A smart operation and maintenance solution method in an Internet of Things application scenario

Abstract

The application discloses a kind of wisdom operation and maintenance solutions under the application scene of Internet of Things, belong to the technical field of Internet of Things application, and first, target sends business request, the node where target equipment is to Internet of Things platform sends request, Internet of Things platform receives request, searches the equipment satisfying demand in access platform, and sends request to these source equipment, wherein Internet of Things platform will these source equipment with target equipment with response form a group organization, then carry out joint learning, and the focus of joint learning is on joint learning sample weight, the sample of source equipment and target equipment is combined by the way of joint learning, then local model training and upload, i.e. model is uploaded to Internet of Things platform server, then model aggregation is carried out and is issued, the whole wisdom operation and maintenance solutions use the label data of same equipment in whole Internet of Things, to realize the wisdom operation and maintenance of local single device, while ensuring the privacy and security of data of each party.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of Internet of Things application, and in particular to a smart operation and maintenance solution in an Internet of Things application scenario. BACKGROUND

[0002] Smart operation and maintenance of equipment in an industrial scenario refers to intelligently predicting the probability of equipment failure or monitoring and estimating the real-time operation of equipment according to the state data of the equipment. Data-driven machine learning methods are commonly used and effective methods, and a large amount of high-quality labeled data is very important for this process. However, for a single device, it is relatively difficult to obtain a large amount of high-quality labeled data. For example, there is limited fault data for a single device, and there is a limited number of sensors installed on a single device. An effective method is to jointly obtain sufficient data from multiple devices of the same type, especially in an Internet of Things environment, where tens of thousands of devices are connected to the Internet of Things, providing a good basis for data joint. At the same time, problems that need to be considered include how to deal with the data differences of different devices and how to ensure the data security of each device. Therefore, a solution is needed that can complete smart operation and maintenance with the help of the Internet of Things while ensuring the privacy and security of data from all parties. SUMMARY

[0003] The present application aims to provide a smart operation and maintenance solution in an Internet of Things application scenario, which has the advantage of being able to use labeled data from the same equipment in the entire Internet of Things to realize smart operation and maintenance of a single device locally, while ensuring the privacy and security of data from all parties, to solve the problems raised in the background.

[0004] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0005] A smart operation and maintenance solution in an Internet of Things application scenario, comprising the following steps:

[0006] S1: A target device sends a business request, and the node where the target device is located sends a request to the Internet of Things platform, the request containing the target device model and the attribute information corresponding to the target device;

[0007] S2: According to the target business request, request joint learning, the Internet of Things platform receives the request, matches the devices in the access platform that meet the demand, and sends a request to these source devices, the request containing an incentive label for the target device;

[0008] S3: Request response;

[0009] S4: Joint the samples of the source device and the target device through joint learning, and adjust the source device data distribution to determine the joint learning sample weight;

[0010] S5: Local model training and uploading, the source device node trains the local model according to the joint learning sample weight data, and uploads the model to the Internet of Things platform server;

[0011] S6: Model aggregation and distribution, the Internet of Things platform server aggregates each source device node model and distributes it to the source device node respectively;

[0012] S7: Target business model distribution, iteratively execute S4 and S5 steps until convergence, and distribute the converged target model to the target node.

[0013] Preferably, the target device in S1 can be any device, and any device can send request information to the Internet of Things platform. The request information sent by different devices is not disturbed by others, and the device sending the request information can carry not only non-labeled data but also data lacking annotation. The request information can include a preliminary service fee offer invitation contract, and the content of the offer invitation contract is not encrypted. The offer invitation contract is broadcast to all devices in a decentralized manner.

[0014] Preferably, after receiving the request in S2, the Internet of Things platform retrieves all devices that meet the requirements in the access platform, and also retrieves those devices that meet part of the requirements, i.e. those devices of the same type, with the same sensors, and with the same labeled data.

