KR102472368B1 - Apparatus and method for constructing big data to share production information between multiple companies producing one finished product in the textile industry - Google Patents

Abstract

The big data construction apparatus of the present invention includes a communication unit for communication with terminals of a plurality of companies producing one finished product, a storage unit for integrating and storing production information of the plurality of companies, and the plurality of terminals through the communication unit. When receiving an access request message from any one of the terminals, the access processing unit performs authentication on whether the terminal has permission to access the production information of the plurality of companies based on the received access request message. And, as a result of the authentication, if the authentication is successful, a work space, which is a volatile memory space of a size corresponding to the terminal, is allocated, and among the production information of the storage unit, production information corresponding to a job performed by a company to which the terminal belongs is assigned and a work management unit including a work assignment unit for copying and loading the assigned work space and then transmitting an access response message including an access address of the assigned work space to the terminal through the communication unit.

Description

Apparatus and method for constructing big data to share production information between multiple companies producing one finished product in the textile industry}

The present invention relates to big data construction technology, and more particularly, to a big data construction apparatus and method for sharing production information between a plurality of companies that produce one finished product.

Production is the main activity of human economic activity and refers to the act or process of making goods that meet human needs from land or raw materials.

Korean Patent Publication No. 2006-0126749 published on December 08, 2006 (Title: Method, device and computer program for making corporate management data accessible by web-application) Korean Patent Publication No. 2008-0098304 published on November 7, 2008 (Name: Authorization management system for database access authentication and database account protection method)

An object of the present invention is to provide a big data construction device and method for sharing production information between a plurality of companies producing one finished product.

Big data construction apparatus according to a preferred embodiment of the present invention for achieving the above object is to integrate a communication unit for communication with terminals of a plurality of companies producing one finished product, and production information of the plurality of companies. When an access request message is received from any one of the plurality of terminals through the storage unit and the communication unit, the terminal can access production information of the plurality of companies based on the received connection request message An access processing unit that authenticates whether or not there is an authority, and as a result of the authentication, if the authentication is successful, allocates a workspace that is a volatile memory space of a size corresponding to the terminal, and among the production information of the storage unit, the terminal After copying the production information corresponding to the work performed by the belonging company and loading it into the assigned work space, a connection response message including the access address of the assigned work space is transmitted to the terminal through the communication unit. It includes a work management unit that includes a request.

The access request message includes an identifier and password of a user of the terminal and a context of the terminal, and the context includes a device identifier of the terminal, a type of terminal, access location, access time, access IP, access port, and access network information. It includes the type, identification information of the access network and the application used for the connection.

The access processing unit includes an account authentication unit that authenticates whether the user's account is a pre-registered account through the user's identifier and password of the terminal, and a weight is applied to the calculation result of any one layer to be input to the next layer. and a context authentication unit that authenticates whether the context corresponds to the registered account by using a context authentication network including a plurality of layers including a plurality of operations.

The context authentication network includes an input layer that receives the context, a convolution layer that constructs a first feature map by performing a convolution operation through a first filter composed of a weight matrix for the context of the input layer, and the convolution layer that constructs a first feature map. A pooling layer constituting a second feature map by performing a pooling operation on the first feature map of the solution layer through a second filter composed of a weight matrix, and converting the second feature map of the pooling layer and arranging them into one column An operation corresponding to each of the plurality of operation nodes is performed by performing an operation through an activation function to which a weight is applied to the second feature map including a flattening layer and a plurality of operation nodes, each of which is arranged in a single column. It includes a fully-connected layer that calculates a value, and a plurality of output nodes corresponding to each of a plurality of companies, and performs an operation through an activation function on the calculated values of the plurality of operation nodes of the fully-connected layer to obtain the context. and an output layer that calculates a context vector representing a probability that a terminal belongs to each of the plurality of companies and outputs the calculated probability as an output value.

의 손실값이 최소가 되도록 상기 컨텍스트인증망의 가중치를 최적하되, 동일한 복수의 학습용 컨텍스트를 이용하여 상기

의 값을 1, 0.5 및 0으로 변경하면서 상기 컨텍스트인증망에 대한 학습을 수행하는 학습부를 더 포함한다. The apparatus sets a label according to a known company for a known learning context, inputs the learning context to the context authentication network, and then the context authentication network performs a plurality of calculations in which a weight is applied to the learning context. When an output value is calculated, a loss function that calculates a loss value that is the difference between the set label and the output value

Optimizing the weight of the context authentication network so that the loss value of

A learning unit for performing learning on the context authentication network while changing the value of 0 to 1, 0.5, and 0 is further included.

는 하이퍼파라미터이다. Here, Ecm represents a loss function, hy is a one-hot encoding vector representing a company known according to a label, fy is the output value, cy is a center vector among a plurality of context vectors of a company known according to a label, The y is the company index, the N is the maximum value of y, and the

is a hyperparameter.

The work management unit further includes a work authentication unit that monitors a work performed by the terminal in the work space when the terminal accesses the work space through the communication unit and determines whether the work performed by the terminal is a valid work. do.

The work authentication unit sequentially extracts a plurality of logs recorded in chronological order after the terminal accesses the workspace, generates a plurality of log vectors representing characteristics of each of the plurality of extracted logs, and generates a plurality of log vectors. A pre-processing unit that provides vectors in order according to the time order in which the logs were recorded, and generates a time vector representing the time at which the log was recorded, and a time vector corresponding to the plurality of log vectors and the context vector are combined to provide the time order. A vector combiner for generating a plurality of work log vectors arranged in , and a task performed by the terminal in the work space by performing one or more operations to which weights are applied in order of a plurality of stages to the plurality of work log vectors. A task attribute vector representing the attribute of is calculated, a probability indicating whether a task performed by the terminal in the workspace is a task authorized to be performed by the terminal is calculated from the calculated task attribute vector, and the calculated probability is It includes the work certification network that outputs.

The work authentication network is characterized in that it consists of a plurality of stages arranged in sequence. In addition, the work certification network includes a work input layer that receives a plurality of work log vectors arranged in chronological order corresponding to the plurality of stages for each of the plurality of stages, and a state value of the previous stage and an input value of the current stage. A state value of the current stage is calculated by performing an operation to which a state and an input weight are applied to the log vector, and then a plurality of hidden cells are passed to the next stage, wherein the last of the plurality of hidden cells The hidden cell calculates the state value of the current stage by performing an operation to which the state weight is applied to the state value of the previous stage, and then applies the output weight to the calculated state value of the current stage. A job hidden layer that calculates a job attribute vector representing the attribute of , a job output layer that divides the job attribute vector into a plurality of attribute values and outputs a plurality of separated attribute values, and a weight value for the calculated plurality of attribute values. Discrimination layer for calculating a discriminant value representing a probability that a task performed by the terminal in the workspace is a task authorized to be performed by the terminal and a probability that the task is not authorized to be performed by the terminal by performing a plurality of applied operations. includes

If the work performed in the work space is a work authorized for the terminal to be performed, the device determines the production information modified according to the result of the work performed by the terminal in the work space, as the modified production information. A recording processing unit for updating the production information of the storage unit is further included.

In the big data construction method according to a preferred embodiment of the present invention for achieving the above object, in a state in which production information of a plurality of companies producing one finished product is integrated and stored, the connection processing unit among the plurality of companies Receiving an access request message from a terminal of any one company, and authenticating whether the terminal has permission to access production information of the plurality of companies based on the connection request message received by the connection processing unit. and, as a result of the authentication, if the authentication is successful, allocating a work space, which is a volatile memory space of a size corresponding to the terminal by the work management unit, and copying production information corresponding to the work to be performed and loading it into the assigned work space; and the work manager providing a connection response message including an access address of the assigned work space to the terminal through the communication unit. It includes steps to

The access request message includes an identifier and password of a user of the terminal and a context of the terminal, and the context includes a device identifier of the terminal, a type of terminal, access location, access time, access IP, access port, and access network information. It includes the type, identification information of the access network and the application used for the connection. Here, the step of performing the authentication includes the step of authenticating, by the connection processing unit, whether the user's account is a pre-registered account through the user's identifier and password of the terminal, and the connection processing unit's operation result of any one layer. and authenticating whether the context corresponds to the registered account by using a context authentication network including a plurality of layers including a plurality of operations that are input to the next layer by applying a weight to .

The step of authenticating whether the context corresponds to the registered account using the context authentication network includes: receiving the context by an input layer of the context authentication network; Constructing a first feature map by performing a convolution operation on the context of the layer through a first filter composed of a weight matrix; Constructing a second feature map by performing a pooling operation through a second filter composed of matrices; converting the second feature map of the pooling layer by the flattening layer of the context authentication network and arranging them into one column; An operation value corresponding to each of the plurality of operation nodes is calculated by performing an operation through an activation function with a weight applied to the second feature map in which a fully connected layer including a plurality of operation nodes of the context authentication network is arranged in a single column. calculating, and an output layer including a plurality of output nodes corresponding to each of the plurality of companies of the context authentication network performs an operation through an activation function on the calculation values of the plurality of calculation nodes of the fully connected layer and calculating a context vector representing a probability that a terminal having the context belongs to each of the plurality of companies, and outputting the calculated probability as an output value.

