Data modeling method, edge node and video edge network system

Abstract

The present application discloses a data modeling method, an edge node, and a video edge network system, which relate to the field of edge intelligence technology. The data modeling method is applied to a first edge node, and includes: determining model loading parameters according to the application requirements of the created application; sending a model loading request to at least one second edge node participating in the application modeling; loading a first target model according to the model loading parameters; sending a modeling job information request to at least one second edge node according to the modeling job request; performing joint modeling based on the acquired first data to obtain a local model; the model loading parameters are determined by the first data and the first synchronization information of the at least one second edge node. The above scheme can improve the accuracy of single-node training, and at the same time enable more intelligent analysis tasks to have higher accuracy.

Description

Technical Field

[0001] The present application belongs to the field of edge intelligence technology, and in particular relates to a data modeling method, an edge node, and a video edge network system. Background Art

[0002] The continuous development of the Internet of Things, mobile computing, and big video services has put pressure on network bandwidth. As more and more devices connect to the internet and generate video data, cloud computing cannot meet the low-latency service requirements and is prone to network congestion and service quality degradation. Consequently, the trend is to move computing power and intelligent analysis to edge nodes. Intelligent video artificial intelligence (AI) algorithms, such as facial recognition, chef hat / mask detection accuracy, vehicle type recognition, fire smoke detection, and plate recognition, have been widely used in commercial, financial, smart buildings, smart campuses, smart business halls, and safe cities.

[0003] Operators are vigorously developing edge clouds at the municipal level. In some industrial scenarios, the collection, storage, and AI processing units are moved from the cloud to the field-level edge boxes. AI inference is performed on the field-level end side, and video edge intelligence is gradually transferred on demand, greatly improving computing efficiency, reducing business latency, achieving data non-exit, and alleviating network bandwidth pressure.

[0004] While the shift in computing power and intelligent analysis capabilities to the edge can indeed reduce data generation and improve efficiency, it also presents challenges in data aggregation. Cloud-based AI intelligent analysis can improve algorithm accuracy by aggregating large amounts of data. However, as computing power shifts to the edge, data becomes increasingly fragmented. Data from individual edge computing points is becoming small-scale and fragmented, leading to a growing phenomenon of data silos. Ensuring algorithm accuracy while reducing latency and improving efficiency is a critical issue that needs to be addressed. Summary of the Invention

[0005] The embodiments of the present application provide a data modeling method, edge node, and video edge network system, which can solve the problem that as computing power points sink, data becomes increasingly dispersed and the accuracy of AI algorithms cannot be guaranteed.

[0006] In order to solve the above technical problems, an embodiment of the present application provides a data modeling method, which is applied to a first edge node, including:

[0007] Determine model loading parameters based on the application requirements of the created application;

[0008] Sending a model loading request to at least one second edge node participating in application modeling, wherein the model loading request carries a model loading parameter;

[0009] Loading a first target model according to the model loading parameters;

[0010] Sending a modeling job information request to the at least one second edge node according to the modeling job request;

[0011] Perform joint modeling based on the acquired first data to acquire a local model;

[0012] Among them, the model loading parameters are determined by the first data and the first synchronization information of the at least one second edge node. The first synchronization information is obtained by the second edge node based on the saved second data and sent to the first edge node. The first synchronization information includes: a data dictionary corresponding to the second data.

[0013] Optionally, after sending the model loading request to at least one second edge node participating in application modeling, the method further includes:

[0014] A model loading response is received from the at least one second edge node, where the model loading response is sent from the second edge node to the first edge node after the second edge node loads the first target model.

[0015] Optionally, after sending the modeling job information request to the at least one second edge node according to the modeling job request, the method further includes:

[0016] Receive a job information request response sent by at least one second edge node.

[0017] Optionally, before determining the model loading parameters according to the application requirements of the created application, the method further includes:

[0018] Activate the node;

[0019] Add an organization corresponding to the at least one second edge node.

[0020] Optionally, the performing node activation includes:

[0021] Configuring the organization information and public network permissions of the first edge node according to the node activation request;

[0022] The organization information includes: organization identification, IP and port of distributed algorithm core engine.

[0023] Optionally, adding an organization corresponding to the at least one second edge node includes:

[0024] Obtaining an organization addition work order for the at least one second edge node;

[0025] Saving the organization information corresponding to the at least one second edge node;

[0026] sending a mechanism addition confirmation request to the at least one second edge node;

[0027] receiving an organization addition confirmation response sent by the at least one second edge node;

[0028] Confirm that the organization has completed adding the work order.

[0029] Optionally, after performing joint modeling based on the acquired first data to acquire a local model, the method further includes:

[0030] Selecting a second target model from the local models obtained through joint modeling according to the service identifier in the inference request;

[0031] Performing calculation using the second target model based on the local data corresponding to the feature data in the inference request to obtain a first calculation result;

[0032] Sending the feature data and model information corresponding to the second target model to the at least one second edge node, and receiving a second calculation result fed back by the at least one second edge node;

[0033] Determining an inference result based on the first calculation result and the second calculation result;

[0034] The characteristic data includes parameters involved in model calculation, and the model information corresponding to the second target model includes parameters used by the second target model.

[0035] This embodiment of the present application further provides a data modeling method, applied to a second edge node, comprising:

[0036] Receive a model loading request sent by the first edge node, where the model loading request carries model loading parameters;

[0037] Loading a first target model according to the model loading request;

[0038] receiving a modeling job information request sent by the first edge node;

[0039] performing joint modeling based on the second data according to the job information request to obtain a local model;

[0040] Among them, the model loading parameters are determined by the first edge node based on the first data and the first synchronization information of at least one second edge node. The first synchronization information is obtained by the second edge node based on the saved second data and sent to the first edge node. The first synchronization information includes: a data dictionary corresponding to the second data.

[0041] Optionally, before receiving the model loading request sent by the first edge node, wherein the model loading request carries the model loading parameters, the method further includes:

[0042] Activate the node;

[0043] Add an organization corresponding to the first edge node.

[0044] Optionally, the performing node activation includes:

[0045] Configuring the organization information and public network permissions of the second edge node according to the node activation request;

[0046] The organization information includes: organization identification, IP and port of distributed algorithm core engine.

[0047] Optionally, the adding of an organization corresponding to the first edge node includes:

[0048] Obtaining an organization addition work order for the first edge node;

[0049] Saving the organization information corresponding to the first edge node;

[0050] Sending a mechanism addition confirmation request to the first edge node;

[0051] receiving an organization addition confirmation response sent by the first edge node;

[0052] Confirm that the organization has completed adding the work order.

[0053] Optionally, after performing joint modeling based on the second data according to the job information request and obtaining a local model, the method further includes:

[0054] Receiving feature data sent by the first edge node and model information corresponding to the second target model;

[0055] performing calculations based on the local data corresponding to the characteristic data using a third target model determined by the model information to obtain a second calculation result;

[0056] Sending the second calculation result to the first edge node;

[0057] The characteristic data includes parameters involved in model calculation, and the model information corresponding to the second target model includes parameters used by the second target model.

[0058] An embodiment of the present application further provides an edge node, wherein the edge node is a first edge node, including:

[0059] A first determining module, configured to determine model loading parameters according to application requirements of the created application;

[0060] A first sending module is configured to send a model loading request to at least one second edge node participating in application modeling, wherein the model loading request carries a model loading parameter;

[0061] A first loading module, configured to load a first target model according to the model loading parameters;

[0062] A second sending module, configured to send a modeling job information request to the at least one second edge node according to the modeling job request;

[0063] A first modeling module is used to perform joint modeling based on the acquired first data to obtain a local model;

[0064] Among them, the model loading parameters are determined by the first data and the first synchronization information of the at least one second edge node. The first synchronization information is obtained by the second edge node based on the saved second data and sent to the first edge node. The first synchronization information includes: a data dictionary corresponding to the second data.

