KR102456900B1 - Data analysis system based on edge computing and method thereof - Google Patents

Abstract

An edge computing based data analysis system is provided. The data analysis system is connected to a server for generating and distributing an analysis model for update based on data collected from at least some of a plurality of terminal devices forming a plurality of groups, and at least some of the plurality of terminal devices, the server In response to receiving the update analysis model from Edge devices may be included. In this case, the server may adjust a group belonging to at least some of the plurality of terminal devices based on a result of performing the update of the analysis model of the plurality of edge devices.

Description

DATA ANALYSIS SYSTEM BASED ON EDGE COMPUTING AND METHOD THEREOF

The present invention relates to an edge computing-based data analysis system and a method therefor. More particularly, it relates to a data analysis system including a server for generating and distributing an analysis model, and an edge device for analyzing data of a connected terminal device using the distributed analysis model, and a method performed in the system.

As the Internet of Things (IoT) technology becomes common, IoT platforms are being built in various environments. For example, an IoT platform may be built in an environment such as a home or a building to implement a smart home, a smart building, a smart city, and the like.

The conventional IoT platform consists of an IoT device, an IoT gateway that connects multiple IoT devices to the Internet, and a server that analyzes sensor data collected from the IoT device to perform overall management and control of the IoT device. Typically, the server performs analysis using the same analysis model for the same type of IoT device. For example, when a server controls a plurality of air conditioners in a smart building, the server analyzes sensor data collected from each air conditioner using one analysis model regardless of the installation location of the air conditioners, and based on the analysis results, the server analyzes the plurality of air conditioners. control each.

However, in the case of the conventional IoT platform as described above, several problems exist.

First, since the data analysis is delayed until the sensor data of the IoT device reaches the server through the IoT gateway, there is a problem in that it is difficult to immediately control the IoT device.

In addition, since one analysis model is collectively applied to the same type of IoT device, there is a problem in that the accuracy of the analysis result is lowered. For example, assuming that the analysis model is a model that predicts whether an IoT device is abnormal, even if it is the same type of IoT device (e.g. air conditioner), it is possible to judge normal and abnormal according to the manufacturer (e.g. Samsung air conditioner, LG air conditioner) or device model. Standards may vary. Alternatively, even for the same IoT device up to the manufacturer, the criteria for judging normality and abnormality may vary depending on the installation location due to the influence of the surrounding environment. Therefore, even for the same type of IoT device, if data collected from a plurality of IoT devices is analyzed through one analysis model, the accuracy of the analysis result is inevitably reduced.

Korean Patent Publication No. 10-2017-0112467 (published on October 12, 2017)

A technical problem to be solved through embodiments of the present invention is to provide a data analysis system having a gradual update function for an analysis model and a method performed in the system.

Another technical problem to be solved through an embodiment of the present invention is to provide a data analysis system supporting a group adjustment function for a plurality of terminal devices and a method performed in the system.

Another technical problem to be solved through the embodiments of the present invention is to provide a data analysis system capable of efficiently and accurately analyzing a large amount of IoT data in an IoT environment and a method performed in the system.

Another technical problem to be solved through an embodiment of the present invention is to provide a server for automatically generating and distributing an analysis model specialized for a specific terminal device, and a method performed by the server.

Another technical problem to be solved through embodiments of the present invention is to provide an edge device for performing an update to an optimal analysis model suitable for a connected terminal device, and a method performed by the edge device.

The technical problems of the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art from the following description.

In order to solve the above technical problem, an edge computing-based data analysis system according to an embodiment of the present invention generates an analysis model for update based on data collected from at least some of a plurality of terminal devices forming a plurality of groups, and It is connected to at least a part of a server to distribute and the plurality of terminal devices, and in response to receiving the analysis model for update from the server, performs performance evaluation on the existing analysis model and the analysis model for update, the performance evaluation and a plurality of edge devices for determining whether to update the existing analysis model based on a result, wherein the server includes at least one of the plurality of terminal devices based on the results of performing the update of the analysis model of the plurality of edge devices. Some groups can be adjusted.

In an embodiment, the server selects an edge device that has been updated with the analysis model for update from among the plurality of edge devices, and the plurality of terminals so that the terminal device connected to the selected edge device belongs to the same group You can adjust the group membership of the device.

In an embodiment, the existing analysis model is a model generated based on data collected from all of the plurality of terminal devices, and the analysis model for update is generated based on data collected from some of the plurality of terminal devices. can be a model.

In an embodiment, the existing analysis model is a model generated based on data collected during a first period from at least some of the plurality of terminal devices, and the analysis model for update is generated from at least some of the plurality of terminal devices. It may be a model generated based on data collected after the first period.

In an embodiment, the plurality of groups are configured in a hierarchical structure, and a first group of the plurality of groups includes a terminal device belonging to a second group that is a lower hierarchical group of the first group, and the server includes , a first update analysis model generated based on data collected from the first group of terminal devices is distributed among the plurality of edge devices to an edge device connected to the first group of terminal devices, and the second group of The second update analysis model generated based on data collected from the terminal device may be distributed to edge devices connected to the second group of terminal devices among the plurality of edge devices.

In an embodiment, the first edge device among the plurality of edge devices performs performance evaluation on the existing analysis model and the update analysis model using pre-stored data for evaluation, wherein the pre-stored data for evaluation is It may contain data in which a false positive error or a false negative error occurred.

In an embodiment, a first edge device of the plurality of edge devices, in response to determining that an available resource is above a threshold, based on data collected from a terminal device connected to the first edge device, an analysis model for a first update can be created directly.

In order to solve the above technical problem, an edge computing-based data analysis system according to another embodiment of the present invention generates and distributes an analysis model based on data collected from at least some of a plurality of terminal devices including a target terminal device an edge device that is connected to a server and the target terminal device, collects data from the target terminal device, transmits the data to the server, and analyzes data received from the target terminal device using an analysis model distributed by the server wherein the server generates and distributes a general-purpose analysis model associated with the plurality of terminal devices based on data collected from the plurality of terminal devices to the edge device, and in response to determining that a model update condition is satisfied An update analysis model associated with the target terminal device may be generated based on the data collected from the target terminal device and distributed to the edge device.

In an embodiment, the model update condition is that an update request message for the general-purpose analysis model is received from the edge device, and the edge device applies data received from the target terminal device to the general-purpose analysis model to predict output a result, increase an update index in response to determining that an error exists in the prediction result, and transmit the update request message to the server in response to determining that the increased update index is greater than or equal to a threshold.

In an embodiment, the model update condition is that an update request message for the general-purpose analysis model is received from the edge device, and the general-purpose analysis model is generated based on data collected from the plurality of terminal devices during a first period analyzed model, wherein the server, when the update request message is received, determines whether the amount of data collected from the target terminal device is less than a threshold value, and in response to the determination that the amount of data collected from the target terminal device is less than the threshold value, after the first period may update the general-purpose analysis model based on data collected from the plurality of terminal devices, and distribute the updated general-purpose analysis model to the edge device.

In an embodiment, the plurality of terminal devices may belong to a first group among a plurality of groups grouped in a hierarchical structure, and the target terminal device may belong to a second group that is a lower hierarchical group of the first group.

A server according to another embodiment of the present invention for solving the above technical problem includes a communication interface for communicating with a plurality of edge devices connected to at least some of a plurality of terminal devices including a target terminal device, and one or more instructions ( instructions) and executing the one or more instructions to generate a general-purpose analysis model based on data collected from the plurality of terminal devices, and distribute the general-purpose analysis model to edge devices connected to the plurality of terminal devices and a processor for generating a target analysis model based on data collected from the target terminal device in response to a target model generation condition being satisfied, and distributing the target analysis model to a target edge device connected to the target terminal device can do.