의 손실값이 최소가 되도록 상기 컨텍스트인증망의 가중치를 최적하되, 동일한 복수의 학습용 컨텍스트를 이용하여 상기

의 값을 1, 0.5 및 0으로 변경하면서 상기 컨텍스트인증망에 대한 학습을 수행하는 단계를 더 포함한다. 여기서, 상기 Ecm은 손실함수를 나타내고, 상기 hy는 레이블에 따라 알려진 기업을 나타내는 원핫인코딩벡터이고, 상기 fy는 상기 출력값이고, 상기 cy는 레이블에 따라 알려진 기업의 복수의 컨텍스트벡터 중 중심벡터이고, 상기 y는 기업 인덱스이고, 상기 N은 y의 최댓값이고, 상기

는 하이퍼파라미터인 것을 특징으로 한다. Before the step of receiving the access request message, the learning unit sets a label according to the known company for the known learning context, and after the learning unit inputs the learning context to the context authentication network, the context authentication network When an output value is calculated through a plurality of operations to which weights are applied for this learning context, a loss function that calculates a loss value that is the difference between the set label and the output value

Optimizing the weight of the context authentication network so that the loss value of

The step of performing learning on the context authentication network while changing the value of 1, 0.5, and 0 is further included. Here, Ecm represents a loss function, hy is a one-hot encoding vector representing a company known according to a label, fy is the output value, cy is a center vector among a plurality of context vectors of a company known according to a label, The y is the company index, the N is the maximum value of y, and the

is a hyperparameter.

In the method, after the work management unit provides the terminal with an access response message including the access address of the assigned workspace through the communication unit, when the work authentication unit of the work management unit accesses the workspace, the terminal accesses the workspace. , Monitoring a task performed by the terminal in the workspace to determine whether the task performed by the terminal is a legitimate task.

The step of determining whether or not the task performed by the terminal is a legitimate task may include a plurality of log vectors arranged in chronological order from a log recorded in chronological order by the pre-processing unit of the work authentication unit, in which the task performed by the terminal in the workspace is recorded in chronological order. a vector combiner for generating a plurality of work log vectors arranged in chronological order by combining the time vector and the context vector corresponding to the plurality of log vectors; A negative work authentication network performs one or more calculations to which weights are applied to the plurality of work log vectors in order of a plurality of stages to calculate a work attribute vector representing attributes of work performed by the terminal in the work space; and calculating a probability indicating whether the task performed by the terminal in the workspace is a task authorized to be performed by the terminal from the calculated task attribute vector, and outputting the calculated probability.

In the step of outputting the calculated probability, the work input layer of the work certification network composed of a plurality of stages arranged in sequence generates a plurality of work log vectors arranged in time order corresponding to the plurality of stages for each of the plurality of stages. In the step of receiving an input, each of a plurality of hidden cells of the work output layer of the work authentication network performs an operation in which the state and input weight are applied to the state value of the previous stage and the work log vector, which is the input value of the current stage, to obtain the current stage After calculating the state value, the calculated state value is passed to the next stage, and the last hidden cell among the plurality of hidden cells performs an operation in which a state weight is applied to the state value of the previous stage to obtain the state value of the current stage After calculating, calculating a job attribute vector representing a property of a job performed by a terminal in a workspace, which is an output value, by applying an output weight to the calculated state value of the current stage; dividing the attribute vector into a plurality of attribute values and outputting a plurality of attribute values;

The determination layer of the work authentication network performs a plurality of calculations to which weights are applied to the plurality of attribute values, and the probability that the work performed by the terminal in the work space is a work authorized to be performed by the terminal and the terminal and calculating a discriminant value representing a probability that the task is not authorized to be performed.

In the method, after the step of determining whether a job performed by the terminal is a legitimate job, if the job performed in the workspace is a job authorized to be performed by the terminal, the recording processing unit determines whether the terminal performed in the workspace. The method further includes updating the production information of the storage unit with the corrected production information when production information corrected according to the operation result exists.

According to an embodiment of the present invention, by building big data, it is possible to share production information between a plurality of companies that produce one finished product. Accordingly, it is possible to easily manage an overall business process requiring collaboration between a plurality of companies, and productivity may be improved. At this time, security can be improved by determining users who can share information using a deep neural network.

1 is a diagram for explaining the configuration of a big data construction system for sharing production information between a plurality of companies producing one finished product according to an embodiment of the present invention.
2 is a diagram for explaining the configuration of an apparatus for building big data for sharing production information between a plurality of companies that produce one finished product according to an embodiment of the present invention.
3 is a diagram for explaining the configuration of a control unit of an apparatus for constructing big data for sharing production information between a plurality of companies producing one finished product according to an embodiment of the present invention.
4 is a diagram for explaining the detailed configuration of the connection processing unit of the apparatus for building big data for sharing production information between a plurality of companies producing one finished product according to an embodiment of the present invention.
5 and 6 are diagrams for explaining the configuration of the context authentication network of the connection processing unit of the device for building big data for sharing production information among a plurality of companies that produce one finished product according to an embodiment of the present invention. .
7 and 8 are diagrams for explaining learning of a context authentication network of an apparatus for constructing big data for sharing production information between a plurality of companies producing one finished product according to an embodiment of the present invention.
9 is a diagram for explaining a detailed configuration of a work management unit of an apparatus for constructing big data for sharing production information among a plurality of companies producing one finished product according to an embodiment of the present invention.
10 is a diagram for explaining the detailed configuration of a work authentication unit of an apparatus for constructing big data for sharing production information between a plurality of companies producing one finished product according to an embodiment of the present invention.
11 to 13 are diagrams for explaining the configuration of a work authentication network of a device for building big data for sharing production information between a plurality of companies producing one finished product according to an embodiment of the present invention.
14 is a flowchart illustrating a big data construction method for sharing production information between a plurality of companies producing one finished product according to an embodiment of the present invention.

Prior to the detailed description of the present invention, the terms or words used in this specification and claims described below should not be construed as being limited to a common or dictionary meaning, and the inventors should use their own invention in the best way. It should be interpreted as a meaning and concept corresponding to the technical idea of the present invention based on the principle that it can be properly defined as a concept of a term for explanation. Therefore, the embodiments described in this specification and the configurations shown in the drawings are only the most preferred embodiments of the present invention, and do not represent all of the technical ideas of the present invention, so various equivalents that can replace them at the time of the present application. It should be understood that there may be water and variations.

In the description and claims of embodiments of the present invention, a "network" is defined as one or more data links that enable the transfer of electronic data between computer systems and/or modules. When information is transmitted or provided to a computer system over a network or over a communication connection of any kind or kind, i.e., wired, wireless, or a combination of wired and wireless, the connection may be understood as a computer-readable medium. . Computer readable instructions include, for example, instructions and data that cause a general purpose or special purpose computer system to perform a particular function or group of functions. Computer executable instructions may be, for example, binary, intermediate format instructions, such as assembly language, or even source code. Moreover, the present invention relates to personal computers, laptop computers, handheld devices, multiprocessor systems, microprocessor-based or programmable consumer electronics, network PCs, minicomputers, mainframe computers, mobile telephones, cellular phones, and mobile devices. In a network computing environment having various types of computer system configurations including communication terminals, smart phones, PDAs, pagers, etc., the above computer systems may be applied.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. At this time, it should be noted that the same components in the accompanying drawings are indicated by the same reference numerals as much as possible. In addition, detailed descriptions of well-known functions and configurations that may obscure the gist of the present invention will be omitted. For the same reason, some components in the accompanying drawings are exaggerated, omitted, or schematically illustrated, and the size of each component does not entirely reflect the actual size.

First, a system for building big data for sharing production information between a plurality of companies that produce one finished product will be described. 1 is a diagram for explaining the configuration of a system for building big data for sharing production information between a plurality of companies producing one finished product.

Referring to Figure 1, a system for building big data for sharing production information between a plurality of companies producing one finished product (hereinafter abbreviated as 'big data system') is a system in which a plurality of companies producing one finished product It is a system for sharing all information related to the production of finished products. Here, it can be assumed that one finished product is a 'textile fabric product'. In this case, a plurality of companies are a dyeing company, which is a fabric processing company that focuses on dyeing textile fabrics, a processing company, which is a fabric processing company that gives various characteristics to textile fabrics, and a fabric processing company that checks for defects in the produced fabric and reports the results of the defects. Inspection company, a company that provides fabric products, a local company that receives orders from brand companies that plan, develop and produce fabric products, directs and manages production, and delivers them to brand companies, a local company that makes and sells textile products with manufactured fabrics It may include branded companies and the like.