[0065] An embodiment of the present application further provides an edge node, wherein the edge node is a first edge node, including a transceiver and a processor;

[0066] The processor is configured to: determine model loading parameters according to application requirements of the created application;

[0067] The transceiver is configured to: send a model loading request to at least one second edge node participating in application modeling, wherein the model loading request carries a model loading parameter;

[0068] The processor is configured to: load a first target model according to the model loading parameters;

[0069] The transceiver is configured to: send a modeling job information request to the at least one second edge node according to the modeling job request;

[0070] The processor is configured to: perform joint modeling based on the acquired first data to acquire a local model;

[0071] Among them, the model loading parameters are determined by the first data and the first synchronization information of the at least one second edge node. The first synchronization information is obtained by the second edge node based on the saved second data and sent to the first edge node. The first synchronization information includes: a data dictionary corresponding to the second data.

[0072] An embodiment of the present application further provides an edge node, which is a second edge node, including:

[0073] A first receiving module is configured to receive a model loading request sent by a first edge node, wherein the model loading request carries model loading parameters;

[0074] A second loading module, configured to load the first target model according to the model loading request;

[0075] A second receiving module, configured to receive a modeling job information request sent by the first edge node;

[0076] A second modeling module is used to perform joint modeling based on the second data to obtain a local model;

[0077] Among them, the model loading parameters are determined by the first edge node based on the first data and the first synchronization information of at least one second edge node. The first synchronization information is obtained by the second edge node based on the saved second data and sent to the first edge node. The first synchronization information includes: a data dictionary corresponding to the second data.

[0078] An embodiment of the present application further provides an edge node, wherein the edge node is a second edge node and includes a transceiver and a processor;

[0079] The transceiver is configured to: receive a model loading request sent by a first edge node, wherein the model loading request carries a model loading parameter;

[0080] The processor is configured to: load a first target model according to the model loading request;

[0081] The transceiver is configured to: receive a modeling job information request sent by the first edge node;

[0082] The processor is configured to: perform joint modeling based on the second data to obtain a local model;

[0083] Among them, the model loading parameters are determined by the first edge node based on the first data and the first synchronization information of at least one second edge node. The first synchronization information is obtained by the second edge node based on the saved second data and sent to the first edge node. The first synchronization information includes: a data dictionary corresponding to the second data.

[0084] An embodiment of the present application also provides a video edge network system, comprising the above-mentioned first edge node and at least one of the above-mentioned second edge nodes.

[0085] An embodiment of the present application also provides an edge node, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned data modeling method when executing the program.

[0086] An embodiment of the present application also provides a readable storage medium having a computer program stored thereon, which implements the steps in the above-mentioned data modeling method when executed by a processor.

[0087] The beneficial effects of this application are:

[0088] The above scheme determines the model loading parameters based on the first data and the first synchronization information of at least one second edge node, loads the model, and then performs joint modeling through the obtained first data to obtain a local model, thereby realizing horizontal federated modeling of video edge nodes, thereby improving the accuracy of single-node training and enabling more intelligent analysis tasks to have higher accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Figure 1 is a flow chart of a data modeling method applied to a first edge node according to an embodiment of the present application;

[0090] Figure 2 This is the system architecture diagram of the edge intelligence engine applied to the video edge node;

[0091] Figure 3 This is a network diagram;

[0092] Figure 4 It is a business process diagram;

[0093] Figure 5 It is a flowchart of node activation;

[0094] Figure 6 It is a flowchart created by the organization;

[0095] Figure 7 It is a flowchart of the application creation process;

[0096] Figure 8 It is a flowchart of data management;

[0097] Figure 9 It is a flowchart of model management;

[0098] Figure 10 It is a flowchart of horizontal modeling training;

[0099] Figure 11 This is a flowchart of online reasoning;

[0100] Figure 12 This is one of the module schematics of the edge node in an embodiment of the present application;

[0101] Figure 13 This is a structural diagram of an edge node in an embodiment of the present application;

[0102] Figure 14is a flow chart of a data modeling method applied to a second edge node according to an embodiment of the present application;

[0103] Figure 15 This is the second module schematic diagram of the edge node in the embodiment of the present application. DETAILED DESCRIPTION

[0104] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0105] The terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of this application can be implemented in an order other than that illustrated or described herein, and that the objects distinguished by "first," "second," and the like are generally of the same type, and do not limit the number of objects; for example, the first object can be one or more. In addition, the term "and / or" in the specification and claims refers to at least one of the connected objects, and the character " / " generally indicates that the objects connected are in an "or" relationship.

[0106] It is worth noting that the technology described in the embodiments of the present application is not limited to the Long Term Evolution (LTE) / LTE-Advanced (LTE-A) system, but can also be used in other wireless communication systems, such as Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Frequency Division Multiple Access (FDMA), Orthogonal Frequency Division Multiple Access (OFDMA), Single-carrier Frequency Division Multiple Access (SC-FDMA) and other systems. The terms "system" and "network" in the embodiments of the present application are often used interchangeably, and the described technology can be used for the systems and radio technologies mentioned above as well as for other systems and radio technologies. The following description describes a New Radio (NR) system for example purposes, and NR terminology is used in most of the following description, but these technologies can also be applied to applications other than NR system applications, such as 6th Generation (6G) communication systems.

[0107] The data modeling method, edge node, and video edge network system provided in the embodiments of the present application are described in detail below with reference to specific embodiments and their application scenarios in conjunction with the accompanying drawings.

[0108] like Figure 2 As shown, at least one embodiment of the present application provides a data modeling method, applied to a first edge node, comprising:

[0109] Step 101: Determine model loading parameters according to the application requirements of the created application;

[0110] It should be noted that the application here refers to an application created according to an application scenario. For example, the application can be campus security intrusion detection, rodent infestation detection, etc.

[0111] The application requirements refer to the data requirements, algorithm requirements, etc. needed to implement the application.

[0112] Step 102: Send a model loading request to at least one second edge node participating in application modeling;

[0113] It should be noted that the model loading request carries model loading parameters. It should be noted here that the model loading parameters are determined by the first data obtained by the first edge node and the first synchronization information of the at least one second edge node. The first synchronization information is obtained by the second edge node based on the saved second data and sent to the first edge node. The first synchronization information includes: a data dictionary corresponding to the second data. Optionally, the first synchronization information can also include the data storage location corresponding to the second data.

[0114] It should be noted here that the first edge node mentioned in this application refers to the modeling party, that is, the party that actively triggers the modeling task, and the second edge node refers to the party that provides modeling data during the modeling process, which can be regarded as the data party.

[0115] Step 103: Load the first target model according to the model loading parameters;

[0116] It should be noted that since there are multiple models built into the edge node, different application requirements use different models. This step is to determine which one or several models to use based on the model loading parameters, and then load the model.

[0117] Step 104: Send a modeling job information request to the at least one second edge node according to the modeling job request;

[0118] It should be noted that the modeling job information request is used to instruct at least one second edge node to start a local modeling process.

[0119] Step 105: performing joint modeling based on the acquired first data to acquire a local model;

[0120] It should be noted that in the embodiment of the present application, when the parties involved in the modeling are conducting local modeling, the data does not leave the local area, which can ensure the security of the data and realize horizontal federated modeling of video edge nodes, thereby improving the accuracy of single-node training and enabling more intelligent analysis tasks to have higher accuracy.

[0121] It should be noted here that the specific implementation process of the joint modeling process between the modeling party and the data party is: after the modeling party obtains the local model based on the local data (i.e., the first data), the model parameters (the model parameters refer to the relevant information corresponding to the model, for example, the learning rate of the model, etc.) are sent to the data party. After the data party obtains the local model based on its own local data, the model parameters of the local model obtained are sent to the modeling party. The modeling party then iteratively updates the local model according to the model parameters sent by the data party, and the data party iteratively updates the local model according to the model parameters sent by the modeling party. The modeling party and the data party repeat the process of iterative updating of the local model until both parties establish a stable local model, that is, the local model finally obtained is converged.

[0122] It should be further explained that in order to enable the first edge node to control the entire modeling process, after the first edge node sends a model loading request to at least one second edge node participating in application modeling, each second edge node should also send a model loading response to the first edge node after loading the first target model.