A computer program according to another embodiment of the present invention for solving the above-described technical problem is combined with a computing device, generating a general-purpose analysis model based on data collected from a plurality of terminal devices, the general-purpose analysis model distributing to an edge device connected to the plurality of terminal devices; generating a target analysis model based on data collected from the target terminal device in response to a target model generation condition being satisfied; and generating the target analysis model as the target analysis model. In order to execute the step of distributing to the target edge device connected to the target terminal device, it may be stored in a computer-readable recording medium.

1 is a block diagram showing a data analysis system according to an embodiment of the present invention.
2 is a block diagram illustrating a data analysis system according to another embodiment of the present invention.
3 is a block diagram illustrating a server according to an embodiment of the present invention.
4 is a block diagram illustrating an edge device according to an embodiment of the present invention.
5 is a hardware configuration diagram illustrating an exemplary computing device capable of implementing a server and an edge device according to some embodiments of the present invention.
6 is a flowchart illustrating a method of updating an analysis model of a data analysis system according to an embodiment of the present invention.
7 is a detailed flowchart of the analysis model distribution step (S30 or S50) shown in FIG.
FIG. 8 is a detailed flowchart of the step S40 of generating an analysis model for updating shown in FIG. 6 .
9 is a detailed flowchart of the analysis model update step S60 shown in FIG. 6 .
10 is a flowchart illustrating a method for requesting an analysis model update of an edge device according to an embodiment of the present invention.
11 is a flowchart illustrating a method of generating an analysis model of an edge device according to an embodiment of the present invention.
12 to 15 are exemplary diagrams for explaining an analysis model update method performed in a data analysis system having a hierarchical structure according to an embodiment of the present invention.
16 is a flowchart illustrating a device group adjustment method of a data analysis system according to an embodiment of the present invention.
17 and 18 are flowcharts illustrating a device group adjustment method performed in a data analysis system having a hierarchical structure according to an embodiment of the present invention.
19 is a configuration diagram illustrating a smart building management system according to an application example of the present invention.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. Advantages and features of the present invention and methods of achieving them will become apparent with reference to the embodiments described below in detail in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but may be implemented in various different forms, and only these embodiments allow the disclosure of the present invention to be complete, and common knowledge in the technical field to which the present invention pertains It is provided to fully inform those who have the scope of the invention, and the present invention is only defined by the scope of the claims.

In adding reference numerals to the components of each drawing, it should be noted that the same components are given the same reference numerals as much as possible even though they are indicated on different drawings. In addition, in describing the present invention, if it is determined that a detailed description of a related known configuration or function may obscure the gist of the present invention, the detailed description thereof will be omitted.

Unless otherwise defined, all terms (including technical and scientific terms) used herein may be used with the meaning commonly understood by those of ordinary skill in the art to which the present invention belongs. In addition, terms defined in a commonly used dictionary are not to be interpreted ideally or excessively unless clearly defined in particular. The terminology used herein is for the purpose of describing the embodiments and is not intended to limit the present invention. In this specification, the singular also includes the plural, unless specifically stated otherwise in the phrase.

In addition, in describing the components of the present invention, terms such as first, second, A, B, (a), (b), etc. may be used. These terms are only for distinguishing the elements from other elements, and the essence, order, or order of the elements are not limited by the terms. When it is described that a component is “connected”, “coupled” or “connected” to another component, the component may be directly connected or connected to the other component, but another component is between each component. It should be understood that elements may be “connected,” “coupled,” or “connected.”

As used herein, "comprises" and/or "comprising" refers to the presence of one or more other components, steps, operations and/or elements mentioned. or addition is not excluded.

Hereinafter, some embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram showing a data analysis system according to an embodiment of the present invention.

Referring to FIG. 1 , the data analysis system may include a server 100 , a plurality of edge devices 200 , and a plurality of terminal devices 300 . However, this is only a preferred embodiment for achieving the object of the present invention, and it goes without saying that some components may be added or deleted as needed. In addition, it should be noted that each of the components of the data analysis system shown in FIG. 1 represents functionally separated functional elements, and a plurality of components may be implemented in a form that is integrated with each other in an actual physical environment.

In the data analysis system, the server 100 is a computing device that manages the data of the terminal device 300 collected through the edge device 200 and creates and distributes an analysis model for analyzing the collected data. . Here, the computing device may be a notebook computer, a desktop computer, a laptop computer, and the like, but is not limited thereto and may include any type of device equipped with a computing function and a communication function. However, if a plurality of terminal devices 300 and edge devices 200 exist, the server 100 may be preferably implemented as a high-performance computing device.

The analysis model refers to any model used by the edge device 200 to analyze data collected from the terminal device 300 . For example, the analysis model may be a model for predicting whether the terminal device 300 is abnormal, but the scope of the present invention is not limited thereto.

In some embodiments, the server 100 may generate and distribute an analysis model specialized for a specific terminal device. For example, in the early stage of the system in which data is not sufficiently collected, the server 100 generates a general-purpose analysis model that can be universally applied to a plurality of terminal devices 300 based on data collected from a plurality of terminal devices 300 . can be created and distributed. Thereafter, when sufficient data is collected from the specific terminal device, the server 100 may generate and distribute an analysis model specialized for the specific terminal device based on the data collected from the specific terminal device. According to the present embodiment, a customized analysis model specialized for each terminal device 300 may be provided in consideration of the installation environment of the terminal device 300 , its own characteristics, and the like. Accordingly, the accuracy of data analysis may be improved. A more detailed description of the present embodiment will be described later with reference to FIGS. 6 to 15 .

For reference, it should be noted that, in this specification, a general-purpose analysis model is used in a relative meaning with an analysis model specialized for a specific terminal device. For example, the first analysis model may be a general-purpose analysis model when compared to the second analysis model, but may be a specialized analysis model when compared to the third analysis model.

In some embodiments, the server 100 sets affiliation groups for the plurality of terminal devices 300 , generates various update analysis models, and distributes them to the edge devices 200 connected to the plurality of terminal devices 300 . can In this case, the edge device 200 may perform or withhold the update of the existing analysis model based on the performance evaluation result of the existing analysis model and the analysis model for update. Also, the server 100 may adjust the group belonging to the terminal device 300 based on the update result of the analysis model of the edge device 200 . For example, the server 100 may adjust the belonging group so that the terminal devices 300 connected to the edge device 200 using the same analysis model belong to the same group. According to the present embodiment, by adjusting terminal devices having similar characteristics or surrounding environments to belong to the same group, ease of management may be improved. In addition, since an analysis model for update can be generated again by synthesizing data of terminal devices with similar characteristics (ie, terminal devices belonging to the same group) after group membership is adjusted, the accuracy of analysis results is further improved can be A more detailed description of the present embodiment will be described later with reference to FIGS. 16 to 18 .

In some embodiments, as shown in FIG. 2 , the functions of the server 100 may be implemented in the form of a cloud 10 composed of a plurality of computing devices. In the same context, the function of the edge device 200 may also be implemented in the form of a fog consisting of a plurality of computing devices. However, in the following, in order to provide convenience of understanding, the description will be continued by assuming the system configuration shown in FIG. 1 .

A more detailed description of the detailed configuration and operation of the server 100 will be described later with reference to FIGS. 3 and 6 to 18 .

In the data analysis system, the edge device 200 is a computing device that analyzes data collected from the connected terminal device 300 using the analysis model distributed by the server 100 and performs processing according to the analysis result. . Also, the edge device 200 transmits data collected from the connected terminal device 300 to the server 100 . When the data analysis system is built in an Internet of Things (IoT) environment, the edge device 200 may be implemented as an IoT gateway.

In some embodiments, the edge device 200 may generate the analysis model itself without receiving the analysis model from the server 100 . Specifically, in response to determining that the available resources are sufficient, the edge device 200 may generate the analysis model itself without requesting the server 100 to update the analysis model. According to the present embodiment, the load on the server 100 may be reduced, and the resources of the edge device 200 may be efficiently utilized. A detailed description of the present embodiment will be described later in more detail with reference to FIG. 11 .

