upgrade command to the edge node management service. The edge node management service, based on the edge node associated with the algorithm, triggers the corresponding algorithm service container of the edge node to perform rolling upgrades through the kubectl command of the K3S Server. This achieves autonomous learning and dynamic upgrade of the algorithm.

[0086] This invention is ingeniously designed. By using cloud-edge collaboration, it leverages cloud computing resources to enable machine learning algorithms for optimizing the control of industrial equipment to learn and train autonomously. The algorithms can then be automatically updated and deployed to edge computing devices, allowing the industrial control algorithms to autonomously optimize as data accumulates. When production conditions change significantly, the algorithms can be trained, updated, and deployed in a timely manner.

[0087] Example 2

[0088] In Embodiment 2 of this application, a method for self-learning optimization of industrial control algorithms based on cloud-edge collaboration includes the following steps:

[0089] Step S1: The cloud platform manages all edge nodes, subscribes to edge node device data topics through the IoT communication service module, and stores them in the data warehouse;

[0090] Step S2: Users create machine learning algorithms and configure training cycles and parameters through the algorithm management service module;

[0091] Step S3: The distributed task scheduler periodically triggers training tasks, pulls data from the data warehouse, and iteratively trains the algorithm;

[0092] Step S4: After training is complete, export the model file, package it into a container image, and push it to the image repository;

[0093] Step S5: The edge node management service module triggers a rolling upgrade of the algorithm container through the K3S Server to complete dynamic optimization.

[0094] In the embodiments of this application, such as Figure 3 As shown, the timing steps for the cloud platform to manage edge nodes are as follows:

[0095] Step 301: Add the edge nodes to be managed in the edge node management service and enter the relevant edge node access information, including: node code, node name, remote connection host, remote connection port, remote connection user and password;

[0096] Step 302: Trigger the inclusion action in the edge node management service;

[0097] Step 303: The edge node management service module uses the remote connection information of the edge node entered in step 301 to establish a remote SSH connection with the edge node and executes the initialization script on the edge node through the SSH connection channel.

[0098] In step 304, the edge node downloads the offline installation package of the edge node program through the edge node management service according to the authorized download address of the offline installation package in the initialization script.

[0099] In step 305, the edge node decompresses and installs the edge node program, including the K3S Agent program, according to the installation command script in the initialization script.

[0100] In step 306, after installing the edge node program, start the K3S Agent using the K3S command-line tool;

[0101] Step 307: The K3S Agent program starts and executes the node initialization actions;

[0102] Step 308: The K3S Agent registers the node with the K3S Server;

[0103] Step 309: The edge node management service obtains the status of currently managed edge nodes from the K3S Server through periodic polling.

[0104] Step 310: The edge node management service updates the management status to "management successful" or "management failed" based on the edge node status reported by the K3S Server.

[0105] In the embodiments of this application, such as Figure 4 As shown, the initial deployment sequence of the algorithm management service module is as follows:

[0106] Step 401: Select the control algorithm to be deployed on the edge node from the algorithm library;

[0107] Step 402: Select the managed edge nodes where the algorithm service needs to be deployed;

[0108] Step 403: Orchestrate the algorithm service container according to the algorithm service Kubernetes Deployment template, and specify the edge node for algorithm deployment in the orchestration file through the node selector;

[0109] Step 404: Using the orchestration file output in step 403, trigger the container deployment action by calling the K3S Server API interface;

[0110] Step 405: K3S Server selects the corresponding edge nodes for deployment based on the orchestration file;

[0111] Step 406: K3SServer triggers the deployment of the algorithm service container to the specified edge node;

[0112] Step 407: The K3S Agent pulls the image from the container image repository according to the algorithm service container image version in the orchestration file and creates the algorithm service container;

[0113] Step 408: Create and start the algorithm service container to complete the initial deployment of the edge node algorithm.

[0114] In the embodiments of this application, such as Figure 5 As shown, the timing steps of the algorithm's self-learning dynamic optimization are as follows:

[0115] Step 501: The distributed task scheduler of the cloud platform triggers the algorithm training task periodically according to the configured algorithm training cycle.