[0123] It should also be noted that in order to ensure the synchronization of the modeling process, the first edge node sends a modeling job information request to the at least one second edge node based on the modeling job request. After receiving the modeling job information request, each second edge node should also feedback the job information request response to the first edge node. The second edge node starts the local modeling process after feedback of the job information request response, and the first edge node starts the local modeling process after receiving the feedback of the job information request response from the second edge node, so as to ensure that the modeling party and the data party perform modeling synchronously and ensure the consistency of the joint modeling.

[0124] It should also be noted that before the modeling party and the data party load the model, they need to activate the node first and then add each other's information. Specifically, the process of node activation by the first edge node is: according to the node activation request, configure the organization information and public network permissions of the first edge node;

[0125] The organization information includes: organization identification, IP and port of distributed algorithm core engine.

[0126] It should be noted that the node activation request may be input by a manager through a human-computer interaction interface on a management platform, that is, the manager may select which edge nodes to activate.

[0127] The first edge node then adds the organization corresponding to the at least one second edge node. The specific implementation process is: obtaining the organization addition work order of the at least one second edge node, saving the organization information corresponding to the at least one second edge node, sending an organization addition confirmation request to the at least one second edge node, receiving the organization addition confirmation response sent by the at least one second edge node, and determining that the organization addition work order is completed.

[0128] The process of node activation of the second edge node is similar to that of the first edge node, and the organization information and public network permissions of the second edge node are also configured according to the node activation request.

[0129] After the second edge node is activated, it is necessary to add the organization corresponding to the first edge node. The specific implementation process is: obtain the organization addition work order of the first edge node, save the organization information corresponding to the first edge node, send an organization addition confirmation request to the first edge node, receive the organization addition confirmation response sent by the first edge node, and determine that the organization addition work order is completed.

[0130] It should also be noted that before joint modeling is performed, the modeling party and the data party need to interact with each other to ensure that the joint modeling is performed accurately. Specifically, the first edge node acquires and saves the first data, and acquires the second synchronization information based on the first data. The second synchronization information includes: a data dictionary corresponding to the first data. Optionally, the second synchronization information may further include the data storage location corresponding to the first data; the second synchronization information is sent to the at least one second edge node, and the second edge node saves the second synchronization information. Correspondingly, the second edge node acquires and saves the second data, and acquires the first synchronization information based on the second data. The first synchronization information includes: a data dictionary corresponding to the second data. Optionally, the first synchronization information may further include the data storage location corresponding to the first data; the first synchronization information is sent to the first edge node, and the first edge node saves the first synchronization information.

[0131] It should be noted that after the modeling party and the data party have completed joint modeling and obtained their respective local models, they can use the created model to perform online reasoning, that is, use the created model to perform detection and analysis. The specific implementation process is: the first edge node selects the second target model from the local model obtained by the joint modeling according to the service identifier in the reasoning request, and uses the second target model to perform calculations based on the local data corresponding to the feature data in the reasoning request to obtain a first calculation result; the first edge node sends the feature data and the model information corresponding to the second target model to at least one second edge node. After receiving the feature data and the model information corresponding to the second target model, each second edge node selects the corresponding model according to the model information, and then uses the corresponding model to perform calculations based on the local data corresponding to the feature data to obtain a second calculation result. The second edge node then sends the second calculation result to the first edge node, and the first edge node determines the reasoning result (i.e., analysis result) based on the first calculation result and the second calculation results of each second edge node.

[0132] It should be noted here that the characteristic data mentioned above includes the parameters involved in the model calculation, and the model information corresponding to the second target model includes the parameters used by the second target model; it should be noted here that the parameters are not the data content actually performed, but the type information, attribute information, etc. of the data involved in the calculation.

[0133] The embodiments of the present application are described in detail below.

[0134] This application proposes a method of using distributed edge intelligent engines in the same business application scenario (such as campus security intrusion detection) to horizontally federate the video data collected by different edge intelligent engines through a decentralized networking architecture. In this way, under the condition of small data samples, data can be kept out of the scene, latency can be reduced, and reasoning efficiency can be improved while ensuring the accuracy of the algorithm.

[0135] The video edge node system architecture proposed in this application using the edge intelligence engine is as follows:

[0136] like Figure 2 As shown, each edge node (also called a video edge node) includes: a video stream processing module, a communication processing module, a video / image storage module, a log module and an edge intelligent engine.

[0137] Specifically, the functions of each module are as follows:

[0138] The video stream processing module is responsible for interacting with video front-ends such as cameras, pulling real-time video streams, and performing frame extraction processing.

[0139] The communication processing module is responsible for external communication interface and adapting communication protocol.

[0140] The video / image storage module is responsible for recording and backing up the captured video stream or storing the extracted frame images.

[0141] The log module is responsible for recording event information such as video acquisition and intelligent analysis tasks.

[0142] The edge intelligence engine consists of two parts: a distributed training management engine and a distributed algorithm core engine. Its training data comes from the video / image storage module, and its algorithm reasoning results are output to the external business platform system. The specific network is as follows: Figure 3 As shown, three video edge nodes are interconnected to form a video edge network system, and video edge node 1 and video edge node 2 are connected to service platform 1, and video edge node 3 is connected to service platform 2.

[0143] Considering the distribution of video edge nodes, and considering that video edge nodes from different customers but with the same business scenarios can also participate in federation, a decentralized distributed networking model is adopted. Multiple video edge nodes are distributedly networked to form a video edge intelligent federation network.

[0144] like Figure 4 As shown, the business process of the video edge intelligent federated network involved in this application mainly includes the following links, namely: node activation, organization creation, application creation, data management, model management, horizontal modeling training and online reasoning.

[0145] The following is a detailed description of each link.

[0146] 1. Node Activation

[0147] It should be noted that when Video Edge Node 1 needs to be federated with Video Edge Node 2, the two nodes first need to activate their own edge intelligence engines and configure their own organization information;

[0148] like Figure 5 As shown in the figure, the specific implementation process of this link is:

[0149] Step S501: The distributed training management engine collects node activation requests;

[0150] It should be noted that the node activation request may be input by a manager through a human-computer interaction interface on a management platform, that is, the manager may select which edge nodes to activate.

[0151] Step S502: the distributed training management engine configures the organization information;

[0152] It should be noted that the organization information includes the identifier of the organization corresponding to this node (for example, organization ID), the IP and port of the distributed algorithm core engine;

[0153] Step S503: The distributed training management engine configures an IP port for the distributed algorithm core engine;

[0154] In step S504, the distributed training management engine configures public network permissions, that is, allows external networks to access the node.

[0155] Step S505: Node activation is completed.

[0156] II. Organization Creation

[0157] It should be noted that this link mainly realizes the creation of organizations in activated nodes, that is, to add / synchronize external organization information that requires federation.

[0158] like Figure 6 As shown in the figure, the specific implementation process of this link is:

[0159] Step S601: Video edge node 1 and video edge node 2 reach a business cooperation and provide organization information to each other;

[0160] Step S602: Both parties obtain the task of adding the other party's organization information;

[0161] That is, the distributed training management engine of video edge node 1 obtains the task of adding organization B corresponding to video edge node 2, and the distributed training management engine of video edge node 2 obtains the task of adding organization A corresponding to video edge node 1.

[0162] Step S603: The distributed training management engine of video edge node 1 and the distributed training management engine of video edge node 2 each generate their own organization addition work order;

[0163] Step S604: the distributed training management engine of the video edge node 1 and the distributed training management engine of the video edge node 2 each save the organization information;

[0164] That is, the distributed training management engine of video edge node 1 stores organization B corresponding to video edge node 2, and the distributed training management engine of video edge node 2 stores organization A corresponding to video edge node 1.