A more detailed description of the configuration and operation of the edge device 200 will be described later with reference to FIGS. 4 and 6 to 18 .

In the data analysis system, the terminal device 300 is a device that provides data to be analyzed. For example, when the data analysis system is built in an IoT environment, the terminal device 300 may correspond to an IoT device that provides various sensor data.

The components of the data analysis system shown in FIG. 1 may communicate via a network. Here, the network includes all types of wired/wireless networks such as a local area network (LAN), a wide area network (WAN), a mobile radio communication network, and a Wibro (Wireless Broadband Internet). can be implemented.

So far, a data analysis system according to an embodiment of the present invention has been briefly described with reference to FIGS. 1 and 2 . Hereinafter, the configuration and operation of the server 100 and the edge device 200, which are components of the data analysis system, will be described in more detail with reference to FIGS. 3 and 4 .

3 is a block diagram illustrating the server 100 according to an embodiment of the present invention.

Referring to FIG. 3 , the server 100 may include a data collection unit 110 , a data DB 130 , a model management unit 150 , and a device/edge information management unit 170 . However, only the components related to the embodiment of the present invention are shown in FIG. 3 . Accordingly, those skilled in the art to which the present invention pertains can see that other general-purpose components other than those shown in FIG. 3 may be further included. Of course, the server 100 may be implemented in a form in which some of the components shown in FIG. 3 are omitted.

In addition, it should be noted that each of the components of the server 100 shown in FIG. 3 represents functionally separated functional elements, and a plurality of components may be implemented in a form that is integrated with each other in an actual physical environment.

Looking at each component, the data collection unit 110 collects data of the terminal device 300 from the plurality of edge devices 200 . The collected data may be stored in the data DB 130 .

Next, the data DB 130 is a storage for integrated management of data collected from the terminal device 300 .

Next, the model management unit 150 performs overall management of the analysis model. The model management unit 150 may include a model information management unit 151 , a model generation unit 153 , and a model distribution unit 155 .

The model information management unit 150 manages various types of information related to distribution and update of an analysis model, such as a distribution history of the analysis model and analysis model information of each edge device 200 .

Next, the model generator 153 generates various analysis models based on the data stored in the data DB 130 .

Next, the model distribution unit 155 distributes the analysis model generated by the model generation unit 153 to the edge device 200 . Specifically, the model distribution unit 155 determines the terminal device 300 to which the analysis model is applied, and the edge device 200 connected to the determined terminal device 300 based on the management information of the device/edge information management unit 170 . ) as the distribution target.

Next, the device/edge information management unit 170 performs various types of information related to the terminal device 300 and the edge device 200 , such as group information of the terminal device 300 and a connection relationship between the terminal device 300 and the edge device 200 . Manage information

For a detailed description of the components of the server 100 illustrated in FIG. 3 , refer to the description portion of FIGS. 6 to 18 .

4 is a block diagram illustrating an edge device 200 according to an embodiment of the present invention.

Referring to FIG. 4 , the edge device 200 includes a data collection unit 210 , a data transmission unit 220 , an analysis unit 230 , a data DB 240 , a model management unit 250 , and a resource monitoring unit 260 . and a feedback collecting unit 270 . However, only the components related to the embodiment of the present invention are shown in FIG. 4 . Accordingly, those skilled in the art to which the present invention pertains can see that other general-purpose components other than the components shown in FIG. 4 may be further included. Of course, the edge device 200 may be implemented in a form in which some of the components shown in FIG. 4 are omitted.

In addition, it should be noted that each component of the edge device 200 shown in FIG. 4 represents functionally separated functional elements, and a plurality of components may be implemented in a form integrated with each other in an actual physical environment.

The data collection unit 210 collects data from the connected terminal device 300 . Some or all of the collected data may be stored in the data DB 240 .

Next, the data transfer unit 220 transfers the data collected from the connected terminal device 300 to the server 100 . In addition, the data transmitting unit 220 may transmit various data such as feedback information collected by the feedback collecting unit 270 and the analysis result provided by the analyzing unit 230 to the server 100 .

Next, the analysis unit 230 analyzes data collected from the connected terminal device 300 based on the analysis model of the model management unit 250 . The analysis result may also be stored in the data DB 240 .

Next, the data DB 240 is a storage for managing various data such as data collected from the connected terminal device 300 and analysis results of the data.

Next, the model manager 250 performs overall management of the analysis model of the edge device 200 . The model manager 250 may include a model updater 251 , a model evaluator 253 , and a model generator 255 .

The model update unit 251 updates the existing analysis model. Specifically, when the model update unit 251 receives the update analysis model from the server 100 , the model evaluation unit 253 provides the update analysis model and performance evaluation results for the existing analysis model. Also, when the performance of the analysis model for update is better, the model update unit 251 may replace the existing analysis model with the analysis model for update.

Also, the model update unit 251 may increase the update index based on the error of the analysis model, and transmit an update request message to the server 100 in response to determining that the update index is greater than or equal to a threshold value.

Next, the model evaluation unit 253 performs performance evaluation on the analysis model. In this case, accuracy, recall, precision, sensitivity, specificity, F-score (e.g. F1, F2, etc.) are used as metrics of the performance evaluation. can be

Next, the model generator 255 generates an analysis model by itself based on the data stored in the data DB 240 . For example, the model generating unit 255 may receive information on available resources from the resource monitoring unit 260 , and generate the analysis model by itself in response to determining that the available resources are greater than or equal to a threshold.

Next, the resource monitoring unit 260 monitors the resource state of the edge device 200 .

Next, the feedback collecting unit 270 collects feedback on the analysis result of the analyzing unit 230 . In this case, the feedback may be about whether an error (e.g. false positive, false negative error, etc.) exists in the analysis result. The feedback collection unit 270 may use any method for collecting the feedback.

For a detailed description of the components of the edge device 200 shown in FIG. 4 , refer to the description part of FIGS. 6 to 18 .

Each component of FIGS. 3 and 4 may mean software or hardware such as a field programmable gate array (FPGA) or an application-specific integrated circuit (ASIC). However, the above components are not meant to be limited to software or hardware, and may be configured to be in an addressable storage medium or configured to execute one or more processors. The functions provided in the components may be implemented by more subdivided components, or may be implemented as one component that performs a specific function by combining a plurality of components.

5 is a hardware configuration diagram illustrating an exemplary computing device 400 that may implement the server 100 and the edge device 200 according to some embodiments of the present invention.

Referring to FIG. 5 , the computing device 400 loads a computer program executed by one or more processors 410 , a bus 450 , a communication interface 470 , and the processor 410 . ) may include a memory 430 and a storage 490 for storing data analysis software 491 . However, only the components related to the embodiment of the present invention are illustrated in FIG. 5 . Accordingly, those skilled in the art to which the present invention pertains can see that other general-purpose components other than the components shown in FIG. 5 may be further included.

The processor 410 controls the overall operation of each component of the computing device 400 . The processor 410 includes a central processing unit (CPU), a micro processor unit (MPU), a micro controller unit (MCU), a graphic processing unit (GPU), or any type of processor well known in the art. can be In addition, the processor 410 may perform an operation on at least one application or program for executing the method according to the embodiments of the present invention. Computing device 400 may include one or more processors.

The memory 430 stores various data, commands and/or information. The memory 430 may load one or more programs 491 from the storage 490 in order to execute the data analysis method according to embodiments of the present invention.

When the computer program 491 is loaded into the memory 430 , the module shown in FIG. 3 or FIG. 4 may be implemented on the memory 430 . For example, when the computer program 491 is a program that performs an operation of the server 100 , the module of the server 100 illustrated in FIG. 3 may be implemented on the memory 430 . As another example, when the computer program 491 is a program that performs the operation of the edge device 200 , the module of the edge device 200 illustrated in FIG. 4 may be implemented on the memory 430 .

The bus 450 provides communication functions between components of the computing device 400 . The bus 450 may be implemented as various types of buses, such as an address bus, a data bus, and a control bus.

