Edge device supporting artificial intelligence inference

By designing modules and model adaptive technologies that support artificial intelligence inference in edge devices, traditional edge devices cannot meet the problem of real-time and privacy data leakage in complex industrial scenarios, achieving better adaptability and privacy protection.

CN118963999APending Publication Date: 2024-11-15HANGZHOU JIJIAN MATERIAL CONTROL TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410956801.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-17
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

Traditional edge devices do not support artificial intelligence inference, making it difficult to meet real-time requirements in complex industrial scenarios and pose a risk of privacy data leakage.

Method used

An edge device supporting artificial intelligence inference is designed, including communication modules, lightweight database modules, data processing modules, data assembly modules, artificial intelligence inference engine modules and control modules, using model storage, model fine-tuning and model adaptive technologies. Implement distributed artificial intelligence inference and reduce privacy data leakage through upper-level system training and edge-side inference mode.

Benefits of technology

It realizes better adaptability of edge devices to complex industrial scenarios, has real-time optimization and model adaptability, and reduces the risk of privacy data leakage.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118963999A_ABST
    Figure CN118963999A_ABST
Patent Text Reader

Abstract

The invention provides an edge device supporting artificial intelligence inference. The edge device comprises a communication module, a lightweight database module, a data processing module, a data assembly module, an artificial intelligence inference engine module and a control module. The communication module is used for collecting data from external equipment, obtained measurement data are stored in the lightweight database module, and after the artificial intelligence inference engine module completes solving, a solving result is sent to the control module and a control instruction is formed, the communication module issues the control instruction to the external equipment. The edge device supports artificial intelligence inference and solution, so that the edge device has better adaptability to more and more complex industrial scenes. Meanwhile, an artificial intelligence inference engine module in the edge device has the characteristics of real-time optimization and model self-adaption, and can better adapt to complex industrial scenes. In addition, the equipment adopts an upper-layer system training and edge side inference mode, so that leakage of privacy data of edge equipment can be effectively reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of edge devices, and in particular to an edge device supporting artificial intelligence inference. Background Art

[0002] Edge devices refer to small computers installed on on-site production equipment or Internet or IoT nodes. Traditional edge devices do not support artificial intelligence inference and only have basic functions such as data collection and data processing. Complex model training and inference generally upload data to central devices through edge devices for centralized processing. However, as computing problems in various industrial scenarios become more and more complex, traditional centralized processing methods are limited by data transmission delays and limited computing resources, and it is difficult to meet real-time requirements. At the same time, some privacy data of some devices sometimes need to be involved in calculations. If they are uploaded to the upper system in a unified manner, privacy leakage is inevitable. Therefore, it is urgent to invent an edge device that supports artificial intelligence inference to support complex problem solving in a distributed manner. Summary of the invention

[0003] In response to the problem that current edge devices cannot support artificial intelligence model inference, the present invention designs an edge device that supports artificial intelligence model inference.

[0004] The present invention is implemented by the following technical solutions:

[0005] An edge device supporting artificial intelligence inference includes: a communication module, a lightweight database module, a data processing module, a data assembly module, an artificial intelligence inference engine module and a control module; wherein:

[0006] The communication module is used to collect data from external devices, store the obtained measurement data in the lightweight database module, and at the same time receive the inference trigger instruction issued by the upper system, and pass it to the artificial intelligence inference engine module, triggering the artificial intelligence inference engine module to start solving. After the artificial intelligence inference engine module completes the solution and forms the control instruction, the communication module sends the control instruction to the external device and uploads the optimization control completion status to the upper system.

[0007] The lightweight database module is used to store the measurement data collected by the communication module and the solution results of the artificial intelligence inference engine module;

[0008] The data processing module is used to obtain the original data stored in the lightweight database module, and perform outlier detection and repair to obtain cleaned data, and transmit the cleaned data to the data assembly module;

[0009] The data assembly module is used to assemble the cleaned data and edge device privacy data into the data format required by the artificial intelligence inference engine module, and transmit it to the artificial intelligence inference engine module;

[0010] The artificial intelligence inference engine module is used to infer the assembled data based on the artificial intelligence model deployed by the upper system after receiving the inference trigger instruction, obtain the solution result, and transmit it to the control module;

[0011] The control module is used to receive the solution results of the artificial intelligence inference engine module, form control instructions according to the solution results, and then transmit the control instructions to the communication module.