[0165] Step S605: the distributed training management engine of the video edge node 1 and the distributed training management engine of the video edge node 2 respectively configure mechanism routing;

[0166] Step S606: The distributed training management engine of the video edge node 1 sends a request for confirmation of adding information of organization B to the distributed training management engine of the video edge node 2;

[0167] Step S607: The distributed training management engine of video edge node 2 sends a confirmation response of adding information of mechanism B to the distributed training management engine of video edge node 1;

[0168] Step S608: The distributed training management engine of the video edge node 2 sends a request for confirmation of adding information of the organization A to the distributed training management engine of the video edge node 1;

[0169] Step S609: The distributed training management engine of video edge node 1 sends a confirmation response of adding information of mechanism A to the distributed training management engine of video edge node 2;

[0170] It should be noted that there is no obvious order relationship between the execution of step S606 and step S608, that is, either step can be executed first.

[0171] In step S610 , the distributed training management engine of video edge node 1 and the distributed training management engine of video edge node 2 respectively confirm that the organization has completed adding the work order.

[0172] 3. Application Creation

[0173] It should be noted that this link mainly realizes the creation of applications based on intelligent analysis application scenarios, such as campus security intrusion detection, rodent infestation detection, etc.

[0174] like Figure 7 As shown in the figure, the specific implementation process of this link is:

[0175] Step S701: The distributed training management engine of the video edge node 1 obtains a request to create an application;

[0176] The application creation request may be input by a manager through a human-computer interaction interface on the management platform, that is, the manager may decide what application to create.

[0177] Step S702: The distributed training management engine of the video edge node 1 saves the application information;

[0178] Step S703: The distributed training management engine of the video edge node 1 synchronizes the application information request to the distributed training management engine of the video edge node 2;

[0179] Step S704: the distributed training management engine of the video edge node 2 saves the application information;

[0180] Step S705: The distributed training management engine of the video edge node 2 returns a synchronization application information response;

[0181] Step S706: The distributed training management engine of the video edge node 1 determines that the application creation is complete;

[0182] In step S707 , the distributed training management engine of the video edge node 1 and the distributed training management engine of the video edge node 2 respectively maintain their own modeling administrator lists.

[0183] It should be noted that the modeling administrator list refers to the login permissions of the administrator, that is, only those with login permissions can create applications.

[0184] 4. Data Management

[0185] It should be noted that this link mainly realizes the loading of video / image data by the organization according to application needs, and synchronizes information such as data dictionary and data storage location to the federal agency. The federal agency also needs to load its own video / image data and store the data information of the other agency.

[0186] It should be noted here that before formal modeling, each node in the video edge intelligent federated network needs to interact with each other for data information. What a node sends to other nodes is not the specific content of the local data, but rather the data dictionary, data storage location and other information synchronized to other nodes.

[0187] It should be noted that the data dictionary referred to in this application refers to the definition and description of data items, data structures, data flows, data storage, processing logic, etc. Its purpose is to provide a detailed description of each element in the data flow diagram. The data dictionary is used for simple modeling projects. In short, the data dictionary is a collection of information that describes the data and is a collection of definitions for all data elements used in the system.

[0188] like Figure 8 As shown in the figure, the specific implementation process of this link is:

[0189] Step S801: The distributed training management engine of video edge node 1 and the distributed training management engine of video edge node 2 respectively obtain their respective data information;

[0190] Step S802, uploading data to the distributed algorithm core engine;

[0191] Step S803: The distributed training management engine of the video edge node 1 and the distributed training management engine of the video edge node 2 respectively save their respective data information;

[0192] Step S804: the distributed training management engine of the video edge node 1 sends a request for synchronizing Party A's data information to the application participating organization to the distributed training management engine of the video edge node 2;

[0193] The request includes synchronization information of Party A's data, including information such as a data dictionary and a data storage location of Party A's data.

[0194] Step S805: The distributed training management engine of the video edge node 2 saves the synchronization information of the party A data;

[0195] Step S806: The distributed training management engine of the video edge node 2 returns a response to synchronize data information of party A;

[0196] Step S807: The distributed training management engine of the video edge node 2 sends a request for synchronizing the B-party data information to the application participating organization to the distributed training management engine of the video edge node 1;

[0197] The request includes synchronization information of Party B's data, including: a data dictionary of Party B's data, a data storage location, and other information.

[0198] Step S808: The distributed training management engine of the video edge node 1 saves the synchronization information of the party B data;

[0199] In step S809 , the distributed training management engine of the video edge node 1 returns a synchronization B-party data information response.

[0200] It should be noted that because modeling has not yet been started in this process, the nodes in the video edge intelligent federation network do not distinguish between the modeling party and the data party. At this time, information interaction is required between each node so that each node can obtain the synchronization information of other nodes.

[0201] 5. Model Management

[0202] It should be noted that this link mainly implements the selection of algorithms and configuration of model parameters based on application requirements, and notifies federal agencies to load the model.

[0203] like Figure 9 As shown in the figure, the specific implementation process of this link is:

[0204] Step S901: The distributed training management engine of the modeling party collects the model loading parameters configured by the administrator on the model loading page;

[0205] It should be noted that the edge node used by the administrator to configure the model loading parameters is the modeling party, and the other related nodes are the data parties.

[0206] Step S902: The distributed training management engine of the modeling party sends a model loading request to the distributed algorithm core engine;

[0207] Step S903: The distributed algorithm core engine of the modeling party synchronizes the model loading request to the distributed algorithm core engine of the data party;

[0208] Step S904: the distributed algorithm core engine of the modeling party and the distributed algorithm core engine of the data party respectively load the model;

[0209] Step S905: The distributed algorithm core engine of the data entity returns a model loading response;

[0210] Step S906: The distributed algorithm core engine of the modeling party returns a model loading response to the distributed training management engine.

[0211] 6. Horizontal Modeling Training

[0212] It should be noted that this link mainly achieves horizontal modeling training among federal agencies.

[0213] like Figure 10 As shown in the figure, the specific implementation process of this link is:

[0214] Step S1001: The distributed training management engine of the modeling party obtains a job creation request;

[0215] The job creation request may be input by a manager through a human-computer interaction interface on the management platform, that is, the manager may decide what kind of job to create.

[0216] Step S1002: The distributed training management engine of the modeling party generates a component diagram and configures components and operation parameters;

[0217] It should be noted that the components mentioned here refer to the algorithm modules and data processing modules used in modeling. The component diagram refers to the component DSL diagram: a directed acyclic graph (DAG) that defines the combination of algorithm components, execution order, and input and output through DSL. It should be noted that the corresponding execution order is formed by dragging, dropping, and connecting components.

[0218] Job parameters define the job runtime parameters, including job parameters (job resource parameters), component parameters, etc.

[0219] Step S1003: The distributed training management engine of the modeling party sends a synchronization job information request to the distributed training management engine of the data party;

[0220] Step S1004: The distributed training management engine of the data entity returns a synchronization job information response;

[0221] Step S1005: the distributed algorithm core engine of the modeling party and the distributed algorithm core engine of the data party read their own modeling data respectively;

[0222] Step S1006: The distributed algorithm core engine of the modeling party and the distributed algorithm core engine of the data party respectively perform feature engineering;

[0223] In step S1007 , the distributed algorithm core engine of the modeling party and the distributed algorithm core engine of the data party respectively implement joint modeling to obtain local models.

[0224] It should be noted that this application only takes the video edge intelligent federated network including two edge nodes as an example, that is, one edge node is the modeling party and the other is the data party. The specific implementation process of joint modeling is: the distributed algorithm core engine of the modeling party uses local data and model parameters obtained from other data parties for joint modeling until a converged local model is obtained, while the distributed algorithm core engine of the data party uses local data and model parameters obtained from the modeling party for joint modeling until a converged local model is obtained.

[0225] It should be noted here that when the video edge intelligent federation network includes more than two edge nodes, one of these edge nodes is the modeling party and the rest are data parties. The specific implementation process of joint modeling is as follows: the distributed algorithm core engine of the modeling party uses local data and model parameters obtained from other data parties for joint modeling until a converged local model is obtained, and the distributed algorithm core engine of each data party uses local data and model parameters obtained from the modeling party and other data parties for joint modeling until a converged local model is obtained.

[0226] In step S1008, the distributed training management engine of the modeling party and the distributed training management engine of the data party monitor their respective job statuses according to the job information.