The communication interface 470 supports wired/wireless Internet communication of the computing device 400 . Also, the communication interface 470 may support various communication methods other than Internet communication. To this end, the communication interface 470 may be configured to include a communication module well known in the art.

The storage 490 may non-temporarily store the one or more programs 491 and various data.

The storage 490 is a non-volatile memory, such as a read only memory (ROM), an erasable programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), a flash memory, a hard disk, a removable disk, or well in the art to which the present invention pertains. It may be configured to include any known computer-readable recording medium.

Computer program 491 may include one or more instructions to perform methods according to some embodiments of the present invention. Here, the instruction refers to a series of instructions grouped based on a function, which is a component of a computer program and is executed by the processor 410 . Methods according to some embodiments of the present invention may be performed by loading the computer program 491 into the memory 430 and the processor 410 executing one or more instructions constituting the computer program 491 .

For example, the computer program 491 may generate a general-purpose analysis model based on data collected from a plurality of terminal devices, distribute the general-purpose analysis model to an edge device connected to the plurality of terminal devices, and update a model In response to a condition being satisfied, instructions for generating an analysis model for update based on the data collected from the target terminal device and distributing the analysis model for update to an edge device connected to the target terminal device may include When the computer program 491 is loaded into the memory 430 and the one or more instructions are executed by the processor 410 , the computing device 400 may operate as the server 100 .

For another example, for example, the computer program 491 performs an operation of transmitting data collected from a connected terminal device to the server, an operation of receiving an analysis model for update distributed by the server, an existing analysis model and the and instructions for determining whether to update the existing analysis model based on a performance evaluation result of the analysis model for update. When the computer program 491 is loaded into the memory 430 and the one or more instructions are executed by the processor 410 , the computing device 400 may operate as the edge device 200 .

So far, the configuration and operation of an exemplary computing device 400 capable of implementing the server 100 or the edge device 200 according to an embodiment of the present invention has been described with reference to FIG. 5 . Hereinafter, methods performed in the data analysis system according to an embodiment of the present invention will be described.

For convenience of description, the description of the operating subject of each step included in the method to be described below may be omitted. In addition, each step of the method may be implemented with instructions of a computer program executed by a processor.

6 is a flowchart illustrating a method for updating an analysis model according to an embodiment of the present invention. However, this is only a preferred embodiment for achieving the object of the present invention, and it goes without saying that some steps may be added or deleted as necessary.

Referring to FIG. 6 , the method for updating the analysis model starts with the server 100 collecting data from at least one edge device 200 ( S10 ). The data is data collected from the terminal device 300 connected to the edge device 200 . That is, in this step S10 , each edge device 200 collects data from the connected terminal device 300 , and transmits the collected data to the server 100 .

For reference, the number of edge devices may vary according to embodiments. Also, it should be noted that one edge device need not be implemented as one physical device, and multiple edge devices may be implemented in one physical device.

In step S20, the server 100 generates an analysis model based on the collected data. For example, the server 100 may generate an analysis model through machine learning on the collected data, but the scope of the present invention is not limited thereto.

In this step (S20), the server 100 determines whether generation of an analysis model is possible based on the number, quality, ratio of data by class, etc. of the collected data, and generates an analysis model in response to the determination that generation is possible can do.

In step S30 , the server 100 distributes the generated analysis model to the edge device 200 . Specifically, as shown in FIG. 7 , the server 100 determines a distribution target edge device from among a plurality of edge devices 200 ( S31 ), and distributes the analysis model to the distribution target edge device ( S33 ).

In some embodiments, the server 100 may determine, as a distribution target, an edge device connected to a terminal device that has provided basic data of the analysis model. For example, when a first analysis model is generated based on data collected from a plurality of first terminal devices, the server 100 may distribute the first analysis model to an edge device connected to the first plurality of terminal devices. have.

In step S40, the server 100 generates an analysis model for update in response to a predetermined model update condition being satisfied.

In some embodiments, the analysis model for update may be an analysis model specialized for a specific terminal device. For example, the server 100 generates a general-purpose analysis model based on data collected from a plurality of terminal devices in step S20, and collects data collected from some of the plurality of terminal devices in step S40. Based on the data, it is possible to generate an analysis model specialized for some of the terminal devices. According to the present embodiment, the edge device 200 will analyze the data collected from the some terminal devices using the specialized analysis model, so that the accuracy of data analysis can be improved.

In some embodiments, the analytic model for updating may be based on recent data. For example, the server 100 generates a first analysis model by learning the data collected from the terminal device during the first period in step S20, and the data collected after the first period in step S40 may update the first analysis model by performing additional learning. Alternatively, the server 100 may generate a second analysis model by newly learning the data collected after the first period.

In some embodiments, at least one element of a hyperparameter and a learning algorithm of the analysis model for update may be different from an existing analysis model. For example, the server 100 may generate a first analysis model using the first learning algorithm in step S20 and generate a second analysis model using the second learning algorithm in this step S40. have.

In some embodiments, the analysis model for update may be generated by a combination of the above-described embodiments. For example, the server 100 generates an analysis model specialized for the partial terminal device based on recent data collected from some terminal devices, or applies a learning algorithm different from the existing analysis model to the collected recent data to apply the partial It is possible to create an analysis model specialized for a terminal device. Also, the server 100 may generate and distribute a plurality of update analysis models according to the above-described embodiments.

Meanwhile, the above-described model update condition may vary according to embodiments.

In some embodiments, the model update condition may be to receive an update request message from the edge device 200 . For example, the server 100 may generate an analysis model for update in response to receiving the update request message from the edge device 200 . A detailed description of an operation of the edge device 200 transmitting the update request message will be described later with reference to FIG. 10 .

In some embodiments, the model update condition may be a condition preset in the server 100 . That is, the operation of generating the analysis model for update may be triggered by the server 100's own determination regardless of the request of the edge device 200 . For example, the server 100 may generate and distribute an analysis model based on recent data according to a specified period. In another example, the server 100 generates a first analysis model specialized for the first terminal device in response to determining that the data collected from the first terminal device is greater than or equal to a threshold, and the data collected from the second terminal device A second analysis model specialized for the second terminal device may be generated and distributed in response to the determination that the threshold is higher. As another example, the server 100 collects performance evaluation results for an analysis model from a plurality of edge devices 200 , and when the performance of a specific analysis model is less than a threshold, an update analysis for the specific analysis model You can create a model. In this case, the server 100 may distribute the update analysis model to the edge device 200 using the specific analysis model.

In some embodiments, the model update condition may be determined based on a combination of the above-described embodiments.

Meanwhile, in some embodiments, the detailed process of this step S40 may be performed according to the flowchart shown in FIG. 8 . Specifically, when an update request message is received from the edge device 200 , the server 100 determines whether an analysis model more specialized than the existing analysis model being used in the edge device 200 can be generated ( S41 , S43 ). Here, the analysis model more specialized than the existing analysis model means an analysis model generated based on data collected from a smaller number of terminal devices than the existing analysis model. For example, the existing analysis model of the edge device 200 is generated based on data of a plurality of terminal devices including the target terminal device connected to the edge device 200 , and the data data collected from the target terminal device is equal to or greater than a threshold value. In this case, the server 100 may determine that it is possible to generate an analysis model specialized for the target terminal device. Also, in response to the generation possibility determination, the server 100 may generate an analysis model specialized for the target terminal device. In the opposite case, the server 100 may update the existing analysis model of the edge device 200 based on the latest data (S45 and S47).

In step S50 , the server 100 distributes the analysis model for update to the edge device 200 . Specifically, the server 100 may determine a distribution target edge device from among the plurality of edge devices 200 , and distribute an analysis model for update to the distribution target edge device.

In some embodiments, the server 100 may determine the edge device 200 that has transmitted the update request message as a distribution target.

