KR102665023B1 - Method and Apparatus for Determining a Monitoring Variable for a Target System - Google Patents

Abstract

A method for determining observation variables for a target system is disclosed. The disclosed method includes collecting measurement values from sensors for the target system, defining variables representing measurement values collected from each of the sensors, and performing hierarchical clustering on the variables. generating at least one cluster, and performing dimension reduction on each of the at least one cluster to determine at least one observation variable for the target system. there is.

Description

{Method and Apparatus for Determining a Monitoring Variable for a Target System}

The disclosure below relates to system monitoring techniques.

Recently, the industry is accelerating the development of smart ships and smart solutions with the goal of digital transformation in line with the development of artificial intelligence and Internet of Things technologies. For digital transformation, the industry replaces analog-type measuring equipment with digital-type sensors and creates an environment where data collected using these digital-type sensors at a certain location can be transmitted to another location through a satellite communication network to perform analysis. has been building. And by applying intelligent technologies such as machine/deep learning and advanced analysis (big data analysis) to the collected data, many technologies are developed with the goal of enabling beginners or intelligent models to make unmanned decisions that used to be made by experienced experts. It is being developed.

Taking the shipbuilding industry as an example, ship engines are the most important system on ships as a power/electricity source for ships, so the development of digitalization and intelligent failure detection and diagnosis solutions for ship engines is actively underway. However, in the case of an engine, it is made up of countless elements from the lowest components to the highest engine system, and all elements operate in an organic cycle, making it difficult to develop a data-based intelligent smart solution.

In addition, in the case of marine engines, it is very difficult to collect high-quality data due to the extreme environment of marine operation and the nature of operating conditions changing very rapidly depending on driving/weather conditions, etc. As a result, there are limits to management through the development of a complete AI model. Therefore, currently, a professional controller at the land support center observes data received from sensors in real time, determines abnormal signs, and analyzes the results. However, in the case of a ship engine, there are dozens/hundreds of various types of sensors attached to just one engine, so there is a limit to observing multiple engines and fleets.

The problem to be solved by the present disclosure is to provide a technology that allows determining and using a smaller number of observation variables than the number of sensors connected to the target system to observe the target system.

The problems to be solved by the present disclosure are not limited to the problems mentioned above, and other problems not mentioned will be clearly understood by those skilled in the art from the description below.

According to one feature of the present disclosure, a method for determining observed variables for a target system is provided. The method includes collecting measurement values from sensors for the target system, defining variables representing measurement values collected from each of the sensors, and performing hierarchical clustering on the variables. generating at least one cluster, and performing dimension reduction on each of the at least one cluster to determine at least one observation variable for the target system. there is.

In one embodiment, the step of generating at least one cluster by performing hierarchical clustering on the variables includes using a plurality of normalization methods for the variables, generating a plurality of cluster models for the variables by applying a plurality of distance measures and a plurality of linkage functions for clustering the variables, a cope associated with each of the plurality of cluster models It includes calculating a value of a cophenetic correlation coefficient, and determining a cluster model associated with the largest cophenetic correlation coefficient among the cophenetic correlation coefficient values as the optimal cluster model.

In one embodiment, a plurality of normalization methods for the variables, a plurality of distance measures between the variables, and a plurality of linkage functions for clustering the variables are applied to the variables. Generating a plurality of cluster models includes a plurality of combinations, each of the plurality of combinations being one of the plurality of normalization methods, one of the plurality of distance measures, and one of the plurality of link functions. configured - defining , and applying each of the plurality of combinations to generate a cluster model that hierarchically clusters the variables and defines cophenetic distances between the variables.

In one embodiment, the step of calculating the value of the cophenetic correlation coefficient associated with each of the plurality of cluster models is calculated using the equation below:

- Here i and j are indices representing variables, is a variable calculated using any one normalization method and any one distance measure applied to generate the corresponding cluster model. and variable represents the distance between represents the average of the distances, is a variable defined by the corresponding cluster model and variable represents the copenetic distance between represents the average of the copenetic distances - and includes calculating the value of the copenetic correlation coefficient associated with the corresponding cluster model using -.