[0227] 7. Online Reasoning

[0228] It should be noted that this link mainly realizes the joint online reasoning of the two federal agencies, Video Edge Node 1 and Video Edge Node 2.

[0229] like Figure 11 As shown in the figure, the specific implementation process of this link is:

[0230] Step S1101: The business platform sends an inference request to the distributed training management engine of the modeling party;

[0231] Step S1102: The distributed training management engine of the modeling party sends an inference request to the distributed algorithm core engine;

[0232] Step S1103: the distributed algorithm core engine of the modeling party maps the model information according to the service ID and selects a model;

[0233] Step S1104: the distributed algorithm core engine of the modeling party sends remote feature data and model information to the distributed algorithm core engine of the data party;

[0234] Step S1105: the distributed algorithm core engine of the modeling party performs local calculations based on the features in the inference request;

[0235] Step S1106: the distributed algorithm core engine of the data entity selects a model based on the model information;

[0236] Step S1107: The distributed algorithm core engine of the data entity performs local calculations based on the acquired features.

[0237] Step S1108: the distributed algorithm core engine of the data entity returns the calculation result;

[0238] Step S1109: the distributed algorithm core engine of the modeling party merges the calculation results of both parties;

[0239] Step S1110: The distributed algorithm core engine returns the calculation result to the distributed training management engine;

[0240] Step S1111: The distributed training management engine returns the calculation results to the business platform.

[0241] In summary, since the edge intelligent nodes are sunk to the municipal-level edge cloud or even the field-level edge box, the data stored on the edge side cannot be aggregated, and the data sample size of a single video collection point is scarce, which poses a great challenge to the video intelligent modeling work. It is difficult for the algorithm to have high accuracy by relying solely on single-node data collection for modeling. At least one embodiment of the present application introduces the concept of federated learning into the video edge intelligent analysis node, and forms a video edge intelligent federated network through distributed networking of multiple video edge intelligent analysis nodes. Through multiple process links such as creating nodes, creating organizations, creating applications, data management, model loading, horizontal federated training and online reasoning, the horizontal federated modeling and reasoning of the video edge nodes are realized, thereby improving the accuracy of single-node training, while enabling more intelligent analysis tasks to have higher accuracy, thereby facilitating the development of business applications.

[0242] like Figure 12 As shown, at least one embodiment of the present application further provides an edge node 1200, which is a first edge node and includes:

[0243] A first determining module 1201 is configured to determine model loading parameters according to application requirements of the created application;

[0244] A first sending module 1202 is configured to send a model loading request to at least one second edge node participating in application modeling, wherein the model loading request carries a model loading parameter;

[0245] A first loading module 1203 is configured to load a first target model according to the model loading parameters;

[0246] A second sending module 1204 is configured to send a modeling job information request to the at least one second edge node according to the modeling job request;

[0247] A first modeling module 1205 is configured to perform joint modeling based on the acquired first data to acquire a local model;

[0248] Among them, the model loading parameters are determined by the first data and the first synchronization information of the at least one second edge node. The first synchronization information is obtained by the second edge node based on the saved second data and sent to the first edge node. The first synchronization information includes: a data dictionary corresponding to the second data.

[0249] Optionally, after the first sending module 1202 sends the model loading request to at least one second edge node participating in application modeling, the method further includes:

[0250] The third receiving module is configured to receive a model loading response sent by the at least one second edge node, where the model loading response is sent by the second edge node to the first edge node after loading the first target model.

[0251] Optionally, after the second sending module 1204 sends the modeling job information request to the at least one second edge node according to the modeling job request, the method further includes:

[0252] The fourth receiving module is configured to receive a job information request response sent by at least one second edge node.

[0253] Optionally, before the first determining module 1201 determines the model loading parameters according to the application requirements of the created application, the method further includes:

[0254] A first activation module, used for node activation;

[0255] The first adding module is configured to add an organization corresponding to the at least one second edge node.

[0256] Optionally, the first activation module is configured to implement:

[0257] Configuring the organization information and public network permissions of the first edge node according to the node activation request;

[0258] The organization information includes: organization identification, IP and port of distributed algorithm core engine.

[0259] Optionally, the first adding module is used to implement:

[0260] Obtaining an organization addition work order for the at least one second edge node;

[0261] Saving the organization information corresponding to the at least one second edge node;

[0262] sending a mechanism addition confirmation request to the at least one second edge node;

[0263] receiving an organization addition confirmation response sent by the at least one second edge node;

[0264] Confirm that the organization has completed adding the work order.

[0265] Optionally, after the first modeling module 1205 performs joint modeling based on the acquired first data and acquires a local model, the method further includes:

[0266] A selection module, configured to select a second target model from the local models obtained by the joint modeling according to a service identifier in the inference request;

[0267] a first acquisition module, configured to perform calculation using the second target model according to the local data corresponding to the feature data in the inference request to obtain a first calculation result;

[0268] a fifth receiving module, configured to send the feature data and model information corresponding to the second target model to the at least one second edge node, and receive a second calculation result fed back by the at least one second edge node;

[0269] a first determining module, configured to determine an inference result based on the first calculation result and the second calculation result;

[0270] The characteristic data includes parameters involved in model calculation, and the model information corresponding to the second target model includes parameters used by the second target model.

[0271] It should be noted that the relay terminal provided in at least one embodiment of the present application is a relay terminal capable of executing the above-mentioned information transmission method. All embodiments of the above-mentioned information transmission method are applicable to the relay terminal and can achieve the same or similar beneficial effects.

[0272] At least one embodiment of the present application further provides a relay terminal, comprising a transceiver and a processor;

[0273] The processor is configured to: determine model loading parameters according to application requirements of the created application;

[0274] The transceiver is configured to: send a model loading request to at least one second edge node participating in application modeling, wherein the model loading request carries a model loading parameter;

[0275] The processor is configured to: load a first target model according to the model loading parameters;

[0276] The transceiver is configured to: send a modeling job information request to the at least one second edge node according to the modeling job request;

[0277] The processor is configured to: perform joint modeling based on the acquired first data to acquire a local model;

[0278] Among them, the model loading parameters are determined by the first data and the first synchronization information of the at least one second edge node. The first synchronization information is obtained by the second edge node based on the saved second data and sent to the first edge node. The first synchronization information includes: a data dictionary corresponding to the second data.

[0279] Optionally, the processor is configured to implement at least one of the following:

[0280] Optionally, the transceiver is further configured to:

[0281] A model loading response is received from the at least one second edge node, where the model loading response is sent from the second edge node to the first edge node after the second edge node loads the first target model.

[0282] Optionally, the transceiver is further configured to:

[0283] Receive a job information request response sent by at least one second edge node.

[0284] Optionally, the processor is further configured to:

[0285] Activate the node;

[0286] Add an organization corresponding to the at least one second edge node.

[0287] Optionally, the processor is further configured to:

[0288] Configuring the organization information and public network permissions of the first edge node according to the node activation request;

[0289] The organization information includes: organization identification, IP and port of distributed algorithm core engine.

[0290] Optionally, the processor is further configured to:

[0291] Obtaining an organization addition work order for the at least one second edge node;

[0292] Saving the organization information corresponding to the at least one second edge node;

[0293] The transceiver is further configured to: send a mechanism addition confirmation request to the at least one second edge node;

[0294] receiving an organization addition confirmation response sent by the at least one second edge node;

[0295] The processor is further configured to:

[0296] Confirm that the organization has completed adding the work order.

[0297] Optionally, the processor is further configured to:

[0298] Selecting a second target model from the local models obtained through joint modeling according to the service identifier in the inference request;

[0299] Performing calculation using the second target model based on the local data corresponding to the feature data in the inference request to obtain a first calculation result;

[0300] The transceiver is further configured to: send the feature data and model information corresponding to the second target model to the at least one second edge node, and receive a second calculation result fed back by the at least one second edge node;

[0301] The processor is further configured to: determine an inference result based on the first calculation result and the second calculation result;

[0302] The characteristic data includes parameters involved in model calculation, and the model information corresponding to the second target model includes parameters used by the second target model.