In some embodiments, the server 100 may determine, as a distribution target, the edge device 200 connected to the terminal device that has provided the basic data of the analysis model for update. For example, when a first analysis model specialized for a specific terminal device 300 is generated based on data collected from the specific terminal device 300 , the server 100 may be configured to connect the edge device 200 to the specific terminal device 300 . may be determined as a distribution target of the first analysis model.

In some embodiments, the server 100 may determine a distribution target based on a performance evaluation result of the analysis model collected from the plurality of edge devices 200 . For example, the server 100 may determine the edge device 200 using the analysis model whose performance is less than a threshold as a distribution target.

In some embodiments, the server 100 may randomly determine a distribution target from among the plurality of edge devices 200 . For example, the server 100 may generate various analysis models for update in order to provide the edge device 200 with an opportunity to select an optimal analysis model, and randomly distribute the generated analysis models for update.

In some embodiments, a distribution target may be determined based on a combination of the above-described embodiments.

In step S60 , the edge device 200 performs or withholds the update of the existing analysis model in response to the receiving of the analysis model for update. The detailed process of this step (S60) is shown in FIG. As shown in FIG. 9 , the edge device 200 performs performance evaluation on the existing analysis model and the analysis model for update in response to receiving the analysis model for update ( S61 and S63 ). Also, the edge device 200 updates the existing analysis model in response to the determination that the performance of the analysis model for update is better ( S65 and S67 ). For example, the edge device 200 may replace an existing analysis model with an analysis model for update.

In step S70 , the edge device 200 receiving the analysis model for update reports the update performance result to the server 100 .

Among the above-described steps (S10 to S70), step (S10) is performed by the data collection unit 110, steps (S20, S40) are performed by the model generation unit 153, and steps (S30, S50) may be performed by the model distribution unit 155 . Also, steps S60 and S70 may be performed by the model updater 251 and the model evaluator 253 .

So far, an analysis model update method according to an embodiment of the present invention has been described with reference to FIGS. 6 to 9 . According to the above-described method, a customized analysis model specialized for each terminal device 300 may be provided in consideration of the installation environment of the terminal device 300 , its own characteristics, and the like. Accordingly, the accuracy of data analysis may be improved.

10 is a flowchart illustrating a method for requesting an analysis model update by the edge device 200 according to an embodiment of the present invention. However, this is only a preferred embodiment for achieving the object of the present invention, and it goes without saying that some steps may be added or deleted as necessary.

Referring to FIG. 10 , the analysis model update request method starts in step S110 in which the edge device 200 receives a feedback on a prediction result of the analysis model. Specifically, the edge device 200 may analyze data received from the terminal device 300 through an analysis model, output a prediction result, and receive feedback on whether an error exists in the prediction result. In this case, the feedback may be provided in any way.

For example, it is assumed that the analysis model is a model for predicting whether the terminal device 300 is abnormal. Also, it is assumed that the edge device 200 performs a control operation such as shutting off the power of the terminal device 300 or providing an alarm to the manager when the state of the terminal device 300 is predicted to be abnormal. Then, the edge device 200 may receive feedback on the control operation from the manager or the terminal device. For example, when the manager checks the terminal device 300 and there is no abnormality, it is possible to provide feedback that there is an error in the prediction result. As another example, the terminal device 300 may check the state by itself through a built-in test, and if there is no abnormality, it may provide feedback indicating that there is an error in the prediction result.

In step S130 , the edge device 200 determines whether an error exists in the prediction result based on the received feedback. In response to determining that an error exists, the next step S150 may be performed.

In steps S150 to S190 , the edge device 200 may increase the update index, and may transmit an update request message to the server 100 in response to determining that the increased update index is equal to or greater than the threshold.

However, in some embodiments, when the error identified through the feedback corresponds to a preset serious error, the edge device 200 may directly transmit an update request message to the server 100 irrespective of the update index. For example, when a false positive error is a serious error in the corresponding system, the edge device 200 may immediately transmit an update request message to the server 100 if the error corresponds to a false positive error. For another example, when a false negative error is a serious error in the corresponding system, the edge device 200 may immediately transmit an update request message to the server 100 when the error corresponds to a false negative error.

In some embodiments, the edge device 200 may store and utilize a data set in which a false positive error or a false negative error obtained through feedback is generated. For example, the edge device 200 may use the dataset as data for evaluation. Specifically, when performing performance evaluation on the existing analysis model and the analysis model for update (e.g. step S63), the edge device 200 may perform performance evaluation using the dataset. Here, the edge device 200 may perform performance evaluation with a higher evaluation weight on a dataset associated with a serious error. As another example, the edge device 200 may transmit the data set to the server 100 to be utilized for generating an analysis model. Specifically, the server 100 may assign a higher learning weight to the dataset, perform learning based on the assigned learning weight, and generate and distribute an analysis model for update. For example, when receiving a dataset for error information from the first edge device, the server 100 may learn the dataset with a higher learning weight to generate an update analysis model. Also, the server 100 may distribute the update analysis model to the first edge device. Here, the update analysis model will be more robust against corresponding errors compared to the existing analysis model, so that the accuracy of data analysis can be further improved.

In the aforementioned steps S110 to S190 , step S110 may be performed by the feedback collector 270 , and the remaining steps S130 to S190 may be performed by the model updater 251 .

So far, a method for requesting an analysis model update according to an embodiment of the present invention performed in the edge device 200 has been described with reference to FIG. 10 . Hereinafter, a method in which the edge device 200 generates an analysis model by itself will be described with reference to FIG. 11 .

11 is a flowchart illustrating a method of generating an analysis model by the edge device 200 by itself according to an embodiment of the present invention. However, this is only a preferred embodiment for achieving the object of the present invention, and it goes without saying that some steps may be added or deleted as necessary.

Referring to FIG. 11 , the method for generating the analysis model starts at the edge device 200 monitoring a resource ( S210 ).

In step S230, the edge device 200 determines whether the available resources are sufficient. For example, the edge device 200 may determine that the available resources are sufficient when the occupancy of the processor is continuously less than the threshold for a predetermined time. However, the scope of the present invention is not limited thereto. In response to determining that the available resources are sufficient, the next step S250 may be performed.

In step S250 , the edge device 200 generates an analysis model for update based on data collected from the connected terminal device 300 , and updates the existing analysis model. As described above, the edge device 200 may perform the update only when the performance of the analysis model for update is superior to that of the existing analysis model.

In some embodiments, when the data collected from the connected terminal device is not sufficient, the edge device 200 secures additional data through the server 100, and further uses the additional data to generate an analysis model for update. can For example, when receiving a request for additional data from the first edge device, the server 100 may provide data of a terminal device similar to the first terminal device connected to the first edge device to the first edge device. In this case, the similar terminal device may be a terminal device belonging to the same group as the first terminal device, a terminal device disposed at a nearby location, or the like. Alternatively, when a plurality of terminal devices 300 form a hierarchical group as shown in FIG. 12 , in the hierarchical structure, a terminal device of a group located at a close distance to a group belonging to the first terminal device is similar to the group. It may be a terminal device.

In step S270 , the edge device 200 reports the update performance result to the server 100 .

In some embodiments, the edge device 200 may provide the analysis model generated in step S250 to the server 100 together with the update performance result. The server 100 may share the analysis model generated by the edge device 200 with the edge device by distributing the provided analysis model to other edge devices.

Among the above-described steps S210 to S270 , step S210 may be performed by the resource monitoring unit 260 , and steps S230 , S250 , and S270 may be performed by the model management unit 250 .

A method of directly generating an analysis model in the edge device 200 has been described with reference to FIG. 12 . According to the above-described method, the edge device 200 generates an analysis model using its available resources without relying on the server 100 , and shares the generated analysis model with other edge devices through the server 100 . can do. Accordingly, the load on the server 100 may be reduced, and resources of the edge device 200 may be efficiently utilized.