[0015] Preferably, in S3, when all source devices receive the request, some of them will respond. The Internet of Things platform collects these responding source devices and the target device to form a group organization. At the same time, the Internet of Things platform will also receive responses again after a period of time, because some source devices did not respond effectively due to various reasons when they first received the request. These source devices that meet the conditions will resend the response at any time. In this way, the Internet of Things platform can receive the responses of these source devices again and form a group organization, which is directly added to the previously formed group organization.

[0016] Preferably, in S4, the joint learning sample weight is divided into multiple types, including event processing optimal solution, event processing general solution, and event processing lag solution. The sample weight coefficient of the event processing lag solution is much larger than that of the event processing optimal solution. Hash encryption is used in the joint learning process. The Internet of Things platform hashes the confirmed response information and processing data to form a specific data chain, and then forms a data processing data chain according to the time axis and stores it in each node on the blockchain. Each node stores the data processing result of the source device.

[0017] Preferably, in S4 joint learning, different source devices need to enter other source devices when learning from each other, at which time the qualifications of the source devices need to be examined, and only the source devices in the mutual learning process can access other source devices of the group organization; when a source device has the qualification to access other source devices of the group organization, the source device sends an access request to the group organization, the group organization prompts the source device to authorize using the node data on the blockchain, and only when the group organization authorizes can the source device read the processing data of other source devices of the group organization. If the source device is not authorized, it cannot read.

[0018] Preferably, in S4 joint learning, when a source device wants to access the data of other source devices of the group organization, a qualified source device will make a broadcast in the group organization and the Internet of Things platform system and leave a corresponding hash value when using the request query and accessing the group organization. An unqualified source device will not leave a corresponding hash value when using the request query and accessing the group organization, but will leave a timestamp in the group organization and the Internet of Things platform system.

[0019] Preferably, in S5 local model training and uploading, each device of the group organization has a large amount of running data, which will result in a large amount of label data. The label data, which needs to be predicted in intelligent operation and maintenance, such as fault data and relatively high-priced sensor data, is selected by the group organization. The group organization selects the event processing optimal solution label data and the event processing general solution label data from the large amount of label data and uploads the node model and the large amount of label data to the Internet of Things platform server.

[0020] Preferably, in S6 model aggregation and distribution, the Internet of Things platform server first receives the label data and the node model sent by the entire organization. The Internet of Things platform server sorts the node model and the label data of each source device and distributes the optimal solution label data and the node model of each source device to the source device node.

[0021] Preferably, in S7 target business model distribution, S4 joint learning and S5 local model training and uploading are repeatedly executed until convergence, and then the converged target model is distributed to the target node. Specifically, in the process of S4 joint learning, the joint learning samples are divided into multiple types, including event processing optimal solution, event processing general solution, and event processing lag solution. In S5 local model training and uploading, the group organization uploads the event processing optimal solution label data and the event processing general solution label data to the Internet of Things platform server. Therefore, in S7 target business model distribution, the event processing optimal solution and the event processing general solution need to be selected, and the event processing general solution is eliminated and the event processing optimal solution is left, which facilitates the target node to solve technical problems more quickly.

[0022] Compared with the prior art, the beneficial effects of the present application are: the intelligent operation and maintenance solution method under the application scene of the Internet of Things proposed by the present application, first, the target sends a business request, the node where the target device is located sends a request to the Internet of Things platform, the request contains the target device model and the attribute information corresponding to the target device, according to the target sends a business request, then requests joint learning, the Internet of Things platform receives the request, matches the devices in the access platform that meet the demand, and sends a request to these source devices, the request contains the incentive label for the target device, then requests response, then joint learning, the joint learning focuses on joint learning sample weight, through joint learning, the samples of the source device and the target device are combined, the data distribution of the source device is adjusted, which can be realized by adjusting the sample weight, then local model training and uploading, then model aggregation and distribution, the Internet of Things platform server aggregates the models of each source device node and distributes them to the source device nodes respectively, the whole intelligent operation and maintenance solution method uses the label data of the same device in the whole Internet of Things to realize the intelligent operation and maintenance of the local single device, while ensuring the privacy and security of the data of each party. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 The overall principle diagram of the present application is shown in the figure;