[0303] like Figure 13 As shown, an embodiment of the present invention further provides an edge node, which is a first edge node and includes a processor 1300, a transceiver 1310, a memory 1320, and a program stored in the memory 1320 and executable on the processor 1300; wherein the transceiver 1310 is connected to the processor 1300 and the memory 1320 via a bus interface, wherein the processor 1300 is configured to read the program in the memory and execute the following process:

[0304] Determine model loading parameters based on the application requirements of the created application;

[0305] Sending a model loading request to at least one second edge node participating in application modeling through the transceiver 1310 , wherein the model loading request carries a model loading parameter;

[0306] Loading a first target model according to the model loading parameters;

[0307] Sending a modeling job information request to the at least one second edge node through the transceiver 1310 according to the modeling job request;

[0308] Perform joint modeling based on the acquired first data to acquire a local model;

[0309] Among them, the model loading parameters are determined by the first data and the first synchronization information of the at least one second edge node. The first synchronization information is obtained by the second edge node based on the saved second data and sent to the first edge node. The first synchronization information includes: a data dictionary corresponding to the second data.

[0310] The transceiver 1310 is configured to receive and send data under the control of the processor 1300 .

[0311] Among them, Figure 13 In the embodiment, the bus architecture may include any number of interconnected buses and bridges, specifically linking together various circuits of one or more processors represented by processor 1300 and memory represented by memory 1320. The bus architecture may also link together various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. The bus interface provides an interface. The transceiver 1310 may be a plurality of components, i.e., a transmitter and a receiver, providing a unit for communicating with various other devices on a transmission medium, such as a wireless channel, a wired channel, an optical cable, and the like. For different user devices, the user interface 1330 may also be an interface capable of connecting external or internal devices as required, and the connected devices include but are not limited to a keypad, a display, a speaker, a microphone, a joystick, and the like.

[0312] The processor 1300 is responsible for managing the bus architecture and general processing, and the memory 1320 can store data used by the processor 1300 when performing operations.

[0313] Optionally, the processor 1300 may be a CPU (central processing unit), an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array) or a CPLD (Complex Programmable Logic Device), and the processor may also adopt a multi-core architecture.

[0314] The processor calls the computer program stored in the memory to execute any of the methods provided in the embodiments of the present application according to the obtained executable instructions. The processor and the memory can also be arranged physically separately.

[0315] Optionally, the processor 1300 is configured to read a program in a memory and further perform the following process:

[0316] The transceiver 1310 receives a model loading response sent by the at least one second edge node, where the model loading response is sent by the second edge node to the first edge node after loading the first target model.

[0317] The processor 1300 is used to read the program in the memory and further perform the following processes:

[0318] The transceiver 1310 receives a job information request response sent by at least one second edge node.

[0319] The processor 1300 is used to read the program in the memory and further perform the following processes:

[0320] Activate the node;

[0321] Add an organization corresponding to the at least one second edge node.

[0322] The processor 1300 is configured to read a program in a memory and further execute the following processes: configuring the organization information and public network permissions of the first edge node according to the node activation request;

[0323] The organization information includes: organization identification, IP and port of distributed algorithm core engine.

[0324] The processor 1300 is used to read the program in the memory and further perform the following processes:

[0325] Obtaining an organization addition work order for the at least one second edge node;

[0326] Saving the organization information corresponding to the at least one second edge node;

[0327] Sending a mechanism addition confirmation request to the at least one second edge node via the transceiver 1310 ;

[0328] Receiving, through the transceiver 1310, an organization addition confirmation response sent by the at least one second edge node;

[0329] Confirm that the organization has completed adding the work order.

[0330] Optionally, the processor 1300 is configured to read a program in a memory and further perform the following process:

[0331] Selecting a second target model from the locally modeled models according to the service identifier in the inference request;

[0332] Performing calculation using the second target model based on the local data corresponding to the feature data in the inference request to obtain a first calculation result;

[0333] Sending the feature data and the model information corresponding to the second target model to the at least one second edge node through the transceiver 1310, and receiving the second calculation result fed back by the at least one second edge node;

[0334] Determining an inference result based on the first calculation result and the second calculation result;

[0335] The characteristic data includes parameters involved in model calculation, and the model information corresponding to the second target model includes parameters used by the second target model.

[0336] At least one embodiment of the present application also provides an edge node, which is a first edge node, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the various processes in the embodiment of the information transmission method applied to the relay terminal are implemented, and the same technical effect can be achieved. To avoid repetition, they will not be described here.

[0337] At least one embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the program implements the various processes in the data modeling method embodiment applied to the first edge node as described above, and can achieve the same technical effect. To avoid repetition, it is not described here. The computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0338] Corresponding to the implementation on the first edge node side, such as Figure 14 As shown, at least one embodiment of the present application provides a data modeling method, applied to a second edge node, comprising:

[0339] Step 1401: Receive a model loading request sent by a first edge node, wherein the model loading request carries model loading parameters;

[0340] Step 1402: Load the first target model according to the model loading request;

[0341] Step 1403: receiving a modeling job information request sent by the first edge node;

[0342] Step 1404: performing joint modeling based on the second data according to the job information request to obtain a local model;

[0343] Among them, the model loading parameters are determined by the first edge node based on the first data and the first synchronization information of at least one second edge node. The first synchronization information is obtained by the second edge node based on the saved second data and sent to the first edge node. The first synchronization information includes: a data dictionary corresponding to the second data.

[0344] Optionally, before receiving the model loading request sent by the first edge node, wherein the model loading request carries the model loading parameters, the method further includes:

[0345] Activate the node;

[0346] Add an organization corresponding to the first edge node.

[0347] Optionally, the performing node activation includes:

[0348] Configuring the organization information and public network permissions of the second edge node according to the node activation request;

[0349] The organization information includes: organization identification, IP and port of distributed algorithm core engine.

[0350] Optionally, the adding of an organization corresponding to the first edge node includes:

[0351] Obtaining an organization addition work order for the first edge node;

[0352] Saving the organization information corresponding to the first edge node;

[0353] Sending a mechanism addition confirmation request to the first edge node;

[0354] receiving an organization addition confirmation response sent by the first edge node;

[0355] Confirm that the organization has completed adding the work order.

[0356] Optionally, after performing joint modeling based on the second data according to the job information request and obtaining a local model, the method further includes:

[0357] Receiving feature data sent by the first edge node and model information corresponding to the second target model;

[0358] performing calculations based on the local data corresponding to the characteristic data using a third target model determined by the model information to obtain a second calculation result;

[0359] Sending the second calculation result to the first edge node;

[0360] The characteristic data includes parameters involved in model calculation, and the model information corresponding to the second target model includes parameters used by the second target model.

[0361] It should be noted that all descriptions about the network-side devices in the above embodiments are applicable to the embodiments of the information transmission method and can achieve the same technical effects.

[0362] like Figure 15 As shown, at least one embodiment of the present application further provides an edge node 1500, which is a second edge node and includes:

[0363] A first receiving module 1501 is configured to receive a model loading request sent by a first edge node, wherein the model loading request carries model loading parameters;

[0364] The second loading module 1502 is configured to load the first target model according to the model loading request;

[0365] The second receiving module 1503 is configured to receive a modeling job information request sent by the first edge node;

[0366] The second modeling module 1504 is configured to perform joint modeling based on the second data according to the job information request to obtain a local model;

[0367] Among them, the model loading parameters are determined by the first edge node based on the first data and the first synchronization information of at least one second edge node. The first synchronization information is obtained by the second edge node based on the saved second data and sent to the first edge node. The first synchronization information includes: a data dictionary corresponding to the second data.

[0368] Optionally, before the first receiving module 1501 receives the model loading request sent by the first edge node, and the model loading request carries the model loading parameters, the method further includes:

[0369] The second activation module is used to activate the node;

[0370] The second adding module is used to add an organization corresponding to the first edge node.

[0371] Optionally, the second activation module is configured to implement:

[0372] Configuring the organization information and public network permissions of the second edge node according to the node activation request;

[0373] The organization information includes: organization identification, IP and port of distributed algorithm core engine.