Hereinafter, in order to provide more convenience of understanding, a process of updating an analysis model in the data analysis system having a hierarchical structure shown in FIGS. 12 to 15 will be described. In particular, it is assumed that the server 100 gradually distributes the specialized analysis model in FIGS. 12 to 15 . Also, FIGS. 12 to 15 exemplify a case in which a plurality of terminal devices 1 to 8 are managed as hierarchical groups 501 to 527 . However, it is not necessary that the plurality of terminal devices 1 to 8 be hierarchically grouped as shown in FIG. 12 in order to apply the technical spirit of the present invention.

For reference, in the drawings below, the display of the analysis model adjacent to the group nodes (e.g. 521 and 523) means that all edge devices belonging to the corresponding group use the corresponding analysis model. In addition, the display of the analysis model adjacent to the edge device node means that only the corresponding edge device is using the corresponding analysis model. In addition, the identifiers (e.g. 1 and 1-2) displayed on the analysis model mean that the corresponding analysis model is generated based on data of a terminal device belonging to a group matching the identifier.

12 to 15 and 16 and 17, a group is referred to as a reference number indicated on a group node, and the edge device and the terminal device are numbered 1 to 8 indicated on the edge device node and the terminal device node. It is indicated by adding a "#" mark to it. For example, group 1 may be referred to as group 551 and edge device 1 may be referred to as edge device #1. In addition, for convenience of explanation, data collected from a terminal device belonging to a specific group may be summarized and described as data of the specific group, and distributing the analysis model to the edge device belonging to the specific group is the analysis model to the specific group It can be summarized as distributing

Referring to FIG. 12 , a plurality of terminal devices #1 to #8 form hierarchical groups 501 to 527 . Such a group may be manually set by an administrator for convenience of management, or may be automatically set based on the arrangement position of the terminal device (or edge device), the type of the terminal device (or edge device), or the like.

For reference, the edge devices #1 to #8 shown in FIG. 12 correspond to logical devices, and a plurality of edge devices may be implemented as one physical device. In this case, the one physical edge device may belong to a plurality of groups. Here, that the edge device belongs to the group means that the terminal device connected to the edge device belongs to the group.

As shown in FIG. 12 , all terminal devices #1 to #8 belong to the group 501 of the uppermost layer, and five terminal devices (#) are included in the group 511 located in the lower layer of the group 501 . 1 to #5), and three terminal devices #1 to #3 belong to the group 521 located in the lower layer of the group 511.

13 illustrates that the server 100 distributes a general-purpose analysis model 531 that can be applied to a plurality of terminal devices #1 to #8. Specifically, the server 100 may generate and distribute the general-purpose analysis model 531 based on the data of the highest group 501 . In this case, the distribution target of the general-purpose analysis model 531 may be the edge devices #1 to #8 belonging to the group 501 . Since the general-purpose analysis model 531 is generated based on data collected from a plurality of terminal devices #1 to #8, it may be somewhat inaccurate in analyzing data of a specific terminal device.

14 illustrates that the server 100 distributes the analysis model 541 specialized for some terminal devices (#6 to #8) for update.

For example, when it is determined that the data of the group 513 is sufficiently collected, the server 100 may generate the analysis model 541 specialized for the terminal devices #6 to #8. Also, the server 100 may distribute the analysis model 541 to the edge devices #6 to #8 belonging to the group 513 .

As another example, the server 100 receives an update request message from the edge devices #6 to #8 belonging to the group 513 and generates a specialized analysis model 541 in response to the reception of the update request message. can do. Specifically, when the server 100 receives the update request message from the edge device #6, the server 100 checks the analysis model information of the edge device #6. In addition, the server 100 determines whether a more specialized analysis model can be generated than the analysis model 531 of the edge device #6 based on the checked analysis model information. Here, since the analysis model 531 is a model corresponding to the uppermost layer (layer 1), the specialized analysis model becomes an analysis model corresponding to the lower layer (layer 2 or 3) of the uppermost layer. In detail, the specialized analysis model may be a model generated based on data of the group 513 or 525 . Accordingly, the server 100 generates the analysis model 541 based on the data collected from the terminal devices #6 to #8, and distributes the analysis model 541 to the edge devices #6 to #8. . In this case, the edge devices #7 and #8 receive the update analysis model 541 by the update request of the other edge devices #6 even though the update request message is not transmitted. For reference, if the data collected from the terminal devices #6 to #8 is not sufficient, the server 100 updates the existing analysis model 531 of the edge device #6 based on the latest data, and updates The analyzed model can also be distributed to edge devices (#6 to #8).

15 shows the model update results of the edge devices #6 to #8.

Referring to FIG. 15 , each edge device (#6 to #8) performs performance evaluation on the existing analysis model 531 and the analysis model for update 541, and based on the performance evaluation result, the existing analysis model 531 ) can be determined whether to perform the update. In particular, FIG. 15 exemplifies a case where the edge devices #6 and #8 decide to perform the model update, and the edge device 7 decides not to perform the model update.

In some embodiments, the server 100 may receive the performance evaluation result for the analysis model 541 for update from the edge devices #6 to #8. In this case, when the performance of the analysis model 541 for update is equal to or greater than the threshold, the server 100 may additionally distribute the analysis model 541 for update to other edge devices. For example, the server 100 may update the analysis model 541 for updating to the edge devices #1 to #5 of the group (e.g. 511) located close to the group 513 to which the edge devices #6 to #8 belong on the hierarchical structure. ) can be further distributed. Alternatively, the server 100 may additionally distribute the analysis model 541 for update to any edge device.

Up to now, a process in which a model update is performed in a data analysis system having a hierarchical structure has been described with reference to the examples shown in FIGS. 12 to 15 . According to the above description, a general-purpose analysis model can be quickly generated and distributed based on data collected from a plurality of terminal devices at the initial stage of the system. In addition, as data is accumulated from the plurality of terminal devices, more and more specialized analysis models are distributed for update, so that the accuracy of data analysis can be improved as time goes by. Furthermore, the edge device selects an optimal analysis model suitable for the connected terminal device by comparing the performance of the existing analysis model and the analysis model for update, so that overall system performance can be further improved.

Hereinafter, a method of adjusting a group belonging to the terminal device 300 based on the update result of the analysis model will be described with reference to FIGS. 16 to 18 .

16 is a flowchart illustrating a device group adjustment method according to an embodiment of the present invention. However, this is only a preferred embodiment for achieving the object of the present invention, and it goes without saying that some steps may be added or deleted as necessary.

Referring to FIG. 16 , the device group adjustment method starts with the step ( S310 ) of the server 100 setting the group belonging to the plurality of terminal devices 300 . For example, the server 100 may set a group belonging to the plurality of terminal devices 300 based on the selection of the administrator. As another example, the server 100 may set the group belonging to the plurality of terminal devices 300 based on the arrangement position of the terminal device 300 or the edge device 200 .

In steps S320 and S330, the server 100 generates and distributes an analysis model for update based on the affiliated group. For example, the server 100 generates a first analysis model for update based on data collected from the first terminal device belonging to the first group, and converts the first analysis model for update to an edge device belonging to the first group. can be distributed. For the description of these steps (S320, S330), refer to the above-described steps (S40, S50) further.

In step S340 , the edge device 200 performs or withholds the update of the existing analysis model. That is, in this step S340 , the first edge device may update the existing analysis model, and the second edge device may withhold the update. Accordingly, as the update progresses, the analysis model used by each edge device 200 may gradually change from each other. For the description of this step (S340), refer to the above-described step (S60) further.

In step S350 , the edge device 200 reports the update performance result. For the description of this step (S350), refer to the above-described step (S70) further.

In step S360 , the server 100 adjusts the group belonging to the plurality of terminal devices 300 based on the update execution results received from the plurality of edge devices 200 . Specifically, the server 100 determines that the terminal devices 300 connected to the edge device 200 using the same analysis model based on the update performance result (e.g., the currently used analysis model information) of the edge device 200 are in the same group. You can adjust it to fit in.