[0024] Figure 2 The overall principle flow chart of the present application is shown in the figure;

[0025] Figure 3 The target sends a business request schematic diagram of the present application is shown in the figure;

[0026] Figure 4 The joint learning forms a specific database schematic diagram of the present application is shown in the figure;

[0027] Figure 5 The joint learning reference process flow schematic diagram of the present application is shown in the figure;

[0028] Figure 6 The joint learning reference trace schematic diagram of the present application is shown in the figure

[0029] Figure 7 The model aggregation and distribution flow chart of the present application is shown in the figure. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0031] Embodiment one:

[0032] Please refer to Figures 1-7 A smart operation and maintenance solution under the Internet of Things application scenario, comprising the following steps:

[0033] Step 1: The target device sends a business request, and the node where the target device is located sends a request to the Internet of Things platform, which contains the target device model and the attribute information corresponding to the target device. Any device can send request information to the Internet of Things platform, and the request information sent by different devices is not disturbed by others. The device sending the request information can not only carry non-labeled data but also carry data lacking of labeling. The request information can contain a preliminary service fee offer invitation contract, and the content of the offer invitation contract is not encrypted. The offer invitation contract is broadcast to all devices in a decentralized manner.

[0034] Step 2: According to the target business request, request joint learning, the Internet of Things platform receives the request, matches the devices in the access platform that meet the demand, and sends a request to these source devices, which contains the incentive label for the target device and the available incentives. After receiving the request, the Internet of Things platform will retrieve all devices in the access platform that meet the demand, and also retrieve those devices that meet part of the demand.

[0035] Step 3: Request response, after receiving the request, some source devices will respond, and the Internet of Things platform collects the information of the source devices that respond after sending the request message, and forms a group organization with the target device and the source devices that respond.

[0036] Step 4: Joint samples of source devices and target devices through joint learning, adjust source device data distribution to determine joint learning sample weight, and the joint learning samples are divided into multiple types, which can be divided into event processing optimal solution, event processing general solution and event processing lag solution. The sample weight coefficient of the event processing lag solution is much larger than that of the event processing optimal solution. The process of joint learning needs to use hash encryption processing, in which the Internet of Things platform hashes and encrypts the confirmed response information and processing data to form a specific data chain, and then forms a data processing data chain according to the time axis and stores it in each node on the block chain. Each node stores the data processing result of the source device.

[0037] Step 5: Local model training and uploading, the source device node trains the local model according to the joint learning sample weight data and uploads the model to the Internet of Things platform server.

[0038] Step 6: After model aggregation, the IoT platform server aggregates each source device node model and sends it to the source device node. The IoT platform server first accepts the tag data volume and node model sent by the entire organization, and then sorts the optimal solution tag data volume and each source device node model and sends them to the source device node.

[0039] Embodiment two:

[0040] Please refer to Figures 1-7 A smart operation and maintenance solution under the Internet of Things application scenario, comprising the following steps:

[0041] Step 1: The target sends a business request, and the node where the target device is located sends a request to the IoT platform. The request contains the target device model and the attribute information corresponding to the target device. The target device can be any device, and any device can send request information to the IoT platform. The request information sent by different devices is not disturbed by others. The device sending the request information can carry not only non-labeled data but also data lacking labels. The request information can contain a preliminary service fee offer invitation contract, and the content of the offer invitation contract is not encrypted. The offer invitation contract is broadcast to all devices in a decentralized manner.

[0042] Step 2: According to the target business request, request joint learning, the IoT platform receives the request, matches the devices in the access platform that meet the demand, and sends a request to these source devices. The request contains the incentive label for the target device and the available incentives. After receiving the request, the IoT platform retrieves all devices in the access platform that meet the demand, and also retrieves those devices that meet part of the demand, i.e. those devices of the same model, with the same sensors, and with the same labeled data.