The big data system includes a shared server 10 and a plurality of terminals 20. The plurality of terminals 20 are devices used by a plurality of employees and users of a plurality of companies that produce one finished product. Such a terminal 20 may exemplify a desktop, a laptop, a tablet, a smart phone, and the like. A plurality of terminals 20 may access the shared server 10 through a network including an access network and a core network.

The sharing server 10 serves as a database server that builds big data by integrating production information of a plurality of companies according to an embodiment of the present invention. In addition, the sharing server 10 authenticates whether a plurality of terminals 20 are qualified to access production information of a plurality of companies, and allows access to production information only for terminals 20 that have succeeded in authentication. It acts as an authentication server at the same time. Moreover, the sharing server 10 may serve as a security server that separates and manages the terminal 20 and the original production information in order to protect original data, that is, production information, from malicious attacks.

Then, the configuration of the aforementioned sharing server 10 will be described in more detail. 2 is a diagram for explaining the configuration of an apparatus for building big data for sharing production information between a plurality of companies that produce one finished product according to an embodiment of the present invention. 3 is a diagram for explaining the configuration of a control unit of an apparatus for constructing big data for sharing production information between a plurality of companies producing one finished product according to an embodiment of the present invention. First, referring to FIG. 2 , a sharing server 10 according to an embodiment of the present invention includes a communication unit 11 , a storage unit 12 and a control unit 13 .

The communication unit 11 is for communicating with the terminal 20 through a network. The communication unit 11 may receive data from the terminal 20 or transmit data to the terminal 20 under the control of the control unit 13 . The communication unit 11 may include a modem that modulates a transmitted signal and demodulates a received signal in order to transmit and receive data through a network. The communication unit 11 may transmit data received from the terminal 20 through a network, for example, an access request message, etc. to the control unit 13 . In addition, the communication unit 11 may transmit data received from the control unit 13, for example, an access response message, etc. to the terminal 20 through a network.

The storage unit 12 serves to store programs and data necessary for the operation of the sharing server 10 . In particular, the storage unit 12 includes a database for integrating and storing production information of a plurality of companies under the control of the control unit 13 . Furthermore, the storage unit 12 may store an identifier (ID), a password (PW), and the like of each of the plurality of terminals 20 to be managed. Various types of data stored in the storage unit 12 may be registered, deleted, changed, or added by the authenticated terminal 200 .

The control unit 13 may control the overall operation of the sharing server 10 and signal flow between internal blocks of the sharing server 10, and may perform a data processing function of processing data. The controller 13 may be a central processing unit, a digital signal processor, or the like. Referring to FIG. 3 , the control unit 13 includes a connection processing unit 100 , a work management unit 200 , a recording processing unit 300 and a learning unit 400 .

The connection processing unit 100 is for authenticating whether the terminal 20 has permission to access production information of a plurality of companies stored in the storage unit 12 . When the connection processing unit 100 receives an access request message from any one terminal 20 of a plurality of terminals 20 of a plurality of companies through the communication unit 11, the terminal 20 is sent based on the received connection request message. Authentication is performed on whether or not there is authority to access the production information of the plurality of companies. At this time, the access processing unit 100 may perform part or all of the authentication process for the terminal 20 using a deep neural network (DNN) model.

The task management unit 200 may allocate a work space, which is a volatile memory space of a size corresponding to the terminal 20 that has succeeded in authentication. In addition, the work management unit 200 copies production information corresponding to a job performed by a company to which the terminal 20 belongs among production information stored in the storage unit 12 and loads it into an assigned work space. Then, the work management unit 200 transmits an access response message including the access address of the assigned work space to the terminal 20 that has succeeded in authentication through the communication unit 11 . In addition, when the terminal 20 that has succeeded in authentication approaches the workspace, the task management unit 200 monitors the task performed by the terminal 20 that has succeeded in authentication in the workspace so that the task performed by the terminal 20 is a legitimate task. determine whether or not At this time, the task management unit 200 may determine whether the task performed by the terminal 20 is a valid task by using a deep neural network (DNN) model.

If the work performed by the terminal 20 that has succeeded in authentication in the workspace is a job authorized to be performed by the terminal 20, the recording processing unit 300 records the modified work according to the result of the work performed by the terminal 20 in the workspace. If the production information exists, the production information in the storage unit 12 is updated with the corrected production information.

As described above, the learning unit 400 performs learning on the deep neural network model used by the connection processing unit 100 and the deep neural network model used by the task management unit 200 . A specific operation of the learning unit 400 will be described in more detail below.

Next, each of the aforementioned connection processing unit 100, work management unit 200, recording processing unit 300, and learning unit 400 will be described in detail. First, the connection processing unit 100 will be described in more detail. 4 is a diagram for explaining the detailed configuration of the connection processing unit of the apparatus for building big data for sharing production information between a plurality of companies producing one finished product according to an embodiment of the present invention. 5 and 6 are diagrams for explaining the configuration of the context authentication network of the connection processing unit of the device for building big data for sharing production information among a plurality of companies that produce one finished product according to an embodiment of the present invention. . 7 and 8 are diagrams for explaining learning of a context authentication network of an apparatus for constructing big data for sharing production information between a plurality of companies producing one finished product according to an embodiment of the present invention.

Referring to FIG. 4 , the connection processing unit 100 according to an embodiment of the present invention includes an account authentication unit 110 and a context authentication unit 120 . In particular, the context authentication unit 120 includes a context authentication network 121 .

As described above, the connection processing unit 100 may receive an access request message from the terminal 20 through the communication unit 11 . Here, the access request message includes an account including an identifier and password of a user of the terminal 20 and a context of the terminal 20 . In particular, the context includes the device identifier of the terminal, the type of terminal, access location, access time, access IP, access port, access network type, access network identification information, and an application used for access.

The account authentication unit 110 compares the account including the identifier (ID) and password (PW) of the user of the terminal 20 with the user's account previously stored in the storage unit 12, and the user transmitted by the terminal 20. It is possible to authenticate whether the account of is a pre-registered account.

The context authentication unit 120 authenticates whether the context of the terminal 20 matches the registered account of the terminal 20 using the deep neural network context authentication network 121 . A plurality of terminals 20 belonging to the same company, that is, a plurality of terminals 20 used by executives and employees of the same company have contexts having a high degree of correlation with each other. Accordingly, the context authentication network 121 is learned by the learning unit 400 to identify the company to which the terminal 20 belongs through the context of the terminal 20 . Therefore, the context authentication unit 120 identifies the company using the context authentication network 121, and then determines whether the account of the terminal 20 is an account assigned to the executives and employees of the identified company, thereby determining the context of the terminal 20. It is possible to authenticate whether or not matches the registered account of the terminal 20 .

Meanwhile, referring to FIGS. 5 and 6 , the context authentication network 121 includes a plurality of layers. That is, the context authentication network 121 sequentially includes an input layer (IL), a convolution layer (CL), a pooling layer (PL), a flattening layer (FL), and a fully connected layer. It includes a fully-connected layer (FCL) and an output layer (OL).

The context authentication unit 120 converts the context including the device identifier of the terminal, the type of terminal, the access location, the access time, the access IP, the access port, the type of access network, the identification information of the access network, and the application used for the connection into binary. code (binary code), and after arranging the context converted to binary code into a matrix (channel × row × column) having at least one channel of a predetermined size according to the input layer (IL), the context authentication network (121 ) into the

The input layer IL is composed of an input matrix (channel×row×column) having at least one channel of a predetermined size. For example, as shown in FIG. 5, the input matrix may have a size of (channel × row × column) = (2 × 7 × 12). The context converted into binary code is input as an element of an input layer (IL) input matrix.

The convolution layer (CL) and the pooling layer (PL) are composed of at least one feature map (FM). A feature map is composed of a matrix of a predetermined size. The feature map is created as a result of performing weighted operations on the values of the previous layer. These weights are applied via filters. Here, the filter is a matrix having at least one channel of a predetermined size, and the value of each element of the matrix constituting the filter becomes a weight (w). Therefore, in each element of the matrix constituting the feature map, a value calculated according to the weighted operation using the filter W in the previous layer is stored.

The convolutional layer (CL) includes at least one first feature map. The first feature map (FM1) constituting the convolution layer (CL) is the result of performing the convolution operation (convolution) through the first filter (W1) composed of the weight matrix for the context input to the input layer (IL). is created with

The pooling layer PL includes at least one second feature map FM2. The second feature map FM2 constituting the pooling layer PL is generated as a result of performing a pooling or subsampling operation on the first feature map FM1 through the second filter W2 composed of a weight matrix. do.

The flattening layer FL serves to transform the shape of the previous layer to match the shape of the next layer. That is, the flattening layer FL changes the second feature map FM2 of the pooling layer PL, which is in matrix form, into a column form according to an operation node of a fully-connected layer (FCL). For example, if the second feature map FM2 is in the shape of 1×6×6 (channels×rows×columns), the flattening layer FL changes it to a shape of 36 (columns). According to this change, matrix-type values of the second feature map may be sequentially arranged in one column and input to a plurality of operation nodes of a fully-connected layer (FCL).