In one embodiment, the determined optimal clustering model defines a cluster tree, and performing hierarchical clustering on the variables to generate at least one cluster includes defining the cluster tree. It further includes generating the at least one cluster by cutting at a selected cutting position.

In one embodiment, the plurality of normalization methods include at least one of a center method, a standard deviation scaling method, a Z-score method, and a minimum-maximum method. Includes.

In one embodiment, the plurality of distance measures include Euclidean distance, Mahalanobis distance, Minkowski distance, City block distance, and Pearson correlation distance. It includes at least one of (Pearson correlation distance) and Spearman correlation distance.

In one embodiment, the plurality of linkage functions include a nearest neighbor linkage function, a farthest neighbor linkage function, an average linkage function, a weighted average linkage function, and Contains at least one of the Ward linkage functions.

In one embodiment, the step of determining at least one observed variable for the target system by performing dimension reduction on each of the at least one cluster includes linearly transforming the variables belonging to the corresponding cluster. Providing a smaller number of principal components than the number of variables and determining the provided principal component as the at least one observed variable.

In one embodiment, the step of linearly transforming the variables belonging to the corresponding cluster to provide a smaller number of principal components than the number of variables and determining the provided principal component as the at least one observed variable includes the equation below:

- here is a matrix composed of variables belonging to the corresponding cluster, is a principal component matrix composed of principal components, Is cast It is an eigenvector that is linearly converted to - from Sort the variance values of the main components in descending order. It includes steps to find .

In one embodiment, the step of linearly transforming the variables belonging to the corresponding cluster to provide a smaller number of principal components than the number of variables and determining the provided principal component as the at least one observed variable includes: Among the main components of the equation below:

- here Is is the eigenvalue of the main component, and p is is the number of main components, Is is the average of the eigenvalues of the main components of represents d main components to be selected - and further includes the step of selecting d main components according to .

In one embodiment, the step of linearly transforming the variables belonging to the corresponding cluster to provide a smaller number of principal components than the number of variables and determining the provided principal component as the at least one observed variable includes the equation below:

- here Is represents the variance of the ith principal component of Is represents the variance of the jth principal component of represents the cumulative percentage of variance explained - the cumulative percentage of variance explained defined by determining the minimum i such that is greater than or equal to a selected percentage, and It further includes determining the determined minimum i-th main components among the main components as the at least one observed variable.

According to another feature of the present disclosure, an apparatus for determining an observed variable for a target system is provided. The apparatus includes a preprocessing module configured to collect measurement values from sensors for the target system and define variables representing measurement values collected from each of the sensors, and perform hierarchical clustering on the variables to determine at least one It may include a clustering module configured to generate clusters, and a dimensionality reduction module configured to perform dimensionality reduction on each of the at least one cluster to determine at least one observed variable for the target system.

In one embodiment, the clustering module applies a plurality of normalization methods to the variables, a plurality of distance measures between the variables, and a plurality of link functions to cluster the variables to form a plurality of clusters for the variables. Create a model, calculate the value of the copenetic correlation coefficient associated with each of the plurality of cluster models, and select the cluster model associated with the largest copenetic correlation coefficient value among the copenetic correlation coefficient values as the optimal cluster model. further configured to decide.

In one embodiment, the clustering module comprises a plurality of combinations, each of the plurality of combinations consisting of one of the plurality of normalization methods, one of the plurality of distance measures, and one of the plurality of link functions. and further configured to apply each of the plurality of combinations to hierarchically cluster the variables and generate a clustering model defining copenetic distances between the variables.

In one embodiment, the clustering module uses the equation below:

- Here i and j are indices representing variables, is a variable calculated using any one normalization method and any one distance measure applied to generate the corresponding cluster model. and variable represents the distance between represents the average of the distances, is a variable defined by the corresponding cluster model above and variable represents the copenetic distance between represents the average of the copenetic distances - is further configured to calculate the value of the copenetic correlation coefficient associated with the corresponding cluster model using -.

In one embodiment, the determined optimal clustering model defines a cluster tree, and the clustering module is further configured to cut the cluster tree at a selected cutting position to generate the at least one cluster.

In one embodiment, the plurality of normalization methods include at least one of a centering method, a standard deviation scaling method, a Z-score method, and a minimum-maximum method.