[0374] Optionally, the second adding module includes:

[0375] Obtaining an organization addition work order for the first edge node;

[0376] Saving the organization information corresponding to the first edge node;

[0377] Sending a mechanism addition confirmation request to the first edge node;

[0378] receiving an organization addition confirmation response sent by the first edge node;

[0379] Confirm that the organization has completed adding the work order.

[0380] Optionally, the second modeling module 1504 performs joint modeling based on the second data according to the job information request, and after obtaining the local model, further includes:

[0381] a sixth receiving module, configured to receive the feature data sent by the first edge node and model information corresponding to the second target model;

[0382] A second acquisition module is configured to perform calculation based on the local data corresponding to the feature data using a third target model determined by the model information to obtain a second calculation result;

[0383] A third sending module, configured to send the second calculation result to the first edge node;

[0384] The characteristic data includes parameters involved in model calculation, and the model information corresponding to the second target model includes parameters used by the second target model.

[0385] It should be noted that the edge node provided in at least one embodiment of the present application is an edge node capable of executing the above-mentioned data modeling method applied to the second edge node side, then all embodiments of the above-mentioned data modeling method are applicable to the edge node and can achieve the same or similar beneficial effects.

[0386] At least one embodiment of the present application further provides an edge node, wherein the edge node is a second edge node and includes a transceiver and a processor;

[0387] The transceiver is configured to: receive a model loading request sent by a first edge node, wherein the model loading request carries a model loading parameter;

[0388] The processor is configured to: load a first target model according to the model loading request;

[0389] The transceiver is configured to: receive a modeling job information request sent by the first edge node;

[0390] The processor is configured to: perform joint modeling based on the second data according to the job information request to obtain a local model;

[0391] Among them, the model loading parameters are determined by the first edge node based on the first data and the first synchronization information of at least one second edge node. The first synchronization information is obtained by the second edge node based on the saved second data and sent to the first edge node. The first synchronization information includes: a data dictionary corresponding to the second data.

[0392] Optionally, the processor is further configured to:

[0393] Activate the node;

[0394] Add an organization corresponding to the first edge node.

[0395] Optionally, the processor is further configured to:

[0396] Configuring the organization information and public network permissions of the second edge node according to the node activation request;

[0397] The organization information includes: organization identification, IP and port of distributed algorithm core engine.

[0398] Optionally, the processor is further configured to:

[0399] Obtaining an organization addition work order for the first edge node;

[0400] Saving the organization information corresponding to the first edge node;

[0401] The transceiver is further configured to: send an addition confirmation request to the first edge node;

[0402] receiving an organization addition confirmation response sent by the first edge node;

[0403] The processor is further used to: determine whether the organization has completed adding the work order.

[0404] Optionally, the transceiver is further configured to:

[0405] Receiving feature data sent by the first edge node and model information corresponding to the second target model;

[0406] The processor is further configured to:

[0407] performing calculations based on the local data corresponding to the characteristic data using a third target model determined by the model information to obtain a second calculation result;

[0408] The transceiver is further configured to: send the second calculation result to the first edge node;

[0409] The characteristic data includes parameters involved in model calculation, and the model information corresponding to the second target model includes parameters used by the second target model.

[0410] The embodiment of the present invention further provides an edge node, which is a second edge node. Specifically, the structure of the second edge node can be seen in Figure 13 , I will not go into details here.

[0411] Optionally, the processor is configured to read a program in a memory and execute the following process:

[0412] Receive a model loading request sent by the first edge node, where the model loading request carries model loading parameters;

[0413] Loading a first target model according to the model loading request;

[0414] receiving a modeling job information request sent by the first edge node;

[0415] performing joint modeling based on the second data according to the job information request to obtain a local model;

[0416] Among them, the model loading parameters are determined by the first edge node based on the first data and the first synchronization information of at least one second edge node. The first synchronization information is obtained by the second edge node based on the saved second data and sent to the first edge node. The first synchronization information includes: a data dictionary corresponding to the second data.

[0417] Optionally, the processor is configured to read a program in a memory and further perform the following process:

[0418] Activate the node;

[0419] Add an organization corresponding to the first edge node.

[0420] Optionally, the processor is configured to read a program in a memory and execute the following process:

[0421] Configuring the organization information and public network permissions of the second edge node according to the node activation request;

[0422] The organization information includes: organization identification, IP and port of distributed algorithm core engine.

[0423] Optionally, the processor is configured to read a program in a memory and further perform the following process:

[0424] Obtaining an organization addition work order for the first edge node;

[0425] Saving the organization information corresponding to the first edge node;

[0426] Sending a mechanism addition confirmation request to the first edge node via a transceiver;

[0427] receiving, through a transceiver, a mechanism addition confirmation response sent by the first edge node;

[0428] Confirm that the organization has completed adding the work order.

[0429] Optionally, the processor is configured to read a program in a memory and further perform the following process:

[0430] receiving, through a transceiver, feature data sent by the first edge node and model information corresponding to the second target model;

[0431] performing calculations based on the local data corresponding to the characteristic data using a third target model determined by the model information to obtain a second calculation result;

[0432] Sending the second calculation result to the first edge node through the transceiver;

[0433] The characteristic data includes parameters involved in model calculation, and the model information corresponding to the second target model includes parameters used by the second target model.

[0434] At least one embodiment of the present application also provides an edge node, which is a second edge node, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the various processes in the data modeling method embodiment applied to the second edge node side are implemented, and the same technical effect can be achieved. To avoid repetition, they will not be described here.

[0435] At least one embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the various processes in the data modeling method embodiment applied to the second edge node side as described above, and can achieve the same technical effect. To avoid repetition, the details are not described here. The computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0436] At least one embodiment of the present application further provides a video edge network system, including the above-mentioned first edge node and at least one second edge node.

[0437] It should be noted that, in this article, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element. In addition, it should be noted that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in the opposite order according to the functions involved. For example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0438] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0439] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms without departing from the purpose of this application and the scope of protection of the claims, all of which are within the protection of this application.

Claims

1. A data modeling method, applied to a first edge node, characterized in that: include: Determine model loading parameters based on the application requirements of the created application; Sending a model loading request to at least one second edge node participating in application modeling, wherein the model loading request carries a model loading parameter; Loading a first target model according to the model loading parameters; Sending a modeling job information request to the at least one second edge node according to the modeling job request; Perform joint modeling based on the acquired first data to acquire a local model; The model loading parameter is determined by the first data and first synchronization information of the at least one second edge node, the first synchronization information is obtained by the second edge node based on the stored second data and sent to the first edge node, and the first synchronization information includes: a data dictionary corresponding to the second data; Among them, the joint modeling based on the acquired first data includes: after the first edge node obtains the local model according to the first data, the model parameters are sent to the at least one second edge node; after the at least one second edge node obtains the local model according to its own local data, the model parameters of the local model obtained by itself are sent to the first edge node; then the first edge node iteratively updates the local model according to the model parameters sent by the at least one second edge node; the at least one second edge node iteratively updates the local model according to the model parameters sent by the first edge node; the first edge node and the at least one second edge node repeat the process of iteratively updating the local model until both parties establish a stable local model.

2. The method according to claim 1, characterized in that After sending the model loading request to at least one second edge node participating in application modeling, the method further includes: A model loading response is received from the at least one second edge node, where the model loading response is sent by the second edge node to the first edge node after loading the first target model.

3. The method according to claim 1, characterized in that After sending the modeling job information request to the at least one second edge node according to the modeling job request, the method further includes: Receive a job information request response sent by at least one second edge node.

4. The method according to claim 1, wherein Before determining the model loading parameters according to the application requirements of the created application, the method further includes: Activate the node; Add an organization corresponding to the at least one second edge node.

5. The method according to claim 4, characterized in that The node activation includes: Configuring the organization information and public network permissions of the first edge node according to the node activation request; The organization information includes: organization identification, IP and port of distributed algorithm core engine.