Referring to FIG. 6, the fully connected layer (FCL) includes a plurality of operation nodes C1, C2, ..., Cx. Each of the plurality of operation nodes (C1, C2, ..., Cx) of the fully connected layer (FCL) has a weight (w) with respect to the second feature map (FM2) arranged in one column by the flattening layer (FL). An operation value corresponding to each of the plurality of operation nodes C1, C2, ..., Cx is calculated by performing an operation through the applied activation function. Here, the activation function may include sigmoid, hyperbolic tangent (tanh), exponential linear unit (ELU), rectified linear unit (ReLU), leaky ReLU, Maxout, Minout, Softmax, and the like. .

The output layer OL includes a number of output nodes O1, O2, ..., Oy corresponding to a plurality of companies producing one finished product according to an embodiment of the present invention. That is, each of the plurality of output nodes O1, O2, ..., Oy corresponds to a plurality of companies.

The plurality of output nodes (O1, O2, ..., Oy) of the output layer (OL) are activated for the calculation values of the plurality of operation nodes (C1, C2, ..., Cx) of the fully connected layer (FCL) By performing an operation through a function, a context vector representing the characteristics of the company to which the terminal 20 that transmitted the context through the access request message belongs and the employee using the terminal 20 is calculated. That is, the plurality of output nodes (O1, O2, ..., Oy) operate through the activation function for the calculated values of the plurality of operation nodes (C1, C2, ..., Cx) of the fully connected layer (FCL). is performed to calculate the output value of each of the plurality of output nodes O1, O2, ..., Oy. As described above, each of the plurality of output nodes O1, O2, ..., Oy corresponds to a plurality of enterprises. For example, the first output node O1 corresponds to company A, the second output node O1 corresponds to company B, the third output node O1 corresponds to company C, and the y th output node Oy ) can correspond to entity H. In addition, the output value of each of the plurality of output nodes (O1, O2, ..., Oy) determines which one of the output nodes (O1, O2, ..., Oy) the terminal 20 that transmitted the context through the connection request message 1) represents the probability of belonging to the corresponding company. For example, if the first output node O1 corresponds to company A and the output value of the first output node O1 is 0.221987, the terminal 20 that transmits the context through the access request message is used by employees of company A. There is a 22% probability that it is a terminal that As another example, if the second output node O2 corresponds to company B and the output value of the second output node O2 is 0.878911, the terminal 20 transmitting the context through the access request message is used by an employee of company B. There is an 88% probability that the terminal is Accordingly, the context vector F=[f1, f2, f3, ..., fy], which connects all the output values of the plurality of output nodes (O1, O2, ..., Oy), transmits the context through the connection request message. It can be said to be a vector expression representing the characteristics of the company to which one terminal 20 belongs and the executives and employees using the terminal 20 .

Meanwhile, the learning unit 400 may teach the context authentication network 121 to generate a context vector representing the characteristics of the company to which the terminal 20 transmitting the context belongs and the employee using the terminal 20 . The context vector is obtained by connecting the output values of a plurality of output nodes (O1, O2, ..., Oy), and the output values of each of the plurality of output nodes (O1, O2, ..., Oy) [f1, f2, f3, ..., fy] represent the probability that the terminal 20 that has transmitted the input context belongs to the company corresponding to the corresponding output node (O1, O2, ..., Oy). Therefore, the learning unit 400 can also be said to teach the context authentication network 121 to classify the company to which the terminal 20 that has transmitted the input context through the operation on the context belongs.

The learning unit 400 sets the output layer OL of the context authentication network 121 according to the company to learn. For example, for convenience of explanation, it is assumed that four companies including company A, company B, company C, and company D exist. Accordingly, the output layer OL of the context authentication network 121 has four output nodes O1, O2, O3, and O4, and the length of the context vector is 4.

Next, the learning unit 400 prepares the context transmitted by the terminal 20 known to the company as a learning context. Then, the learning unit 400 sets a label according to a known company for a learning context in which the company is known. Then, the learning unit 400 inputs the learning context to the context authentication network 121 . Accordingly, the context authentication network 121 calculates an output value through a plurality of operations to which unlearned weights are applied to the learning context. Then, the learning unit 400 sets the weight of the context authentication network 121 so that the loss value of the loss function of Equation 1 representing the difference between the set label and the output value is minimized by using the loss function of Equation 1 below. optimize

는 하이퍼파라미터로 수학식 1의 첫 번째 텀과 두 번째 텀의 반영 비율을 조절하기 위한 파라미터이다. 수학식 1에 보인 바와 같이, 하이퍼파라미터

는 그 값이 1, 0.5 및 0으로 전환되며, 학습부(400)는 동일한 학습 데이터인 학습용 컨텍스트를 이용하여

의 값을 1, 0.5 및 0으로 전환하면서, 컨텍스트인증망(121)에 대해 1차 내지 3차의 학습을 수행한다. 이러한 1차 내지 3차의 학습에 대해서 설명하면 다음과 같다. In Equation 1, Ecm represents a loss function for calculating a difference between a label and an output value of the context authentication network 121. Here, the output value of the context authentication network 121 may be a context vector (F = [f1, f2, ..., fy] or an output value of each of a plurality of output nodes O1 to Oy. Here, hy is a label It is a one hot encoding vector representing a company known according to fy represents a context vector or an output value of an output node cy represents a center vector among a plurality of context vectors of a known company according to a label y represents a company index , and N is the maximum value of y. For example, assuming that there are 4 companies to learn, y is a positive integer and has a value from 1 to 4, and N is 4. On the other hand, Equation 1 at

Is a hyperparameter and is a parameter for adjusting the reflection ratio of the first term and the second term of Equation 1. As shown in Equation 1, the hyperparameter

is converted to 1, 0.5, and 0, and the learning unit 400 uses the same learning data, the learning context

1st to 3rd learning is performed on the context authentication network 121 while changing the value of 1, 0.5, and 0. The first to third learning is described as follows.

로 설정하여 학습을 수행하는 1차 학습을 수행한다. 1차 학습 시, 학습부(400)는 기업이 알려진 단말(20)이 전송한 컨텍스트를 학습용 컨텍스트로 마련한다. 그런 다음, 학습부(400)는 학습용 컨텍스트의 알려진 기업에 따라 레이블을 설정한다. 1차 학습 시, 레이블은 다음의 표 1과 같이 원핫인코딩(one hot encoding) 벡터로 설정한다. First, the learning unit 400

Set to , and perform the primary learning that performs learning. At the time of primary learning, the learning unit 400 prepares the context transmitted by the terminal 20 known to the company as a context for learning. Then, the learning unit 400 sets a label according to the known company of the learning context. During the first learning, the label is set as a one hot encoding vector as shown in Table 1 below.

Known Enterprises in Learning Context expected value O4 O3 O2 O1 Firm A 0 0 0 One company B 0 0 One 0 company C 0 One 0 0 company D One 0 0 0

이기 때문에 학습부(400)는 다음의 수학식 2와 같은 손실함수를 통해 레이블인 원핫인코딩벡터와 컨텍스트벡터 혹은 복수의 출력노드(O1 내지 Oy, 예컨대, y=4) 각각의 출력값의 차이인 손실함수의 손실값이 최소화되도록 컨텍스트인증망(121)의 가중치를 최적화한다. After setting labels as shown in Table 1, the learning unit 400 inputs the learning context to the context authentication network 121. Then, the context authentication network 121 calculates a context vector by performing a plurality of operations to which the unlearned weight w is applied to the learning context.

Therefore, the learning unit 400 uses a loss function such as Equation 2 below to obtain a loss, which is the difference between the one-hot encoding vector, which is a label, and the context vector, or the output value of each of the plurality of output nodes (O1 to Oy, for example, y = 4) The weight of the context authentication network 121 is optimized so that the loss value of the function is minimized.

The primary learning as described above may be performed through a plurality of learning contexts. After the primary learning is completed, the learning unit 400 inputs the plurality of learning contexts into the context authentication network 121 where the primary learning is completed, and thus, the plurality of learning contexts are obtained from the plurality of learning contexts through the context authentication network 121. vectors can be derived. Then, the learning unit 400 may map each of the derived context vectors onto a vector space formed through the context authentication network 121 and classify each company according to the values of the plurality of context vectors. For example, as assumed above, if the length of the context vector is 4 and the value of the context vector is F= [f4, f3, f2, f1] = [0.01, 0.01, 0.01, 0.97], the corresponding context vector is Since the probability of f1 corresponding to the output value of the first output node O1 corresponding to A is the highest, it can be classified as belonging to company A. For example, FIG. 7 shows the distribution of context vectors of companies A and B mapped to a vector space formed through the context authentication network 121 that has been primarily learned. As such, as a result of the primary learning, the context authentication network 121 can distinguish company A and company B through the context vector based on the borderline (BL) of the vector space.