In one embodiment, the plurality of distance measures include at least one of Euclidean distance, Mahalanobis distance, Minkowski distance, city block distance, Pearson correlation distance, and Spearman correlation distance.

In one embodiment, the plurality of link functions include at least one of a shortest link function, a longest link function, an average link function, a weighted average link function, and a Ward link function.

In one embodiment, the dimensionality reduction module is further configured to linearly transform variables belonging to the corresponding cluster to provide a smaller number of principal components than the number of variables and determine the provided principal components as the at least one observed variable.

In one embodiment, the dimensionality reduction module uses the equation below:

- here is a matrix composed of variables belonging to the corresponding cluster, is a principal component matrix composed of principal components, Is cast It is an eigenvector that linearly transforms from - to Sort the variance values of the principal components in descending order. It is further configured to find .

In one embodiment, the dimensionality reduction module: Among the main components of the equation below:

- here Is is the eigenvalue of the principal component of , and p is is the number of main components, Is is the average of the eigenvalues of the main components of represents d main components to be selected - and is further configured to select d main components according to .

In one embodiment, the dimensionality reduction module uses the equation below:

- here Is represents the variance of the ith principal component of Is represents the variance of the jth principal component of represents the percentage of cumulative variance explained - the percentage of cumulative variance explained defined by Determine the minimum i such that is greater than or equal to a selected percentage, It is further configured to determine the determined minimum ith principal components among the principal components as the at least one observed variable.

According to the disclosed embodiments, there is a technical effect of being able to observe the target system using a smaller number of observation variables than the number of sensors connected to the target system.

1 shows a block diagram of one embodiment of an apparatus for determining an observed variable for a target system, connected to sensors for the target system.
Figure 2 is a diagram illustrating an example of a dendrogram tree schematizing a clustering module generated by a clustering module.
FIG. 3 is a diagram illustrating a method of performing cutting in the dendrogram tree of FIG. 2.
FIG. 4 is a flowchart illustrating an embodiment of a method for determining observation variables for a target system.

Specific structural or functional descriptions of the embodiments are disclosed for illustrative purposes only and may be changed and implemented in various forms. Accordingly, the actual implementation form is not limited to the specific disclosed embodiments, and the scope of the present disclosure includes changes, equivalents, or substitutes included in the technical idea described in the embodiments.

Although terms such as “first” or “second” may be used to describe various components, these terms should be interpreted only for the purpose of distinguishing one component from another component. For example, a “first component” may be named a “second component” and similarly, a “second component” may also be named a “first component”.

When a component is referred to as being “connected” to another component, it should be understood that it may be directly connected or connected to the other component, but that other components may exist in between.

Singular expressions include plural expressions unless the context clearly dictates otherwise. In the present disclosure, terms such as "comprise" or "have" are intended to designate the presence of the described features, numbers, steps, operations, components, parts, or combinations thereof, and are intended to indicate the presence of one or more other features or numbers, It should be understood that this does not exclude in advance the possibility of the presence or addition of steps, operations, components, parts, or combinations thereof.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related technology, and should not be interpreted in an idealized or overly formal sense unless explicitly defined in the present disclosure. No.

Hereinafter, embodiments will be described in detail with reference to the attached drawings. In the description with reference to the accompanying drawings, identical components will be assigned the same reference numerals regardless of the reference numerals, and overlapping descriptions thereof will be omitted.

1 shows a block diagram of one embodiment of an apparatus for determining an observed variable for a target system, connected to sensors for the target system.