The reason for adjusting the belonging group based on the analysis model is that the terminal devices 300 to which the same analysis model is applied have a high probability of having similar characteristics. For example, when the analysis model is a model for determining whether the analysis model is abnormal, it means that the terminal devices 300 to which the same analysis model is applied have very similar criteria for determining whether the analysis model is abnormal. Accordingly, when a group is formed among the terminal devices 300 having similar characteristics, the convenience of managing the bar can be increased for collective management. In addition, the server 100 can generate and distribute an analysis model for update based on data of the terminal devices 300 having similar characteristics, so that the accuracy of data analysis can be further improved.

In some embodiments, the server 100 may adjust the affiliation group of the terminal device 300 in response to a preset condition being satisfied. For example, the server 100 may adjust the group belonging to the plurality of terminal devices 300 every preset period. For another example, when the analysis model of the edge device 200 belonging to the same group is changed by a certain level or more, the server 100 may adjust the belonging of the terminal device 300 belonging to the corresponding group.

So far, a device group adjustment method according to an embodiment of the present invention has been described with reference to FIG. 16 . Hereinafter, in order to provide more convenience of understanding, a process in which a device group is adjusted in a data analysis system having a hierarchical structure will be described with reference to FIGS. 17 and 18 .

17 illustrates an analysis model update result according to some embodiments described above. In FIG. 17 , the shaded edge devices #1 to #3 and #7 mean edge devices connected to the terminal devices #1 to #3 and #7 that are the target of group adjustment.

Referring to FIG. 17 , an edge device #3 uses a different analysis model from other edge devices #1 and #2 belonging to the same group 571 . For example, the edge device #3 may generate and update an analysis model by itself. In addition, the analysis model generated by the edge device (#3) would have been distributed to other edge devices (#1, #2) through the server 100, but the edge devices (#1, #2) are the existing analysis models. It is possible that the update was not performed due to better performance.

The edge device #7 belonging to the group 577 uses a more general analysis model than the edge device #8. In this case, the analysis model for update generated based on the data of the group 577 was distributed to the edge devices #7 and #8 belonging to the group 577, but the edge device #7 had poor performance of the existing analysis model. It may be better that you haven't done the update. Also, the edge device #8 may have performed the update because the performance of the analysis model for update is better.

As shown in FIG. 17 , as the model update is performed, it can be seen that the analysis models of edge devices belonging to the same group are gradually changed from each other. In this case, the server 100 may adjust the group belonging to the terminal devices #1 to #8 connected to the edge devices #1 to #8 based on the analysis model.

18 illustrates a result of the server 100 adjusting the group belonging to the terminal devices #1 to #8.

As shown in FIG. 18 , when the analysis model of the edge devices #1 to #3 within the group 571 is changed, new groups 581 and 583 may be created in the lower layer of the corresponding group 571 . have. According to the group adjustment result, the server 100 may generate an analysis model for update based on the data of the group 581 and distribute the generated analysis model for update to the edge devices #1 and #2. there will be Here, the update analysis model is a more specialized model than the analysis model of the edge devices #1 and #2, so that the accuracy of data analysis can be further improved as a result of group adjustment.

In the case of the edge devices #4 and #5, since the same analysis model is used, the server 100 does not adjust the group belonging to the terminal devices #4 and #5.

Since the edge device #6 belongs to the group 575 alone, there is no need to separately perform group adjustment for the terminal device #6.

Since the edge devices #7 and #8 belonging to the group 577 use different analysis models, the server 100 adjusts the group belonging to the terminal devices #7 and #8. Specifically, the server 100 confirms that the group belonging to the edge device #6 using the same analysis model as the edge device #7 is the group 575, and the terminal device #7 is the group 575. adjust to belong to According to this group adjustment result, the server 100 may generate an analysis model for update based on the data of the group 575 and distribute the generated analysis model for update to the edge devices (#6, #7). there will be Here, the update analysis model is a more specialized model than the analysis model of the edge devices #6 and #7, so that the accuracy of data analysis can be further improved as a result of group adjustment.

On the other hand, there may be a case where a new terminal device or a new edge device dynamically participates in the data analysis system. For example, it is assumed that a new terminal device participates. In this case, the server 100 may distribute a plurality of pre-generated analysis models in response to the request for participation of the new terminal device to a specific edge device connected to the new terminal device. Next, the server 100 may obtain a performance evaluation result for the plurality of pre-generated analysis models from the specific edge device. Specifically, the specific edge device may perform a performance evaluation on each of the plurality of pre-generated analysis models with the data of the new terminal device, and transmit the performance evaluation result to the server 100 . Then, the server 100 may determine the affiliation group of the new terminal device based on the obtained performance evaluation result. For example, when the performance of the first analysis model is the best among the plurality of previously generated analysis models, the server 100 may join the new terminal device to a group to which the first analysis model is applied. Even when a new edge device participates dynamically, a membership group may be determined according to a process similar to the above.

So far, the process of adjusting the device group in the data analysis system having a hierarchical structure has been described with reference to FIGS. 17 and 18 . Hereinafter, an example in which the technical idea of the present invention is utilized in an IoT platform will be briefly mentioned with reference to FIG. 19 . In particular, FIG. 19 exemplifies that a smart building management system is built based on the IoT platform.

19 , the smart building management system may include a server 600 , a plurality of IoT gateways 611 , 631 , and 651 , and a plurality of IoT devices 613 , 633 , 663 . In this case, the first IoT gateway 611 and the first plurality of IoT devices 613 are installed in the first building 610 , and the second IoT gateway 631 and the second plurality of IoT devices are installed in the second building 630 . Devices 633 may be installed. Also, a third IoT gateway 651 and a plurality of third IoT devices 653 may be installed in the third building 650 .

The server 600 may correspond to the server 100 according to the above-described embodiments of the present invention. Each of the plurality of IoT gateways 611 , 631 , and 651 may correspond to the edge device 200 according to the above-described embodiments of the present invention. The plurality of IoT devices 613 , 633 , and 663 may correspond to the aforementioned terminal device 300 .

The functions provided by the smart building management system may vary according to the type of analysis model.

For example, the smart building management system analyzes sensor data collected from the IoT devices 613 , 633 , and 663 to predict the energy use of the buildings 610 , 630 , and 650 , and to reduce the energy use of the building IoT. It may be a system for controlling the devices 613 , 633 , 653 or various facilities. In this case, the server 600 may distribute the analysis model for predicting the energy use of the building by analyzing the collected sensor data to each IoT gateway 611 , 631 , and 651 .

As another example, the smart building management system may be a system that detects and responds to emergency situations such as malfunction of equipment in a building in real time. In this case, the server 600 analyzes the sensor data collected from the IoT devices 613 , 633 , and 653 to generate an analysis model for predicting whether the IoT devices 613 , 633 , 653 or various facilities are abnormal for each IoT gateway. (611, 631, 651).

According to some embodiments of the present invention, data analysis and prediction may be performed at the IoT gateways 611 , 631 , and 651 rather than data analysis and prediction at the server 600 end. Accordingly, the load on the server 600 is greatly reduced, and data analysis and prediction are performed quickly, so that an immediate response to an abnormal situation can be made.

In addition, according to some embodiments of the present invention, the server 600 may gradually distribute an update analysis model specialized for each IoT device 613 , 633 , 653 to the IoT gateways 611 , 631 , 651 . . Accordingly, the accuracy of data analysis may be further improved. For example, even with the same IoT device, the analysis criteria of IoT device data may be slightly different due to the environmental impact of the installed buildings 610 , 630 , and 650 . Therefore, even for the same IoT device, when a customized analysis model is distributed in consideration of the surrounding environment, the accuracy of data analysis may be improved.