[0043] Step 3: Request response, after receiving the request, some source devices will respond. The IoT platform collects the source device node information of the responding source devices after receiving the request message, and forms a group organization with the target device. When all source devices receive the request, some source devices will respond. The IoT platform collects the source devices that respond and the target device to form a group organization. At the same time, the IoT platform will also receive responses again after a period of time, because some source devices did not respond effectively due to various reasons when they first received the request information. These source devices that meet the conditions will resend the response at any time. In this way, the IoT platform can receive the responses of these source devices again and form a group organization, which is directly added to the previously formed group organization.

[0044] Fourth step: adjust the source device data distribution to determine the joint learning sample weight by joint learning of the source device and the target device samples. The joint learning samples are divided into multiple types, which can be divided into event processing optimal solution, event processing general solution and event processing lag solution. The sample weight coefficient of the event processing lag solution is much larger than that of the event processing optimal solution. The joint learning process needs to use hash encryption processing. The Internet of Things platform hashes and encrypts the confirmed response information and processing data to form a specific data chain, and then forms a data processing data chain according to the time axis and stores it on each node of the block chain. Each node stores the data processing result of the source device. Different source devices need to enter other source devices when learning from each other. At this time, the qualifications need to be examined. Only the source devices in the mutual learning process can refer to other source devices of other groups.

[0045] Fifth step: local model training and uploading. The source device node trains the local model according to the joint learning sample weight data and uploads the model to the Internet of Things platform server. Each device in the group organization has a large amount of running data, so a large amount of labeled data is obtained. The labeled data is the data that needs to be predicted in intelligent operation and maintenance, such as fault data and relatively high-priced sensor data. The group organization selects event processing optimal solution labeled data and event processing general solution labeled data from a large amount of data, and uploads a large amount of labeled data and source device node models to the Internet of Things platform server.

[0046] Sixth step: model aggregation and distribution. The Internet of Things platform server aggregates each source device node model and distributes it to the source device node. The Internet of Things platform server first accepts the labeled data and node model sent by the whole organization, sorts the source device node model and labeled data, and distributes the optimal solution labeled data and source device node model to the source device node.

[0047] Step 7: Target business model is issued, and steps S4 and S5 are iteratively executed until convergence, and the converged target model is issued to the target node, that is, the steps of performing S4 joint learning and S5 local model training and uploading are repeatedly executed until convergence, and then the converged target model is issued to the target node; Specifically, in the process of S4 joint learning, the joint learning samples are divided into multiple types, including event processing optimal solution, event processing general solution and event processing lag solution, and in the S5 local model training and uploading step, the event processing optimal solution label data and the event processing general solution label data are uploaded to the Internet of Things platform server. In this way, in the S7 target business model issuing process, the event processing optimal solution and the event processing general solution need to be selected, and the event processing general solution is eliminated and the event processing optimal solution is left, which facilitates the target node to solve technical problems faster.

[0048] Embodiment three

[0049] Please refer to Figures 1-7 A smart operation and maintenance solution under an Internet of Things application scenario, comprising the following steps:

[0050] Step 1: The target issues a business request, and the node where the target device is located issues a request to the Internet of Things platform, the request containing the target device model and the attribute information corresponding to the target device. The target device can be any device, and any device can send request information to the Internet of Things platform. The request information sent by different devices is not disturbed by others. The device sending the request information can carry not only non-labeled data but also data lacking labels. The request information can contain a preliminary service fee offer invitation contract, and the content of the offer invitation contract is not encrypted. The offer invitation contract is broadcast to all devices in a decentralized manner.

[0051] Step 2: According to the target business request, request joint learning, the Internet of Things platform receives the request, matches the devices in the access platform that meet the demand, and issues a request to these source devices, the request containing the incentive label for the target device and the available incentives. After receiving the request, the Internet of Things platform retrieves all devices in the access platform that meet the demand, and also retrieves those devices that meet part of the demand, that is, those devices of the same model, with the same sensors, and with the same labeled data.