As shown in FIG. 1 , apparatus 150 for determining observation variables for a target system may include a preprocessing module 152 . The preprocessing module 152 may be connected to a plurality of sensors 140, where the plurality of sensors 140 are connected to the target system 120 and various types of sensors for observing the operating state of the target system 120. It can be. The target system 120 may be composed of a plurality of subsystems. In this case, the operating state of any one subsystem can be observed by at least one sensor among the plurality of sensors 140. In one embodiment, the target system 120 is a ship's engine, and in this embodiment, the plurality of sensors 140 include pressure sensors and temperature sensors installed at various locations at the engine inlet and pressure sensors installed at various locations at the outlet of the engine. sensors and temperature sensors. The plurality of sensors 140 may partially include sensors that are installed in similar locations and perform similar functions. The preprocessing module 152 may be configured to collect measurement values from the plurality of sensors 140 and define variables representing the measurement values collected from each of the plurality of sensors 140 . The preprocessing module 152 collects measured values over several months from a temperature sensor installed at a specific location at the inlet of the engine, for example, and uses the collected measured values as variables. By assigning to a variable It can be configured to consist of a multi-dimensional vector. The preprocessing module 120 may be configured to define the number of variables corresponding to the number of the plurality of sensors 140, and store and manage measured values for each of the defined variables in association with the corresponding variable. Although not shown, the preprocessing module 120 may include A/D converters for A/D conversion of measured values from the plurality of sensors 140.

Apparatus 150 may further include a clustering module 154 communicatively connected to preprocessing module 152. The clustering module 154 may be configured to generate at least one cluster by performing hierarchical clustering on variables. To this end, the clustering module 154 may be configured to generate a clustering model. A cluster model may be a model that defines which variable belongs to one cluster with which variable. According to the cluster model, variables with similar statistical characteristics (patterns) of data are grouped into the same cluster, and if not, they are grouped into different clusters. The cluster model is a model created through the process of grouping similar clusters one by one, starting from a single cluster corresponding to the number of variables defined in the preprocessing module 152, and grouping similar clusters one by one until these single clusters are finally grouped into one cluster. It can be. The cluster model created in this way can be schematized into a hierarchical tree structure called a ‘dendrogram’. To create a cluster model, first perform normalization such as scaling to ignore unit differences between variables, determine the distance measure, calculate the distance between normalized variables, and calculate the distance between the calculated variables. The process of determining whether to combine the variables into one cluster is performed by determining the proximity between the variables and applying a linkage function based on the determined proximity. Here, the closer the distance between variables means that the statistical characteristics (patterns) of the data are similar, so there is a greater possibility that they will be combined into the same cluster. The greater the distance between variables, the more dissimilar the statistical characteristics (patterns) of the data are, so they are more likely to be combined into the same cluster. The likelihood of merging into a cluster is low. Additionally, different cluster models can be created depending on which normalization method is adopted, which distance measure is adopted, and which link function is adopted. The clustering module 154 according to the present disclosure may be configured to select an optimal cluster model from among several cluster models that can be generated.

The clustering module 154 may be configured to generate a plurality of cluster models for variables by applying a plurality of normalization methods to the variables, a plurality of distance measures between variables, and a plurality of link functions for clustering the variables. . The plurality of normalization methods may include at least one of the center method, standard deviation scaling method, Z-score method, and minimum-maximum method, but the normalization method It should be recognized that the types are not limited to this. A plurality of distance measures include Euclidean distance, Mahalanobis distance, Minkowski distance, City block distance, Pearson correlation distance, and It may include at least one of Spearman correlation distances, but it should be recognized that the type of distance measure is not limited to this. The plurality of linkage functions include the nearest neighbor linkage function, the farthest neighbor linkage function, the average linkage function, the weighted average linkage function, and the ward linkage. It may include at least one of the functions, but it should be recognized that the type of connection function is not limited to this. The clustering module 154 may be configured to form a plurality of combinations by taking one from among a plurality of normalization methods, a plurality of distance measures, and a plurality of connection functions and combining them. For example, if the number of normalization methods is 3, the distance measure is 3, and the link function is 3, the clustering model 154 can form 27 combinations. The clustering module 154 may be configured to hierarchically cluster variables by applying each of the plurality of combinations and generate a cluster model that defines cophenetic distances between variables.

Figure 2 is a diagram illustrating an example of a dendrogram tree schematizing a clustering module generated by a clustering module.