[0116] Step 502: After the algorithm training task is started, the required device data is pulled from the data warehouse, and iterative training of the algorithm is performed.

[0117] Step 503: After the algorithm training is completed, export the trained algorithm model. PyTorch supports exporting PyTorch native format and ONNX format model files, while TensorFlow supports exporting SavedModel format, HDF5 format, and TensorFlow Lite format model files. Then, use the algorithm service container Dockerfile template to build and package the algorithm service container image.

[0118] Step 504: Push the constructed algorithm service container image to the container image repository for storage;

[0119] Step 505: The cloud platform's distributed task scheduler automatically triggers rolling updates of the deployment algorithm service through the kubectl command of K3SServer based on the container information associated with the algorithm.

[0120] Step 506: Update the image version in the K3S Server algorithm service orchestration file;

[0121] Step 507: The K3S Server triggers the corresponding container to perform a rolling update through the communication interface with the K3S Agent;

[0122] Step 508: The K3S Agent pulls the corresponding version of the algorithm service container image from the container image repository;

[0123] Step 509: Perform rolling updates.

[0124] The rolling update step in step 509 is as follows: Step 510, create a new version algorithm service container;

[0125] Step 511: Start the algorithm service container; once the new version of the algorithm service container is ready, proceed to step 512.

[0126] Step 512: Perform container traffic switching;

[0127] Step 513: Start stopping the old version of the algorithm service container; after finally stopping the old version of the algorithm service container, the autonomous learning and dynamic upgrade of the edge node algorithm is completed.

[0128] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and modifications made to the above embodiments based on the present invention without departing from the scope of the present invention are within the scope of the present invention.

Claims

1. A cloud-edge collaboration based industrial control algorithm self-learning optimization system, characterized in that, The cloud platform on the cloud side, the edge node on the edge side, and the controlled industrial equipment on the device side are included. The cloud platform on the cloud side is used for training, model exporting, and container image management of machine learning algorithms, and the cloud platform includes: The IoT communication service module is used for subscribing to the device data topic in the edge node message bus and storing data to the data warehouse. The algorithm management service module is used for supporting users to create algorithms based on the PyTorch or TensorFlow framework and store them to the algorithm library. The distributed task scheduler configures the algorithm training period and parameters and triggers the training task at a fixed time. The edge node management service module manages the edge node and algorithm container version upgrade through the K3S Server. The edge node on the edge side cooperates with the cloud platform on the cloud side through the K3S Agent, and the edge node on the edge side includes: The industrial communication service module communicates with the controlled equipment through the Modbus, OPC, S7, or CoAP protocol, supports the MQTT protocol to publish / subscribe device data and control instructions, and subscribes to the device data topic for sensing decision-making. The algorithm service container outputs the optimized control instruction to the control command topic.

2. The cloud-edge collaboration based industrial control algorithm self-learning optimization system according to claim 1, characterized in that, The controlled industrial equipment on the device side receives the optimized control instruction issued by the edge node and feeds back the running data. The cloud platform further includes:

3. The cloud-edge collaboration based industrial control algorithm self-learning optimization system according to claim 1, characterized in that, The data warehouse is used for storing the device data synchronized from the edge node for algorithm training calling.

4. The cloud-edge collaboration based industrial control algorithm self-learning optimization system according to claim 2, characterized in that, The edge node includes an edge node initialization process, which is to execute an initialization script through SSH connection, download and install the K3S Agent program, register the node to the K3S Server, and complete the management state update.

5. A method for cloud-edge collaborative industrial control algorithm self-learning optimization, characterized in that, The algorithm management service module includes an algorithm service dynamic upgrade process, which is to export the model file after training, build the algorithm service container image, and push it to the container image warehouse. The steps include: Step S1, the cloud platform manages all edge nodes, subscribes to the edge node device data topic through the IoT communication service module, and stores it to the data warehouse. Step S2, the user creates a machine learning algorithm through the algorithm management service module, configures the training period and parameters. Step S3, the distributed task scheduler triggers the training task at a fixed time, pulls the data from the data warehouse, and iteratively trains the algorithm. Step S4, the model file is exported after training, packaged as a container image, and pushed to the image warehouse.