[0052] Third step: request response, after the source device receives the request, part of the response, among which the Internet of Things platform collects the source device node information of the responding source device after sending the request message, and forms a group organization with the target device. When all source devices receive the request, some source devices will respond. Among them, the Internet of Things platform collects these responding source devices and target devices to form a group organization. At the same time, the Internet of Things platform will also receive responses again after a period of time, because some source devices did not respond effectively when they first received the request information due to various reasons. These source devices that meet the conditions will reissue response receipts at any time. In this way, the Internet of Things platform can re-receive the responses of these source devices and form a group organization, which is directly added to the previously formed group organization;

[0053] Fourth step: adjust the sample weight of joint learning of source devices and target devices to determine the joint learning sample weight, and the sample of joint learning is divided into multiple types, which can be divided into event processing optimal solution, event processing general solution and event processing lag solution. The sample weight coefficient of event processing lag solution is much larger than that of event processing optimal solution. The process of joint learning needs to use hash encryption processing. The Internet of Things platform hashes the confirmed response information and processing data to form a specific data chain, and then forms a data processing data chain according to the time axis and stores it in each node on the block chain. Each node stores the data processing result of the source device. Different source devices need to enter other source devices when learning from each other. At this time, the qualifications of the source devices need to be examined. Only the source devices in the mutual learning process can refer to other source devices in other group organizations. When the source device is qualified to refer to other source devices in the group organization, the source device sends a reference request to the group organization, and the group organization prompts the source device to use the node data authorization on the block chain. Only when the group organization authorizes, the source device can read the processing data of other source devices in the group organization. If the source device is not authorized, it cannot read. When the source device wants to refer to the data of other source devices in the group organization, the qualified source device will make a broadcast in the group organization and the Internet of Things platform system when using the request query and referring to the group organization, and leave the corresponding hash value. The unqualified source device will not leave the corresponding hash value when using the request query and referring to the group organization, but will leave the timestamp in the group organization and the Internet of Things platform system;

[0054] Step 5: Local model training and uploading, the source device node trains a local model according to the joint learning sample weight data, and uploads the model to the Internet of Things platform server. Each device in the group organization has a large amount of running data, so a large amount of labeled data is obtained. The labeled data is used for prediction in intelligent operation and maintenance, such as fault data and relatively high-priced sensor data. The group organization selects the event processing optimal solution labeled data and the event processing general solution labeled data from the large amount of data, and uploads the large amount of labeled data and the source device node model to the Internet of Things platform server;

[0055] Step 6: Model aggregation and distribution, the Internet of Things platform server aggregates each source device node model and distributes it to the source device node. The Internet of Things platform server first accepts the labeled data and the node model sent by the whole organization, sorts the source device node model and the labeled data, and distributes the optimal solution labeled data and the source device node model to the source device node.

[0056] Step 7: Target business model distribution, repeat steps S4 and S5 until convergence, and then distribute the converged target model to the target node. Specifically, in the process of S4 joint learning, the joint learning samples are divided into multiple types, including event processing optimal solution, event processing general solution, and event processing lag solution. In the S5 local model training and uploading step, the group organization uploads the event processing optimal solution labeled data and the event processing general solution labeled data to the Internet of Things platform server. In the S7 target business model distribution process, the event processing optimal solution and the event processing general solution need to be selected, and the event processing general solution is eliminated and the event processing optimal solution is left. This facilitates the target node to solve technical problems more quickly.

[0057] S1: Target business request, the node where the target device is located sends a request to the Internet of Things platform, and the request contains the target device model and the attribute information corresponding to the target device;

[0058] S2: Request joint learning according to the target business request, the Internet of Things platform receives the request, matches the devices that meet the demand in the access platform, and sends a request to these source devices, which contains the incentive label for the target device;

[0059] S3: Request response;

[0060] S4: Joint source device and target device samples through joint learning, adjust source device data distribution to determine joint learning sample weight;

[0061] S5: Local model training and uploading, the source device node trains a local model according to the joint learning sample weight data, and uploads the model to the Internet of Things platform server;

[0062] S6: Model aggregation and distribution, the Internet of Things platform server aggregates the models of each source device node and distributes them to the source device nodes respectively;