In addition, according to some embodiments of the present invention, an initial group for the IoT devices 613 , 633 , and 653 is set, and the server 600 is based on the model update result of the IoT gateways 611 , 631 , 651 . A group belonging to the IoT devices 613 , 633 , and 653 may be adjusted. Accordingly, as the IoT devices 613, 633, and 653 having similar characteristics are newly formed as a group, the ease of management can be increased. 633, 653) data are synthesized to generate and redistribute an analysis model for update, so that the accuracy of overall data analysis can be further improved.

Up to now, an example in which the technical idea of the present invention is utilized in an IoT platform has been briefly reviewed with reference to FIG. 19 .

So far, some embodiments of the present invention and effects according to the embodiments have been described with reference to FIGS. 1 to 19 . Effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the following description.

The concept of the present invention described so far with reference to FIGS. 1 to 19 may be implemented as computer readable code on a computer readable medium. The computer-readable recording medium may be, for example, a removable recording medium (CD, DVD, Blu-ray disk, USB storage device, removable hard disk) or a fixed recording medium (ROM, RAM, computer-equipped hard disk). can The computer program recorded in the computer-readable recording medium may be transmitted to another computing device through a network such as the Internet and installed in the other computing device, thereby being used in the other computing device.

In the above, even though all the components constituting the embodiment of the present invention are described as being combined or operating in combination, the present invention is not necessarily limited to this embodiment. That is, within the scope of the object of the present invention, all of the components may operate by selectively combining one or more.

Although acts are shown in a particular order in the drawings, it should not be understood that the acts must be performed in the specific order or sequential order shown, or that all depicted acts must be performed to obtain a desired result. In certain circumstances, multitasking and parallel processing may be advantageous. Moreover, the separation of the various components in the embodiments described above should not be construed as necessarily requiring such separation, and the program components and systems described may generally be integrated together into a single software product or packaged into multiple software products. It should be understood that there is

Although embodiments of the present invention have been described above with reference to the accompanying drawings, those of ordinary skill in the art to which the present invention pertains can realize that the present invention can be embodied in other specific forms without changing the technical spirit or essential features. can understand Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. The protection scope of the present invention should be construed by the following claims, and all technical ideas within the equivalent range should be construed as being included in the scope of the present invention.

Claims (17)

a server for generating and distributing an analysis model for update based on data collected from at least some of a plurality of terminal devices forming a plurality of groups; and
It is connected to at least some of the plurality of terminal devices, and in response to receiving the analysis model for update from the server, performs performance evaluation on the existing analysis model and the analysis model for update, based on the performance evaluation result A plurality of edge devices that determine whether to perform an update to the existing analysis model,
The server is
based on a result of performing the update of the analysis model of the plurality of edge devices, characterized in that at least a part of the group belonging to the plurality of terminal devices is adjusted,
Edge computing based data analysis system. The method of claim 1,
The server is
Selecting an edge device that has been updated with the analysis model for update from among the plurality of edge devices, and adjusting the belonging group of the plurality of terminal devices so that the terminal devices connected to the selected edge devices belong to the same group to do,
Edge computing based data analysis system. The method of claim 1,
The existing analysis model is a model generated based on data collected from all of the plurality of terminal devices,
The update analysis model is a model generated based on data collected from some of the plurality of terminal devices,
Edge computing based data analysis system. The method of claim 1,
The existing analysis model is a model generated based on data collected during a first period from at least some of the plurality of terminal devices,
The analysis model for update is a model generated based on data collected after the first period from at least some of the plurality of terminal devices,
Edge computing based data analysis system. The method of claim 1,
The plurality of groups are configured in a hierarchical structure,
A first group among the plurality of groups includes a terminal device belonging to a second group that is a lower hierarchical group of the first group,
The server is
Distributing a first update analysis model generated based on data collected from the first group of terminal devices to an edge device connected to the first group of terminal devices among the plurality of edge devices,
Distributing a second update analysis model generated based on data collected from the second group of terminal devices to an edge device connected to the second group of terminal devices among the plurality of edge devices,
Edge computing based data analysis system. The method of claim 1,
The server is
In response to a request for participation of a new terminal device, a plurality of pre-generated analysis models are distributed to a first edge device connected to the new terminal device, and performance evaluation of the plurality of analysis models generated from the first edge device Obtaining a result, characterized in that the group belonging to the new terminal device is determined based on the obtained performance evaluation result,
Edge computing based data analysis system. The method of claim 1,
A first edge device among the plurality of edge devices,
Perform performance evaluation on the existing analysis model and the update analysis model using the previously stored evaluation data,
The pre-stored evaluation data is characterized in that it includes data in which a false positive error or a false negative error occurs,
Edge computing based data analysis system. The method of claim 1,
A first edge device among the plurality of edge devices,
Directly generating an analysis model for a first update based on data collected from a terminal device connected to the first edge device in response to determining that the available resources are above a threshold,
Edge computing based data analysis system. 9. The method of claim 8,
The first edge device,
characterized in that the first update analysis model is shared with a second edge device among the plurality of edge devices through the server,
Edge computing based data analysis system. a server for generating and distributing an analysis model based on data collected from at least some of a plurality of terminal devices including a target terminal device; and
an edge device connected to the target terminal device, collecting data from the target terminal device, transmitting the data to the server, and analyzing data received from the target terminal device using an analysis model distributed by the server ,
The server is
A general-purpose analysis model associated with the plurality of terminal devices is generated based on the data collected from the plurality of terminal devices and distributed to the edge device, and data collected from the target terminal device in response to a determination that a model update condition is satisfied Based on the generation of an analysis model for update associated with the target terminal device, characterized in that the distribution to the edge device,
Edge computing based data analysis system. 11. The method of claim 10,
The edge device,
transmits the first data in which a false positive error or a false negative error occurs to the server by the general-purpose analysis model;
The server is
Characterized in that a higher weight is given to the first data among the data collected from the target terminal device, and the analysis model for the update is generated by performing learning based on the assigned weight,
Edge computing based data analysis system. 11. The method of claim 10,
The model update condition is that an update request message for the universal analysis model is received from the edge device,
The edge device,
applying the data received from the target terminal device to the general-purpose analysis model to output a prediction result, increasing the update index in response to determining that an error exists in the prediction result, and determining that the increased update index is greater than or equal to a threshold Characterized in transmitting the update request message to the server in response to,
Edge computing based data analysis system. 13. The method of claim 12,
The edge device,
In response to determining that the error corresponds to a preset specific error, transmitting the update request message to the server regardless of the update index,
Edge computing based data analysis system. 11. The method of claim 10,
The model update condition is that an update request message for the universal analysis model is received from the edge device,
The general-purpose analysis model is an analysis model generated based on data collected during a first period from the plurality of terminal devices,
The server is
When the update request message is received, it is determined whether the amount of data collected from the target terminal device is less than a threshold, and in response to the determination that the amount of data is less than the threshold, collected from the plurality of terminal devices after the first period updating the general-purpose analysis model based on data, and distributing the updated general-purpose analysis model to the edge device,
Edge computing based data analysis system. 11. The method of claim 10,
The plurality of terminal devices belong to a first group among a plurality of groups grouped in a hierarchical structure,
The target terminal device is characterized in that it belongs to a second group that is a lower hierarchical group of the first group,
Edge computing based data analysis system. 16. The method of claim 15,
The data analysis system further includes other edge devices connected to some of the plurality of terminal devices in addition to the edge devices,
The server is
Distributing the general-purpose analysis model also to the other edge devices, and distributing the analysis model for update only to the edge devices,
Edge computing based data analysis system. a communication interface for communicating with a plurality of edge devices connected to at least some of the plurality of terminal devices including the target terminal device;
a memory storing one or more instructions; and
By executing the one or more instructions,
Generating a general-purpose analysis model based on data collected from the plurality of terminal devices, and distributing the general-purpose analysis model to an edge device connected to the plurality of terminal devices, in response to a target model generation condition being satisfied, the target terminal A processor for generating a target analysis model based on data collected from a device and distributing the target analysis model to a target edge device connected to the target terminal device,
server.

Images