로 설정하여 학습을 수행하는 2차 학습을 수행한다. 2차 학습 시, 학습부(400)는 1차 학습 시 사용된 학습용 컨텍스트를 마련한다. 그런 다음, 학습부(400)는 학습용 컨텍스트의 알려진 기업에 따라 레이블을 설정한다. 2차 학습 시, 학습부(400)는 하나의 학습용 컨텍스트에 대응하여 2 종의 레이블을 설정한다. 즉, 학습부(400)는 상기의 표 1과 같이 원핫인코딩벡터를 통해 제1 레이블을 설정한다. 또한, 학습부(400)는 학습부(400)는 1차 학습이 완료된 컨텍스트인증망(121)에 의해 형성되는 벡터공간 상에서 동일 분류, 즉, 동일 기업의 컨텍스트벡터의 중심벡터를 선택하여 제2 레이블로 설정한다. 예컨대, 도 7을 참조하면, 기업 A의 복수의 컨텍스트벡터 중에서 중간값을 가지는 컨텍스트벡터를 중심벡터(Ca1)로 선택하고, 기업 B의 복수의 컨텍스트벡터 중에서 중간값을 가지는 컨텍스트벡터를 중심벡터(Cb1)로 선택한다. Next, the learning unit 400

Set to , and perform secondary learning that performs learning. In the second learning, the learning unit 400 prepares the learning context used in the first learning. Then, the learning unit 400 sets a label according to the known company of the learning context. During secondary learning, the learning unit 400 sets two types of labels corresponding to one learning context. That is, the learning unit 400 sets the first label through the one-hot encoding vector as shown in Table 1 above. In addition, the learning unit 400 selects the center vector of the context vector of the same classification, that is, the same company, on the vector space formed by the context authentication network 121 for which the primary learning has been completed. set to label For example, referring to FIG. 7 , a context vector having a median value among a plurality of context vectors of company A is selected as a center vector (Ca1), and a context vector having a median value among a plurality of context vectors of company B is a center vector ( Select Cb1).

이기 때문에 학습부(400)는 수학식 1로부터 수학식 3과 같은 손실함수를 도출하고, 도출된 손실함수를 통해 레이블과 컨텍스트벡터 혹은 복수의 출력노드(O1 내지 Oy, 예컨대, y=4) 각각의 출력값의 차이인 손실함수의 손실값이 최소화되도록 컨텍스트인증망(121)의 가중치를 최적화한다. The learning unit 400 inputs the learning context to the context authentication network 121 . Then, the context authentication network 121 calculates a context vector by performing a plurality of operations to which a weight w is applied to the learning context.

Therefore, the learning unit 400 derives a loss function such as Equation 3 from Equation 1, and through the derived loss function, a label and a context vector or a plurality of output nodes (O1 to Oy, for example, y = 4), respectively The weight of the context authentication network 121 is optimized so that the loss value of the loss function, which is the difference between the output values of , is minimized.

In Equation 3, hy is the first label and cy is the second label. The secondary learning procedure as described above may be repeatedly performed through a plurality of learning contexts used in the primary learning.

After the secondary learning is completed, the learning unit 400 inputs the plurality of learning contexts used in the secondary learning to the context authentication network 121 in which the secondary learning is completed, so that the plurality of learning contexts are obtained through the context authentication network 121. A plurality of context vectors may be derived from the learning context. Then, the learning unit 400 may map each of the derived context vectors onto a vector space formed through the context authentication network 121 and classify each company according to the values of the plurality of context vectors. For example, FIG. 8 shows the distribution of context vectors of companies A and B mapped to a vector space formed through the context authentication network 121 secondarily learned. In this way, in the secondary learning, learning is performed using not only one-hot encoding vectors but also center vectors (eg, Ca1 and Cb1). Accordingly, compared to the vector space (FIG. 7) formed through the firstly learned context authentication network 121, all context vectors are formed on the vector space (FIG. 8) formed through the secondly learned context authentication network 121. According to the degree of correlation with the same type of center vector, it moves in the direction of the same type of center vector as indicated by the arrow shown in FIG. 8 . For example, when the context vectors derived from any one learning context are compared and observed, the secondly learned context vectors (f1st) on the vector space (FIG. 7) formed through the firstly learned context authentication network 121 are compared. The context vector f2nd on the vector space (FIG. 8) formed through the context authentication network 121 is moved in the direction of the center vector. That is, in the secondary learning, since learning is performed using not only one-hot encoding vectors but also center vectors (eg, Ca1 and Cb1), correlation between homogeneous context vectors based on center vectors as well as classification by one-hot encoding vectors is improved. learn more Therefore, as a result of the secondary learning, the context authentication network 121 can not only distinguish between company A and company B through the context vector based on the boundary line (BL: Borderline) of the vector space, but also within the cluster circle (CCA, CCB). In , the similarity between context vectors is learned through the distance from the center vector.

로 설정하여 학습을 수행하는 3차 학습을 수행한다. 3차 학습 시, 학습부(400)는 1차 및 2차 학습 시 사용된 학습용 컨텍스트와 동일한 학습용 컨텍스트를 이용한다. 이에 따라, 학습부(400)는 학습용 컨텍스트의 알려진 기업에 따라 레이블을 설정한다. 3차 학습 시, 또한, 학습부(400)는 학습부(400)는 2차 학습이 완료된 컨텍스트인증망(121)에 의해 형성되는 벡터공간 상에서 동일 분류, 즉, 동일 기업의 컨텍스트벡터의 중심벡터를 선택하여 레이블로 설정한다. 예컨대, 도 8을 참조하면, 기업 A의 복수의 컨텍스트벡터 중에서 중간값을 가지는 컨텍스트벡터를 중심벡터(Ca2)로 선택하고, 기업 B의 복수의 컨텍스트벡터 중에서 중간값을 가지는 컨텍스트벡터를 중심벡터(Cb2)로 선택한다. Next, the learning unit 400

Set to , and perform the 3rd learning that performs learning. During the 3rd learning, the learning unit 400 uses the same learning context as the learning context used in the 1st and 2nd learning. Accordingly, the learning unit 400 sets a label according to a known company of a learning context. During the tertiary learning, the learning unit 400 also performs the same classification on the vector space formed by the context authentication network 121 where the secondary learning is completed, that is, the central vector of the context vector of the same company. to set it as a label. For example, referring to FIG. 8 , a context vector having a median value among a plurality of context vectors of company A is selected as a center vector (Ca2), and a context vector having a median value among a plurality of context vectors of company B is a center vector ( Select Cb2).

이기 때문에 학습부(400)는 수학식 1로부터 수학식 4와 같은 손실함수를 도출하여 도출된 손실함수를 통해 레이블과 컨텍스트벡터 혹은 복수의 출력노드(O1 내지 Oy, 예컨대, y=4) 각각의 출력값의 차이인 손실함수의 손실값이 최소화되도록 컨텍스트인증망(121)의 가중치를 최적화한다. The learning unit 400 inputs the learning context to the context authentication network 121 . Then, the context authentication network 121 calculates a context vector by performing a plurality of operations to which a weight w is applied to the learning context.

Therefore, the learning unit 400 derives a loss function such as Equation 4 from Equation 1, and through the derived loss function, each of the label and context vector or a plurality of output nodes (O1 to Oy, for example, y = 4) The weights of the context authentication network 121 are optimized so that a loss value of a loss function, which is a difference between output values, is minimized.

In Equation 3, cy is a central vector that is a label. The tertiary learning procedure as described above is repeatedly performed using the same learning context as the plurality of learning contexts used in the first or second learning. In this tertiary learning, since learning is performed using the central vector (Ca2, Cb2) according to the result of the secondary learning, the correlation between context vectors of the same kind based on the central vector (Ca2, Cb2) is additionally learned. . Accordingly, the classification becomes clearer, and since the degree of correlation between a plurality of context vectors is learned, the plurality of context vectors can be used for calculation based on the correlation.

Next, the work management unit 200 according to an embodiment of the present invention will be described. 9 is a diagram for explaining a detailed configuration of a work management unit of an apparatus for constructing big data for sharing production information among a plurality of companies producing one finished product according to an embodiment of the present invention. 10 is a diagram for explaining the detailed configuration of a work authentication unit of an apparatus for constructing big data for sharing production information between a plurality of companies producing one finished product according to an embodiment of the present invention. 11 to 13 are diagrams for explaining the configuration of a work authentication network of a device for building big data for sharing production information between a plurality of companies producing one finished product according to an embodiment of the present invention. Referring to FIG. 9 , the work management unit 200 includes a work allocation unit 210 and a job certification unit 220 .

The task assigning unit 210 is for enabling the terminal 20 that has successfully authenticated to perform tasks such as searching, viewing, uploading, downloading, and updating production information. At this time, the work allocation unit 210 provides a workspace, which is a memory space physically separated from the area of the storage unit 12 in which the original production information is stored, to the terminal 20 that has succeeded in authentication, instead of the original production information. Protect production information by providing copies of production information.