[0063] S7: Target business model distribution, iteratively execute steps S4 and S5 until convergence, and distribute the converged target model to the target node

[0064] The smart operation and maintenance solution in the Internet of Things application scenario, the target sends a business request, the node where the target device is located sends a request to the Internet of Things platform, the request contains the target device model and the attribute information corresponding to the target device, according to the target business request, request joint learning, the Internet of Things platform receives the request, matches the devices that meet the demand in the access platform, and sends a request to these source devices, the request contains the incentive label for the target device, then the request is answered, after the source device receives the request, some will answer, among them, the Internet of Things platform collects the information of the source device nodes that have answered after sending the request message, forms a group organization with the target device and the source devices that have answered, adjusts the sample weight of the joint learning of the source devices and the target device through joint learning, adjusts the data distribution of the source devices to determine the joint learning sample weight, then trains and uploads the local model, the source device node trains a local model according to the joint learning sample weight data, and uploads the model to the Internet of Things platform server, then performs model aggregation and distribution, the Internet of Things platform server aggregates the models of each source device node and distributes them to the source device nodes respectively, finally, the target business model is distributed, the steps S4 and S5 are iteratively executed until convergence, and the converged target model is distributed to the target node, that is, the steps S4 joint learning and S5 local model training and uploading are repeatedly executed until convergence, and then the converged target model is distributed to the target node.

[0065] To sum up, the intelligent operation and maintenance solution method under the application scene of the Internet of Things provided by the application is that the target sends a business request, the node where the target device is sends a request to the Internet of Things platform, the request contains the target device model and the attribute information corresponding to the target device, according to the target sending a business request, then request joint learning, the Internet of Things platform receives the request, matches the devices meeting the demand in the access platform, and sends a request to these source devices, the request contains the incentive label for the target device, then request response, then joint learning, the joint learning focuses on joint learning sample weight, the samples of the source device and the target device are combined through joint learning, the data distribution of the source device is adjusted, which can be realized by adjusting the sample weight, then local model training and uploading, then model aggregation and distribution, the Internet of Things platform server aggregates the models of each source device node and distributes them to the source device nodes respectively, the whole intelligent operation and maintenance solution method realizes the intelligent operation and maintenance of the local single device by means of the label data of the same device in the whole Internet of Things, and meanwhile ensures the privacy and security of the data of each party.

[0066] It should be noted that, in this document, the terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "comprising", or any other variations thereof are intended to cover non-exclusive inclusions, so that a process, method, article, or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or inherent to such a process, method, article, or device.

[0067] The above description is only the preferred embodiment of the application, but the protection scope of the application is not limited thereto, any person skilled in the art can make equivalent replacements or changes to the technical solution and the inventive concept of the application within the technical range disclosed by the application, which should be covered by the protection scope of the application.

Claims

1. A smart operation and maintenance solution for IoT application scenarios, characterized in that, Includes the following steps: S1: The target sends a service request. The node where the target device is located sends a request to the IoT platform. The request contains the target device model and the corresponding attribute information of the target device. S2: A business request is issued according to the target, requesting joint learning. The IoT platform receives the request, matches the devices in the access platform that meet the requirements, and sends a request to these source devices. The request contains an incentive label for the target device. S3: Request for response; S4: Combine samples from the source device and the target device through joint learning, and adjust the data distribution of the source device to determine the weights of the joint learning samples; In the S4 joint learning sample weights, the joint learning samples are divided into multiple types, which can be divided into event processing optimal solutions, event processing general solutions, and event processing lagged solutions. The sample weight coefficient of the event processing lagged solutions is greater than that of the event processing optimal solutions. The joint learning process requires hash encryption processing. The IoT platform hashes and encrypts the confirmed response information and processing data to form a specific data chain. Then, a data processing data chain is formed according to the time axis and stored on each node of the blockchain. Each node stores the data processing results of the source device. S5: Local model training and uploading: The source device node trains a local model based on the joint learning sample weight data and uploads the model to the IoT platform server. S6: After model aggregation, the IoT platform server aggregates the models from each source device node and distributes them to the source device nodes respectively. S7: The target business model is deployed. Steps S4 and S5 are executed iteratively until convergence. The converged target model is then deployed to the target node.