[0012] In the above technical solution, further, the artificial intelligence inference engine module includes a model storage module, a model fine-tuning module, a model adaptation module and an inference module;

[0013] The model storage module is used to store one or more artificial intelligence models trained by the upper system;

[0014] The model fine-tuning module is used to continuously fine-tune the artificial intelligence model according to the data assembled by the data assembly module through online learning or transfer learning technology to adapt to changes in the environment or tasks;

[0015] The model adaptation module is used to dynamically select and load different artificial intelligence models according to the load, environment or user needs when the device is running;

[0016] The inference module performs inference based on the artificial intelligence model selected by the model adaptation module.

[0017] Furthermore, the external device is a device managed by the edge device, and the number of the external device is one or more. The upper layer system is used to manage the edge device.

[0018] The beneficial effects of the present invention are:

[0019] Compared with traditional edge devices, the edge devices designed by the present invention support artificial intelligence inference and solving, so that the edge devices can be better adapted to more and more complex industrial scenarios. At the same time, the artificial intelligence inference engine module in the edge device has the characteristics of real-time optimization and model adaptation, which can better adapt to complex industrial scenarios. In addition, the edge device of the present invention adopts the mode of upper-layer system training and edge-side inference, which can effectively reduce the leakage of privacy data of edge devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A schematic diagram of the structure of an edge device supporting artificial intelligence inference of the present invention;

[0021] Figure 2 This is a schematic diagram of the structure of the artificial intelligence inference engine module. DETAILED DESCRIPTION

[0022] The present invention is further described in detail below with reference to specific embodiments.

[0023] like Figure 1 As shown, the edge device supporting artificial intelligence inference designed by the present invention includes a communication module, a lightweight database module, a data processing module, a data assembly module, an artificial intelligence inference engine module, a control module and other components.

[0024] The communication module is used to collect data from external devices, and the obtained measurement data is stored in the lightweight database module. At the same time, it receives the inference trigger instruction issued by the upper system and passes it to the artificial intelligence inference engine module, triggering the artificial intelligence inference engine module to start solving. After the artificial intelligence inference engine module completes the solution and forms the control instruction, the communication module sends the control instruction to the external device and uploads the optimization control completion status to the upper system. The external device refers to the device managed by the edge device, which may be one or more; the upper system is used to manage the edge device.

[0025] The lightweight database module is used to store the measurement data collected by the communication module and the solution results obtained by the artificial intelligence inference engine module.

[0026] The data processing module is used to obtain the original data stored in the lightweight database module, detect and repair outliers, obtain cleaned data, and transmit it to the data assembly module. Among them, the measurement data is expressed and implemented according to the two-dimensional table structure, and the lightweight database module stores it as structured data.

[0027] The data assembly module is used to assemble the cleaned structured data and edge device privacy data into the data format required by the artificial intelligence inference engine module, and transmit it to the artificial intelligence inference engine module.

[0028] The AI ​​inference engine module is used to infer the assembled data based on the AI ​​model deployed by the upper system after receiving the inference trigger instruction, obtain the solution result, and transmit it to the control module. The upper system can also be used to train the AI ​​model and deploy it on the AI ​​inference engine module of the edge device. Figure 2As shown, after the artificial intelligence inference engine module receives one or more artificial intelligence models trained from the upper system, they are stored in the model storage module. The model fine-tuning module can continuously fine-tune the stored artificial intelligence model according to the new data (data assembled by the data assembly module) through online learning or transfer learning technology to adapt to changes in the environment or tasks; the model adaptation module can dynamically select and load different artificial intelligence models according to the load, environment or user needs during device operation; the inference module performs inference based on the artificial intelligence model selected by the model adaptation module.