More specifically, the task allocation unit 210 determines the authentication result of the connection processing unit 100, and if the authentication is successful, the context vector used for authentication of the terminal 20, that is, the size corresponding to the company to which the terminal 20 belongs. Allocates a working space, a volatile memory space of In addition, the work allocator 210 copies production information corresponding to a job performed by a company to which the terminal 20 belongs among production information stored in the storage unit 12 and loads it into an assigned work space. In addition, the work assignment unit 210 transmits an access response message including an access address of the assigned work space to the terminal 20 through the communication unit 11 .

When the terminal 20 accesses the work space through the communication unit 11, the work authentication unit 220 monitors the work performed by the terminal 20 in the work space to determine whether the work performed by the terminal 20 is a valid work. It is to determine whether If the work authentication unit 220 determines that the work is not legitimate, it may block the connection of the terminal 20 to the sharing server 10 . In addition, the work authentication unit 220 may transmit an alarm to a device used by a manager of the sharing server 10 through the communication unit 11 .

Referring to FIG. 10 , the work authentication unit 220 includes a pre-processing unit 221, a vector combiner 223 and a work authentication network 225.

The pre-processing unit 221 is for generating a plurality of log vectors L1 to Ln and time vectors T1 to Tn corresponding to each of the plurality of log vectors L1 to Ln. More specifically, after the terminal 20 accesses the workspace, logs are recorded in chronological order, and the pre-processing unit 221 sequentially extracts and collects a plurality of logs recorded in chronological order. A plurality of log vectors (L1 to Ln) are generated by extracting features of each of the plurality of logs, and the generated plurality of log vectors (L1 to Ln) are sorted and provided according to the time order in which the logs are recorded. In addition, the pre-processing unit 221 generates time vectors T1 to Tn representing log recording times corresponding to each of the plurality of log vectors L1 to Ln.

The vector combiner 221 combines the time vectors T1 to Tn corresponding to the plurality of log vectors L1 to Ln, and additionally combines the context vector F to a plurality of work log vectors arranged in time order. (X1=L1+T1+F to Xn=Ln+Tn+F).

The work authentication network 225 performs one or more calculations to which weights are applied in order of a plurality of stages S1 to Sn for a plurality of work log vectors Xt: X1 to Xn, so that the terminal 20 is in the work space. A task attribute vector (Y) representing the predicted task property is calculated by predicting the task to be performed. Then, the work authentication network 225 calculates a probability indicating whether the work performed by the terminal 20 in the work space is a work authorized to be performed by the terminal from the calculated work attribute vector Y, and outputs the calculated probability. do.

Referring to FIG. 11 , the task authentication network 225 is a deep neural network (DNN) and includes a task prediction network (WEN) and a discriminating network (DEN). The work prediction network (WEN) may include a recurrent neural network (RNN), long short-term memory models (LTSM), and a gated recurrent unit (GRU). The work prediction network (WEN) is composed of a plurality of stages (S1 to Sn), and includes a work input layer (WIN), a work hidden layer (WHL), and a work output layer (WOL). The length n of the stages S1 to Sn is predetermined.

The task input layer (WIL) is composed of a plurality of (n) inputs for predicting a task to be performed by the terminal 20 in the workspace and calculating a task attribute vector (Y) representing the properties of the predicted task. That is, a plurality of work log vectors X1 to Xn arranged in chronological order are input to the work input layer WIL. The working hidden layer (WHL) includes a plurality of hidden cells (HC) corresponding to the number of stages (n), and the plurality of hidden cells (HC) are weighted for a plurality of work log vectors (X1 to Xn). One or more operations are performed in a recursive manner to calculate the work attribute vector (Y). The work output layer (WOL) outputs the work attribute vector (Y) calculated by the hidden cell (HCn) of the last stage (Sn).

Meanwhile, referring to FIG. 12, the hidden cell HC is composed of one or more calculations to which weights (Weight, Wx, Wh, Wy) are applied. Here, the operation means an operation to which an activation function is applied. Activation functions may include sigmoid, hyperbolic tangent (tanh), exponential linear unit (ELU), rectified linear unit (ReLU), leaky ReLU, Maxout, Minout, and Softmax. In addition, in one hidden cell (HC), the weight is the input weight Wx applied to the work log vector (Xt), the state weight Wh applied to the state value Ht-1 of the previous stage, and the output value Yt applied to the input value Include the output weight Wy. For example, the following Equation 5 may be exemplified as an operation to which the weight applied to the hidden cell HC is applied.

Here, b is a threshold or bias. In particular, the tanh function and the ReLU function can be changed to other activation functions. Referring to Equation 1 and FIG. 9, each of the plurality of hidden cells HCt has a state value Ht- of the previous stage St-1 calculated by the hidden cell HCt-1 of the previous stage St-1. 1) and the state value (Ht) and output value (Yt) of the current stage (St) by performing an operation in which weights (W: Wh, Wx, Wy) are applied to the input value (Xt) of the own stage (St) can be calculated.

The work input layer WIN of the work prediction network WEN converts each of the plurality of work log vectors X1 to Xn arranged in chronological order into corresponding hidden cells of the work hidden layer WHL, that is, the first to nth It is input to the hidden pixels (HC1 to HCn). Then, each of the first to nth hidden cells HC1 to HCn of the working hidden layer WHL has a state value of the previous stage St-1 calculated by the hidden cell HCt-1 of the previous stage St-1 After calculating the state value (Ht) of the current stage (St) by performing an operation to which a weight is applied to (Ht-1) and the input value (Xt) of the stage (St) of itself (HCt), the calculated state The value (Ht) is transferred to the hidden cell (HCt+1) of the next stage (St+1). For example, the first hidden cell HC1 applies the state weight Wh to the initial state value H0 and applies the input weight Wx to the first work log vector X1 to obtain the first stage S1. State value H1 is calculated. In the case of the first hidden pixel HC1, since there is no previous stage, the initial state value H0 is used. Subsequently, the second hidden cell HC2 applies the state weight Wh to the state value H1 of the first stage S1, which is the previous stage, and applies the input weight Wx to the second work log vector X2. By applying, the state value H2 of the second stage S2 is calculated. And the third hidden cell HC3 applies the state weight Wh to the state value H2 of the second stage S2, which is the previous stage, and applies the input weight Wx to the third work log vector X3. This is an equation for calculating the state value H3 of the third stage S3. In this way, the plurality of hidden cells HCt performs an operation in which the state and input weights (Wh, Wx) are applied to the state value (Ht-1) of the previous stage and the work attribute vector (Xt), which is the input value of the current stage. , the state value (Ht) of the current stage is calculated, and the calculated state value (Ht) is transferred to the next stage (St+1).

During the execution of such an operation, the n-th hidden pixel (HCn), which is the last hidden pixel, applies the state weight (Wh) to the state value (Hn-1) of the n-1th stage (Sn-1), which is the previous stage, to obtain the current stage. The state value Hn of the n-th stage Sn is calculated. Then, the n-th hidden cell HCn performs an operation of applying an output weight Wy to the state value Hn of the n-th stage Sn, which is the current stage, to obtain an output value, that is, a work attribute vector Y yield

As such, in the work prediction network (WEN) according to an embodiment of the present invention, the work input layer (WIL) is a plurality of work log vectors (X1 to Xn) arranged in time order corresponding to each of the plurality of stages (S1 to Sn). ) is input for each of the plurality of stages S1 to Sn. Then, the plurality of hidden cells HC1 to HCn-1, except for the last hidden cell HCn of the working hidden layer WHL, have a state value Ht-1 of the previous stage St-1 and the current stage St. After calculating the state value (Ht) of the current stage by performing an operation to which the state and input weights (Wx, Wh) are applied to the work log vector (Xt), which is an input value, the calculated state value is converted to the next stage (St+ 1) forwarded to Then, the last hidden cell (HCn) has a state and an input weight (Wx) for the state value (Hn-1) of the previous stage (Sn-1) and the n-th work log vector (Xn), which is the input value of the current stage (Sn). , Wh) is applied to calculate the state value (Hn) of the current stage (Sn), and then the output weight (Wy) is applied to the calculated state value (Hn) of the current stage so that the terminal 20 can A work attribute vector (Y) representing the attributes of work performed in the work space is calculated.

12 and 13, the work output layer (WOL) of the work prediction network (WEN) has a plurality of attribute nodes (Y1 to Yp), and each of the plurality of attribute nodes (Y1 to Yp) is a work attribute vector (Y = The value (y1, y2, ..., yp) of each digit of [y1, y2, ..., yp]) is input as an attribute value. That is, the work output layer (WOL) divides the work attribute vector (Y) into a plurality of attribute values (y1, y2, ..., yp). Further, the work output layer (WOL) outputs a work attribute vector (Y) representing the attribute of the work performed by the terminal 20 in the work space, and the attribute values (y1, y2, ..., yp) is output to the discriminant network (DEN). That is, the work output layer (WOL) outputs the attribute values (y1, y2, ..., yp) of each of the plurality of attribute nodes (Y1 to Yp) to the discrimination network (DEN).