2. The intelligent operation and maintenance solution for an IoT application scenario according to claim 1, characterized in that, The target device in S1 can be any device. Any device can send a request message to the IoT platform. The request messages sent by different devices are not affected by others. The device sending the request message can carry not only unlabeled data but also data that is not labeled. The request message can include a preliminary service fee offer invitation contract. The content of the offer invitation contract is not encrypted and is broadcast to all devices in a decentralized manner.

3. The intelligent operation and maintenance solution for an IoT application scenario according to claim 1, characterized in that, After receiving a request, the IoT platform in S2 will search all devices in the access platform that meet the requirements. It will also search for devices that meet some of the requirements, namely those with the same model, the same sensor, and the same labeled data.

4. The intelligent operation and maintenance solution for an IoT application scenario according to claim 1, characterized in that, When a request is made in S3, after all source devices receive the request, some source devices will respond. The IoT platform collects these responding source devices and the target device to form a group. At the same time, the IoT platform will receive responses again at regular intervals. This is because some source devices did not respond effectively when they first received the request information for various reasons. These qualified source devices will re-issue response receipts at any time. In this way, the IoT platform can re-receive the responses from these source devices and form a group. This group is directly added to the previously formed group.

5. The intelligent operation and maintenance solution for an IoT application scenario according to claim 1, characterized in that, In S4 collaborative learning, different source devices need to access other source devices when learning from each other. This requires qualification verification. Only source devices in the mutual learning process can access other source devices in other groups. When a source device is qualified to access other source devices in a group, it sends a access request to the group. The group prompts the source device to authorize using node data on the blockchain. Only after the group authorizes can the source device read the processed data of other source devices in the group. If the source device does not authorize, it cannot read the data.

6. The intelligent operation and maintenance solution for an IoT application scenario according to claim 5, characterized in that, In S4 joint learning, when a source device wants to access data from other source devices in the group, qualified source devices will broadcast the request to the group and IoT platform system and leave a corresponding hash value when they query and access the data. Unqualified source devices will not leave a corresponding hash value when they query and access the data in the group, but they will leave a timestamp in the group and IoT platform system.

7. The intelligent operation and maintenance solution for an IoT application scenario according to claim 1, characterized in that, In the S5 local model training and uploading process, each device in the swarm organization has a large amount of operational data, which results in a large amount of labeled data. This labeled data is used as the data to be predicted in smart operation and maintenance, including fault data and relatively expensive sensor data. The swarm organization filters the large amount of data, selecting the labeled data of the optimal solution for event processing and the labeled data of the general solution for event processing. The swarm organization then uploads the large amount of labeled data and the source device node model to the IoT platform server.

8. The intelligent operation and maintenance solution for an IoT application scenario according to claim 1, characterized in that, During the S6 model aggregation and distribution process, the IoT platform server first receives the tag data volume and node model sent by the entire organization. The IoT platform server sorts the tag data volume and the optimal solution tag data volume and the individual source device node model and distributes them to the source device nodes respectively.

9. The intelligent operation and maintenance solution for an IoT application scenario according to claim 1, characterized in that, During the deployment of the S7 target business model, the S4 joint learning and S5 local model training and uploading steps are repeatedly executed until convergence. Then, the converged target model is deployed to the target node. During the S4 joint learning process, the joint learning samples are divided into multiple types, including the optimal solution for event handling, the general solution for event handling, and the lagged solution for event handling. At the same time, during the S5 local model training and uploading step, the group organization uploads the label data of the optimal solution for event handling and the label data of the general solution for event handling to the IoT platform server. Thus, during the deployment of the S7 target business model, it is necessary to select between the optimal solution for event handling and the general solution for event handling. The general solution for event handling will be eliminated, and the optimal solution for event handling will be retained.

Images