[0029] The control module is used to receive the solution results of the artificial intelligence inference engine module, form control instructions according to the solution results, and then transmit the control instructions to the communication module.

[0030] Example

[0031] The edge device designed by the present invention can be used to perform economic optimization control on the microgrid. Assuming that a microgrid includes energy storage, photovoltaic, power load and other equipment, the external equipment controlled by the edge device includes energy storage, photovoltaic, power load and the like. The upper system is the EMS (Energy Management System) of the microgrid, which pre-trains the microgrid photovoltaic power prediction and power load power prediction models. The models include DNN, xgboost, LSTM, GNN, and GAN models. The five models are deployed in the model storage module of the artificial intelligence inference engine module.

[0032] The communication module collects energy storage SoC, photovoltaic power, power load power, etc., and stores these measurement data in the lightweight database SQLite of the edge device. The data processing module extracts the original data of the lightweight database for outlier detection and repair, and transmits the cleaned data to the data assembly module. The data assembly module assembles the collected information into the data format required for deploying the model in the artificial intelligence engine.

[0033] The artificial intelligence inference engine module model can fine-tune the DNN, xgboost, LSTM, GNN, and GAN models deployed by the upper system based on the latest collected data. After receiving the data transferred by the data assembly module, different models can be dynamically selected and loaded according to the load, environment, or user needs of the device during operation. For the load of the device during operation, for example, if the device load is heavy and the device power is insufficient during operation, the xgboost model with simpler calculation can be selected. If the device load is light and the device power is sufficient during operation, the DNN model with more complex calculation can be selected. For the environment or user needs, for example, if the processed data shows strong time series characteristics, the LSTM model can be selected. When the processed data contains graph structure characteristics, the GNN model can be selected. For user needs that include generation tasks, the GAN model can be selected. The results of the model prediction will be transmitted to the control module to form control instructions for energy storage and power loads, and issued to the corresponding devices for execution. In this way, the process of economic optimization control of the microgrid is completed.

Claims

1. An edge device supporting artificial intelligence inference, characterized in that: include: Communication module, lightweight database module, data processing module, data assembly module, artificial intelligence inference engine module and control module; wherein: The communication module is used to collect data from external devices, store the obtained measurement data in the lightweight database module, and receive the inference trigger instruction issued by the upper system, and pass it to the artificial intelligence inference engine module, triggering the artificial intelligence inference engine module to start solving. After the artificial intelligence inference engine module completes the solution and forms the control instruction, the communication module sends the control instruction to the external device and uploads the optimization control completion status to the upper system; The lightweight database module is used to store the measurement data collected by the communication module and the solution results of the artificial intelligence inference engine module; The data processing module is used to obtain the original data stored in the lightweight database module, and perform outlier detection and repair to obtain cleaned data, and transmit the cleaned data to the data assembly module; The data assembly module is used to assemble the cleaned data and edge device privacy data into the data format required by the artificial intelligence inference engine module, and transmit it to the artificial intelligence inference engine module; The artificial intelligence inference engine module is used to infer the assembled data based on the artificial intelligence model deployed by the upper system after receiving the inference trigger instruction, obtain the solution result, and transmit it to the control module; The control module is used to receive the solution results of the artificial intelligence inference engine module, form control instructions according to the solution results, and then transmit the control instructions to the communication module.

2. The edge device supporting artificial intelligence inference according to claim 1, characterized in that: The external device is a device managed by an edge device, and the number of the external device is one or more; the upper layer system is used to manage the edge device.

3. The edge device supporting artificial intelligence inference according to claim 1, characterized in that: The artificial intelligence inference engine module includes a model storage module, a model fine-tuning module, a model adaptation module and an inference module; The model storage module is used to store one or more artificial intelligence models trained by the upper system; The model fine-tuning module is used to continuously fine-tune the artificial intelligence model according to the data assembled by the data assembly module through online learning or transfer learning technology to adapt to changes in the environment or tasks; The model adaptation module is used to dynamically select and load different artificial intelligence models according to the load, environment or user needs when the device is running; The inference module performs inference based on the artificial intelligence model selected by the model adaptation module.