Referring to FIGS. 12 and 13 , the discrimination network DEN consists of one layer. That is, the discrimination network DEN includes the discrimination layer DEL. The determination layer (DEL) performs a plurality of operations to which weights are applied to the calculated work attribute vector (Y), and the probability that the work performed by the terminal 20 in the work space is the work authorized to be performed by the terminal 20 and a probability that the terminal 20 is not authorized to perform the task.

The discrimination layer DEL includes two discrimination nodes D1 and D2. The first determination node D1 corresponds to a job authorized for the terminal 20 to perform. The second determination node D2 corresponds to a task that the terminal 20 is not authorized to perform. In the discriminant layer (DEL), the two discriminant nodes (D1 and D2) are weighted through an activation function for the task attribute vector (Y) of the work prediction network (WEN) or a plurality of attribute values (y1 to yp). is performed to calculate the discriminant value of each of the two discriminant nodes D1 and D2. As described above, the first determination node D1 corresponds to a job that the terminal 20 is authorized to perform, and the second determination node D2 corresponds to a job that the terminal 20 is not authorized to perform. do. In addition, the first determination value, which is an output of the first determination node D1, indicates a probability that the operation performed by the terminal 10 is a job authorized for the terminal 20 to perform. And, the second determination value, which is an output of the second determination node D2, represents a probability that a job to be performed by the terminal 10 is a job to which the terminal 20 is not authorized to perform. Accordingly, the work authentication network 225 performs the work authorized for the terminal 20 to perform the work to be performed by the terminal 10 from the discrimination values of each of the first determination node D1 and the second determination node D2. You can judge whether or not to do it. For example, if the respective discriminant values of the first discriminant node D1 and the second discriminant node D2 are 0.282 and 0.718, the operation performed by the terminal 10 according to the first discriminant value is applied to the terminal 20 to be performed. It indicates that the probability that the task has been performed is 28%, and the probability that the task performed by the terminal 10 according to the second determination value is a task that the terminal 20 is not authorized to perform is 72%. Accordingly, the work authentication network 225 may determine that the terminal 10 is currently performing a job that the terminal 20 is not authorized to perform.

The learning unit 400 may teach the work authentication network 225 to calculate a discrimination value indicating whether the current terminal 10 is performing a job that the terminal 20 is not authorized to perform. A learning method for this work authentication network 225 will be described.

The learning unit 400 may acquire a plurality of actually used field data by actually using a plurality of terminals 20 belonging to a plurality of companies producing one finished product. In addition, the learning unit 400 may obtain a context vector F from field data through the context authentication network 121 for which learning has been completed. In addition, the learning unit 400 may obtain log vectors L1 to Ln and time vectors T1 to Tn in addition to the context vector F. Accordingly, the learning unit 400 generates a plurality of work log vectors (Xt: X1 to Xn) by combining the context vector (F), log vectors (L1 to Ln), and time vectors (T1 to Tn) obtained from the field data. A test group of learning data can be prepared by creating a test group. Also, the learning unit 400 may prepare a control group of the learning data by modifying a part of the test group of the learning data. When the test group and control group of the learning data are prepared, the learning unit 150 sets [D1, D2] = (1, 0) as a label for the test group of the learning data, and [ D1, D2] = (0, 1).

Then, the learning unit 400 inputs learning data to the work certification network 225 . Then, the work prediction network (WEN) of the work certification network 225 weights (Wx, Wh, Wy) in order of a plurality of stages (S1 to Sn) for a plurality of work log vectors (Xt: X1 to Xn) for learning ) is applied to predict a task to be performed by the terminal 20 in the workspace, and a task attribute vector (Y) representing properties of the predicted task is calculated. In addition, the determination layer (DEL) of the work authentication network 225 performs a plurality of operations to which a weight (w) is applied to the calculated work attribute vector (Y), and the work performed by the terminal 20 in the work space is authorized. The probability of an approved operation and the probability of an unapproved operation are calculated as discriminant values, and the calculated discriminant value is output. Then, the learning unit 400 calculates the binary cross entropy loss, which is the difference between the previously set label and the discrimination value, and the weight of the work certification network 225 through an optimization algorithm so that the calculated loss is minimized. optimize The learning unit 400 may train the work certification network 225 by repeating a procedure of optimizing the weight of the work certification network 225 using a plurality of different learning data including a test group and a control group a plurality of times. .

Next, a big data construction method for sharing production information between a plurality of companies producing one finished product according to an embodiment of the present invention will be described. 14 is a flowchart illustrating a big data construction method for sharing production information between a plurality of companies producing one finished product according to an embodiment of the present invention. The method for sharing production information of FIG. 14 starts in a state in which the learning unit 400 has completed learning of the previously described context authentication network 121 and work authentication network 225 .

Referring to FIG. 14 , the storage unit 12 of the sharing server 10 stores production information in step S110. The storage of production information may be performed by the recording processing unit 300 of the control unit 13 of the sharing server 10 .

In the state where the production information is stored, the connection processing unit 100 of the control unit 13 of the sharing server 10 may receive an access request message through the communication unit 11 in step S120. The access request message includes an account including an identifier and password of a user of the terminal 20 and a context of the terminal 20 . In particular, the context includes the device identifier of the terminal, the type of terminal, access location, access time, access IP, access port, access network type, access network identification information, and an application used for access.

After receiving the access request message, the access processing unit 100 performs authentication on whether the terminal 20 has permission to access production information of a plurality of companies based on the access request message in step S130. At this time, the account authentication unit 110 of the connection processing unit 100 compares the account including the user's identifier (ID) and password (PW) of the terminal 20 with the user's account previously stored in the storage unit 12. It is authenticated whether the user's account transmitted by the terminal 20 is a pre-registered account. Further, the context authentication unit 120 of the connection processing unit 100 authenticates whether the context of the terminal 20 matches the registered account of the terminal 20 using the deep neural network context authentication network 121 . A plurality of terminals 20 belonging to the same company, that is, a plurality of terminals 20 used by executives and employees of the same company have contexts having a high degree of correlation with each other. Accordingly, the context authentication network 121 is learned by the learning unit 400 to identify the company to which the terminal 20 belongs through the context of the terminal 20 . Therefore, the context authentication unit 120 identifies the company using the context authentication network 121, and then determines whether the account of the terminal 20 is an account assigned to the executives and employees of the identified company, thereby determining the context of the terminal 20. It is possible to authenticate whether or not matches the registered account of the terminal 20 .

Next, the task allocator 210 of the control unit 13 identifies the company to which the terminal 20 belongs from the context vector used for authentication of the terminal 20 that has succeeded in authentication in step S140, and provides information corresponding to the identified company. Allocates a working space, which is a volatile memory space of size . In addition, the work allocator 210 copies the production information corresponding to the work performed by the company to which the terminal 20 belongs among the production information stored in the storage unit 12 in step S150 and stores the production information in the previously allocated work space. load a copy Then, the work assignment unit 210 transmits an access response message including the access address of the work space allocated in step S160 to the terminal 20 through the communication unit 11 . In this way, the work space allocated to the terminal 20 is a memory space physically separated from the area of the storage unit 12 in which the original production information is stored, and provides a copy of the production information instead of the original production information. In this way, the original of the production information can be protected.

When the terminal 20 accesses the workspace through the communication unit 11, the work authentication unit 220 performs operations performed by the terminal 20 in the workspace in step S170, such as search, reading, upload, download, and update. etc. can be monitored. At this time, as described through FIGS. 10 to 13 in step S180, the work authentication unit 220 may determine whether or not the work performed by the terminal 20 is a legitimate work using the work authentication network 225. have. If the work authentication unit 220 determines that the work is not legitimate, it may block the connection of the terminal 20 to the sharing server 10 . In addition, the work authentication unit 220 may transmit an alarm to a device used by a manager of the sharing server 10 through the communication unit 11 .

On the other hand, if the work performed by the terminal 20 in the workspace according to the determination of the work authentication unit 220 is a work authorized to be performed by the terminal 20, the recording processing unit 300 records the terminal 20 in step S190. If production information corrected according to the result of work performed in this workspace exists, the production information in the storage unit 12 is updated with the corrected production information.

On the other hand, the method according to the embodiment of the present invention described above may be implemented in the form of a program readable by various computer means and recorded on a computer-readable recording medium. Here, the recording medium may include program commands, data files, data structures, etc. alone or in combination. Program instructions recorded on the recording medium may be those specially designed and configured for the present invention, or those known and usable to those skilled in computer software. For example, recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magneto-optical media such as floptical disks ( It includes hardware devices specially configured to store and execute program instructions, such as magneto-optical media and ROM, RAM, flash memory, etc. Examples of program commands may include high-level language wires that can be executed by a computer using an interpreter, as well as machine language wires such as those produced by a compiler. These hardware devices may be configured to act as one or more software modules to perform the operations of the present invention, and vice versa.