6. The method according to claim 4, characterized in that The adding of an organization corresponding to the at least one second edge node includes: Obtaining an organization addition work order for the at least one second edge node; Saving the organization information corresponding to the at least one second edge node; sending a mechanism addition confirmation request to the at least one second edge node; receiving an organization addition confirmation response sent by the at least one second edge node; Confirm that the organization has completed adding the work order.

7. The method according to claim 1, characterized in that After performing joint modeling based on the acquired first data to acquire a local model, the method further includes: Selecting a second target model from the local models obtained through joint modeling according to the service identifier in the inference request; Performing calculation using the second target model based on the local data corresponding to the feature data in the inference request to obtain a first calculation result; Sending the feature data and model information corresponding to the second target model to the at least one second edge node, and receiving a second calculation result fed back by the at least one second edge node; Determining an inference result based on the first calculation result and the second calculation result; The characteristic data includes parameters involved in model calculation, and the model information corresponding to the second target model includes parameters used by the second target model.

8. A data modeling method, applied to a second edge node, characterized in that: include: Receive a model loading request sent by the first edge node, where the model loading request carries model loading parameters; Loading a first target model according to the model loading request; receiving a modeling job information request sent by the first edge node; performing joint modeling based on the second data according to the job information request to obtain a local model; The model loading parameters are determined by the first edge node according to the first data and first synchronization information of at least one second edge node, the first synchronization information is obtained by the second edge node according to the stored second data and sent to the first edge node, and the first synchronization information includes: a data dictionary corresponding to the second data; Among them, the joint modeling based on the second data includes: after the second edge node obtains the local model according to the second data, the model parameters are sent to the first edge node; after the first edge node obtains the local model according to its own local data, the model parameters of the local model obtained by itself are sent to the second edge node; then the second edge node iteratively updates the local model according to the model parameters sent by the first edge node; the first edge node iteratively updates the local model according to the model parameters sent by the second edge node; the second edge node and the first edge node repeat the process of iteratively updating the local model until both parties establish a stable local model.

9. The method according to claim 8, characterized in that Before receiving the model loading request sent by the first edge node, wherein the model loading request carries the model loading parameters, the method further includes: Activate the node; Add an organization corresponding to the first edge node.

10. The method according to claim 9, characterized in that The node activation includes: Configuring the organization information and public network permissions of the second edge node according to the node activation request; The organization information includes: organization identification, IP and port of distributed algorithm core engine.

11. The method according to claim 9, characterized in that Adding an organization corresponding to the first edge node includes: Obtaining an organization addition work order for the first edge node; Saving the organization information corresponding to the first edge node; Sending a mechanism addition confirmation request to the first edge node; receiving an organization addition confirmation response sent by the first edge node; Confirm that the organization has completed adding the work order.

12. The method according to claim 8, characterized in that After performing joint modeling based on the second data according to the job information request and obtaining a local model, the method further includes: Receiving feature data sent by the first edge node and model information corresponding to the second target model; performing calculations based on the local data corresponding to the characteristic data using a third target model determined by the model information to obtain a second calculation result; Sending the second calculation result to the first edge node; The characteristic data includes parameters involved in model calculation, and the model information corresponding to the second target model includes parameters used by the second target model.

13. An edge node, wherein the edge node is a first edge node, characterized in that: include: A first determining module, configured to determine model loading parameters according to application requirements of the created application; A first sending module is configured to send a model loading request to at least one second edge node participating in application modeling, wherein the model loading request carries a model loading parameter; A first loading module, configured to load a first target model according to the model loading parameters; A second sending module, configured to send a modeling job information request to the at least one second edge node according to the modeling job request; A first modeling module is used to perform joint modeling based on the acquired first data to obtain a local model; The model loading parameter is determined by the first data and first synchronization information of the at least one second edge node, the first synchronization information is obtained by the second edge node based on the stored second data and sent to the first edge node, and the first synchronization information includes: a data dictionary corresponding to the second data; Among them, the joint modeling based on the acquired first data includes: after the first edge node obtains the local model according to the first data, the model parameters are sent to the at least one second edge node; after the at least one second edge node obtains the local model according to its own local data, the model parameters of the local model obtained by itself are sent to the first edge node; then the first edge node iteratively updates the local model according to the model parameters sent by the at least one second edge node; the at least one second edge node iteratively updates the local model according to the model parameters sent by the first edge node; the first edge node and the at least one second edge node repeat the process of iteratively updating the local model until both parties establish a stable local model.

14. An edge node, wherein the edge node is a first edge node, characterized in that: including a transceiver and a processor; The processor is configured to: determine model loading parameters according to application requirements of the created application; The transceiver is configured to: send a model loading request to at least one second edge node participating in application modeling, wherein the model loading request carries a model loading parameter; The processor is configured to: load a first target model according to the model loading parameters; The transceiver is configured to: send a modeling job information request to the at least one second edge node according to the modeling job request; The processor is configured to: perform joint modeling based on the acquired first data to acquire a local model; The model loading parameter is determined by the first data and first synchronization information of the at least one second edge node, the first synchronization information is obtained by the second edge node based on the stored second data and sent to the first edge node, and the first synchronization information includes: a data dictionary corresponding to the second data; Among them, the joint modeling based on the acquired first data includes: after the first edge node obtains the local model according to the first data, the model parameters are sent to the at least one second edge node; after the at least one second edge node obtains the local model according to its own local data, the model parameters of the local model obtained by itself are sent to the first edge node; then the first edge node iteratively updates the local model according to the model parameters sent by the at least one second edge node; the at least one second edge node iteratively updates the local model according to the model parameters sent by the first edge node; the first edge node and the at least one second edge node repeat the process of iteratively updating the local model until both parties establish a stable local model.

15. An edge node, wherein the edge node is a second edge node, characterized in that: include: A first receiving module is configured to receive a model loading request sent by a first edge node, wherein the model loading request carries model loading parameters; A second loading module, configured to load the first target model according to the model loading request; A second receiving module, configured to receive a modeling job information request sent by the first edge node; a second modeling module, configured to perform joint modeling based on the second data according to the job information request to obtain a local model; The model loading parameters are determined by the first edge node according to the first data and first synchronization information of at least one second edge node, the first synchronization information is obtained by the second edge node according to the stored second data and sent to the first edge node, and the first synchronization information includes: a data dictionary corresponding to the second data; Among them, the joint modeling based on the second data includes: after the second edge node obtains the local model according to the second data, the model parameters are sent to the first edge node; after the first edge node obtains the local model according to its own local data, the model parameters of the local model obtained by itself are sent to the second edge node; then the second edge node iteratively updates the local model according to the model parameters sent by the first edge node; the first edge node iteratively updates the local model according to the model parameters sent by the second edge node; the second edge node and the first edge node repeat the process of iteratively updating the local model until both parties establish a stable local model.

16. An edge node, wherein the edge node is a second edge node, characterized in that: including a transceiver and a processor; The transceiver is configured to: receive a model loading request sent by a first edge node, wherein the model loading request carries a model loading parameter; The processor is configured to: load a first target model according to the model loading request; The transceiver is configured to: receive a modeling job information request sent by the first edge node; The processor is configured to: perform joint modeling based on the second data according to the job information request to obtain a local model; The model loading parameters are determined by the first edge node according to the first data and first synchronization information of at least one second edge node, the first synchronization information is obtained by the second edge node according to the stored second data and sent to the first edge node, and the first synchronization information includes: a data dictionary corresponding to the second data; Among them, the joint modeling based on the second data includes: after the second edge node obtains the local model according to the second data, the model parameters are sent to the first edge node; after the first edge node obtains the local model according to its own local data, the model parameters of the local model obtained by itself are sent to the second edge node; then the second edge node iteratively updates the local model according to the model parameters sent by the first edge node; the first edge node iteratively updates the local model according to the model parameters sent by the second edge node; the second edge node and the first edge node repeat the process of iteratively updating the local model until both parties establish a stable local model.

17. A video edge network system, characterized in that: The method comprises the first edge node according to claim 13 and at least one second edge node according to claim 15 .

18. An edge node, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the data modeling method according to any one of claims 1 to 12 are implemented.

19. A readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the data modeling method according to any one of claims 1 to 12 are implemented.

Images