6. The method of claim 5, wherein the method further comprises: Step S5, the edge node management service module triggers the algorithm container rolling upgrade through the K3S Server to complete dynamic optimization. The timing steps of the cloud platform managing the edge node are: Step 301, add the edge node to be managed in the edge node management service, and enter the edge node access related information, including: node code, node name, remote connection host, remote connection port, remote connection user and password; Step 302, trigger the management action in the edge node management service; Step 303, the edge node management service module uses the edge node remote connection information entered in step 301 to establish a remote SSH connection with the edge node, and executes an initialization script on the edge node through the SSH connection channel; In step 304, the edge node downloads the offline installation package of the edge node program from the edge node management service according to the offline installation package authorization download address in the initialization script; In step 305, the edge node decompresses and installs the edge node program according to the installation command script of the initialization script, which includes the K3S Agent program; In step 306, after installing the edge node program, start the K3S Agent through the K3S command line tool; Step 307, the K3S Agent program starts and performs the initialization action of the node; Step 308, the K3S Agent registers with the K3S Server; Step 309, the edge node management service obtains the current managed edge node state from the K3S Server through timed polling; Step 310, the edge node management service updates the managed state to "managed successfully" or "managed failed" according to the edge node state feedback from the K3S Server.

7. The method of claim 5, wherein the method further comprises: The timing steps of the initial deployment of the algorithm management service module are: Step 401, select the control algorithm that needs to be deployed on the edge node from the algorithm library; Step 402, select the managed edge node that needs to deploy the algorithm service; Step 403, according to the algorithm service Kubernetes Deployment template, arrange the algorithm service container, and specify the edge node for algorithm deployment through the node selector in the arrangement file; Step 404, use the arrangement file output in step 403 to trigger the container deployment action by calling the API interface of the K3S Server; Step 405, the K3S Server deploys the corresponding edge node according to the arrangement file; Step 406, the K3S Server triggers the algorithm service container to be deployed to the specified edge node; Step 407, the K3S Agent pulls the image from the container image warehouse according to the algorithm service container image version in the arrangement file, and creates the algorithm service container; Step 408, create and start the algorithm service container to complete the initial deployment of the edge node algorithm.

8. The method of claim 5, wherein the method further comprises: The timing steps of the algorithm self-learning dynamic optimization are: Step 501, the distributed task scheduler of the cloud platform triggers the algorithm training task at a regular time according to the configured algorithm training period; Step 502, after the algorithm training task is started, the required device data is pulled from the data warehouse to execute the iterative training of the algorithm; Step 503, after the algorithm training is completed, the trained algorithm model is exported, wherein PyTorch supports exporting Pytorch native format and ONNX format model files, TensorFlow supports exporting SavedModel format, HDF5 format and TensorFlow Lite format model files; then an algorithm service container Dockerfile template is used to build a packaged algorithm service container image; Step 504, the built algorithm service container image is pushed to the container image repository for storage; Step 505, the distributed task scheduler of the cloud platform automatically triggers the deployment of algorithm service rolling update according to the container information associated with the deployment of the algorithm through the kubectl command of K3S Server; Step 506, K3S Server updates the image version in the algorithm service orchestration file; Step 507, K3S Server triggers the corresponding container for rolling update through the communication interface with K3S Agent; Step 508, K3S Agent pulls the corresponding version of the algorithm service container image from the container image repository; Step 509, rolling update is performed.

9. The method of claim 5, wherein the method further comprises: The rolling update step in step 509 is as follows: step 510, creating a new version of the algorithm service container; Step 511, starting the algorithm service container; when the new version of the algorithm service container is ready, step 512 is executed; Step 512, container traffic switching is performed; Step 513, starting to stop the old version of the algorithm service container; finally, after stopping the old version of the algorithm service container, the autonomous learning and dynamic upgrading of the edge node algorithm are completed.

Images