The present invention has been described above using several preferred examples, but these examples are illustrative and not limiting. As such, those skilled in the art to which the present invention belongs will understand that various changes and modifications can be made according to the doctrine of equivalents without departing from the spirit of the present invention and the scope of rights set forth in the appended claims.

10: shared server 11: communication unit
12: storage unit 13: control unit
20: terminal 100: connection processing unit
110: account authentication unit 120: context authentication unit
121: context authentication network 200: work management unit
210: work allocation unit 220: work certification unit
221: pre-processing unit 223: coupling unit
225: work authentication network 300: record processing unit
400: learning unit

Claims (12)

In the big data construction device,
A communication unit for communication with terminals of a plurality of companies producing one finished product;
A storage unit for integrating and storing production information, which is information about a process of making the finished product from raw materials for the plurality of companies to produce the finished product, and is information to be shared by the plurality of companies; and
When receiving an access request message from any one of the plurality of terminals through the communication unit, based on the received access request message, whether the terminal has permission to access the production information of the plurality of companies an access processing unit that performs authentication for;
As a result of the authentication, if the authentication is successful, a work space, which is a volatile memory space of a size corresponding to the terminal, is allocated, and among the production information of the storage unit, production information corresponding to a job performed by a company to which the terminal belongs is copied. a work management unit including a work allocating unit that loads the allocated workspace and transmits an access response message including an access address of the allocated workspace to the terminal through the communication unit;
Including,
The connection request message
It includes an identifier and password of a user of the terminal and a context of the terminal,
The context is
Includes device identifier of the terminal, type of terminal, access location, access time, access IP, access port, type of access network, identification information of the access network, and application used for access,
The connection processing unit
an account authentication unit authenticating whether or not the user's account is a pre-registered account through the user's identifier and password of the terminal; and
Authenticating whether the context corresponds to the registered account using a context authentication network including a plurality of layers including a plurality of calculations that are input to the next layer by applying a weight to the operation result of one layer context authentication unit;
Including,
The context authentication network
an input layer receiving the context;
a convolution layer constituting a first feature map by performing a convolution operation on the context of the input layer through a first filter composed of a weight matrix;
a pooling layer constituting a second feature map by performing a pooling operation on the first feature map of the convolution layer through a second filter composed of a weight matrix;
a flattening layer transforming the second feature map of the pooling layer and arranging them into one column;
Including a plurality of calculation nodes, calculating a calculation value corresponding to each of the plurality of calculation nodes by performing an operation through an activation function to which a weight is applied to a second feature map in which each of the plurality of calculation nodes is arranged in a column fully connected layer; and
It includes a plurality of output nodes corresponding to each of a plurality of companies, and performs an operation through an activation function on the calculation values of the plurality of calculation nodes of the fully connected layer so that the terminal having the context is sent to each of the plurality of companies. an output layer that calculates a context vector indicating a probability of belonging and outputs the calculated probability as an output value;
characterized in that it includes
Big data building device. delete delete According to claim 1,
The device
Set labels according to known companies for learning contexts where companies are known;
After entering the learning context into the context authentication network,
When the context authentication network calculates an output value through a plurality of calculations to which weights are applied to the learning context,
A loss function that calculates a loss value that is the difference between the set label and the output value

Optimizing the weight of the context authentication network so that the loss value of

A learning unit that performs learning on the context authentication network while changing the value of 1, 0.5 and 0
The Ecm represents a loss function,
The hy is a one-hot encoding vector representing a known company according to the label,
The fy is the output value,
The cy is a central vector among a plurality of context vectors of companies known according to labels,
Where y is the company index,
Wherein N is the maximum value of y,
remind

is a hyperparameter, characterized in that
Big data building device. According to claim 1,
The work management department
When the terminal accesses the work space through the communication unit, a work authentication unit that monitors the work performed by the terminal in the work space and determines whether the work performed by the terminal is a valid work;
characterized in that it further comprises
Big data building device. According to claim 5,
The work certification department
After the terminal accesses the workspace, a plurality of logs recorded according to time sequence are sequentially extracted, a plurality of log vectors representing characteristics of each of the plurality of extracted logs are generated, and the plurality of log vectors are converted into logs. a pre-processing unit that sorts and provides data according to recorded time order and generates a time vector indicating a time when the log was recorded;
a vector combining unit generating a plurality of work log vectors arranged in chronological order by combining a time vector corresponding to the plurality of log vectors and the context vector; and
One or more operations to which weights are applied are performed on the plurality of work log vectors in order of a plurality of stages to calculate a work attribute vector representing the attribute of a work performed by the terminal in the work space, and the calculated work attribute a work certification network that calculates a probability indicating whether a task performed by the terminal in the workspace is a task authorized to be performed by the terminal from a vector, and outputs the calculated probability;
characterized in that it includes
Big data building device. According to claim 6,
The work authentication network
It consists of a plurality of stages arranged in sequence,
a work input layer which receives a plurality of work log vectors arranged in chronological order corresponding to the plurality of stages for each of the plurality of stages;
The state value of the current stage is calculated by performing an operation to which the state and input weight are applied to the state value of the previous stage and the work log vector, which is the input value of the current stage, and then the calculated state value is passed to the next stage. Including a hidden pixel,
The last hidden cell among the plurality of hidden pixels calculates the state value of the current stage by performing an operation to which state weights are applied to the state value of the previous stage, and then applies the output weight to the calculated state value of the current stage to obtain an output value a task hidden layer that calculates a task attribute vector representing properties of a task performed by the terminal in the workspace;
a job output layer that divides the job attribute vector into a plurality of attribute values and outputs a plurality of attribute values; and
A probability that a task performed by the terminal in the workspace by performing a plurality of operations to which weights are applied to the plurality of attribute values is a task authorized to be performed by the terminal and a probability that the task is not authorized to be performed by the terminal Discrimination layer for calculating a discriminant value representing ;
characterized in that it includes
Big data building device. According to claim 7,
The device
If the work performed in the work space is a work authorized to be performed by the terminal, if there is production information corrected according to the result of the work performed by the terminal in the work space, production of the storage unit is performed with the corrected production information. a recording processing unit that updates information;
characterized in that it further comprises
Big data building device. In the big data construction method,
When there are a plurality of companies producing one finished product, the plurality of companies are information on the process of making the finished product from raw materials to produce the finished product, and the production information, which is information for the plurality of companies to share, is integrated Receiving a connection request message from a terminal of one of the plurality of companies by the connection processing unit in the stored state;
performing authentication on whether or not the terminal has authority to access the production information of the plurality of companies based on the connection request message received by the connection processing unit;
Allocating, by a work management unit, a work space, which is a volatile memory space of a size corresponding to the terminal, if the authentication is successful as a result of the authentication;
copying, by the work manager, production information corresponding to a work performed by a company to which the terminal belongs, among the production information, and loading the copy into the assigned work space;
providing, by the work manager, an access response message including an access address of the assigned work space to the terminal;
Including,
The connection request message
It includes an identifier and password of a user of the terminal and a context of the terminal,
The context is
Includes device identifier of the terminal, type of terminal, access location, access time, access IP, access port, type of access network, identification information of the access network, and application used for access,
The step of performing the authentication is
authenticating whether the user's account is a pre-registered account through the user's identifier and password of the terminal by the access processing unit; and
The connection processing unit checks whether the context corresponds to the registered account by using a context authentication network including a plurality of layers including a plurality of calculations that are input to the next layer by applying weights to calculation results of one layer. certifying whether or not;
Including,
After the work management unit provides an access response message including an access address of the allocated work space to the terminal,
When the terminal accesses the work space, the work authentication unit of the work management unit monitors the work performed by the terminal in the work space and determines whether the work performed by the terminal is a legitimate work;
As a result of the determination, if the work performed in the work space is a work authorized for the terminal to be performed, if production information corrected according to the result of the work performed by the terminal in the work space exists, the recording processor modifies the modified production information. Updating pre-stored production information with production information;
characterized in that it further comprises
How to build big data. delete According to claim 9,
Before receiving the access request message,
setting, by a learning unit, a label according to a known company for a learning context in which the company is known;
After the learning unit inputs the learning context to the context authentication network, when the context authentication network calculates an output value through a plurality of calculations to which weights are applied to the learning context,
A loss function that calculates a loss value that is the difference between the set label and the output value

Optimizing the weight of the context authentication network so that the loss value of

Further comprising: performing learning on the context authentication network while changing the value of
The Ecm represents a loss function,
The hy is a one-hot encoding vector representing a known company according to the label,
The fy is the output value,
The cy is a central vector among a plurality of context vectors of companies known according to labels,
Where y is the company index,
Wherein N is the maximum value of y,
remind

is a hyperparameter, characterized in that
How to build big data.
delete

Images