Referring to Figure 2, the process of creating a cluster model is briefly explained as follows. First, variables according to the determined normalization method and distance measure inside After finding the distances between them, the closest class pair of and and Each pair of { , } and { , } Create clusters. in this case class distance between and and Since the distance between them is the same, clustering is performed on them simultaneously. Then, the variable closest to the created cluster is added to the existing cluster. Accordingly { , , } Clusters are created and finally { , , , , } Clustering is done with one cluster, and the cluster tree is completed. When the cluster tree is completed, cophenetic distances between variables are also calculated. It should be noted that the copenetic distance is a distance that reflects the characteristics of the cluster calculated between two variables, so it may be different from the distance between two variables. class Because it is different class distance between and and distance between are different from each other class copenetic distance between and and distance between may be the same. However, it should be noted that even if any of the normalization method, distance measure, and link function applied to clustering changes, the variable values themselves, the distance between variables, and/or the clustering method will change, and the copenetic distance will also change.

Referring again to FIG. 1, the clustering module 154 calculates the value of the cophenetic correlation coefficient associated with each of the plurality of cluster models, and calculates the value of the cophenetic correlation coefficient, which is the largest among the cophenetic correlation coefficient values. It may be further configured to determine the clustering model associated with the value as the optimal clustering model. The clustering module 154 can be configured to calculate the value of the copenetic correlation coefficient associated with each cluster model using Equation 1 below.

Here, i and j are indices representing variables, is a variable calculated using any one normalization method and any one distance measure applied to create the corresponding cluster model. and variable represents the distance between represents the average of the distances, is a variable defined by the corresponding cluster model and variable represents the copenetic distance between represents the average of the copenetic distances.

The copenetic correlation coefficient is a linear correlation coefficient between the distances between variables and the copenetic distances and has a value between 0 and 1. The copenetic correlation coefficient represents the relationship between the copenetic distances calculated from the cluster model and the distances between the variables used to create the cluster model. Therefore, the closer the value of the copenetic correlation coefficient is to 1, the better the data the cluster model is. It can be seen as a more accurate reflection of the characteristics of .

After determining the cluster model associated with the largest copenetic correlation coefficient value among the calculated copenetic correlation coefficient values as the optimal cluster model, the clustering module 154 generates at least one cluster based on the cluster model. It can be configured. At least one cluster may be created by performing cutting by determining a cutting position in the dendrogram tree of the determined optimal cluster model. Referring to FIG. 3, which is a diagram illustrating a method of performing cutting in the dendrogram tree of FIG. 2, cutting can be performed by selecting one of the two cutting positions 310 and 320. If you perform the cut by selecting the cut position (310), { , }, and { , } Three clusters can be created. On the other hand, if you perform cutting by selecting the cutting position (320), { , , } and { , } Two clusters can be created. It is possible to program the clustering module 154 so that the cutting location can be determined by the user or an expert on the target system 120.

Referring back to FIG. 1 , device 150 may include a dimension reduction module 156 communicatively coupled to clustering module 154 . To improve the efficiency of monitoring the target system 120, the dimension reduction module 156 performs dimension reduction on each of at least one cluster generated by the clustering module 154 to target the target system 120. System 120 may be configured to determine at least one observed variable. Dimensionality reduction is a process of converting a large number of variables into a small number of variables so that the data characteristics can be sufficiently expressed. Through such dimensionality reduction, the number of variables that must be observed can be reduced. The dimensionality reduction module 156 may be configured to linearly transform variables belonging to a given cluster to provide a smaller number of principal components than the number of corresponding variables and to determine the principal component thus provided as at least one observed variable. In one embodiment, the dimensionality reduction module 156 converts variables belonging to a given cluster into reduced features through principal component analysis (PCA), a dimensionality reduction technique. Principal component analysis is a technique that linearly transforms the original variables into principal components that are independent and orthogonal to each other. Here, the sum of the variances of the newly created principal components is equal to the sum of the variances of the original variables.

The dimension reduction module 156 is a principal component matrix composed of the main components in Equation 2 below for principal component analysis. Sort the variance values of the principal components in descending order. may be configured to find, for example, through an optimization method.

here is a matrix composed of variables belonging to the corresponding cluster, is a principal component matrix composed of principal components, Is cast It is an eigenvector that is linearly converted to .

Eigenvector When found, the same number of principal components as the number of variables belonging to the corresponding cluster is generated, but dimensionality reduction can be achieved by selecting only one or some of these principal components. In one embodiment, the dimensionality reduction module 156 prevents the loss of the characteristics of the original variables through 'size of eigenvalue' and 'cumulative percentage of variance explained'. It can be configured to select some of the main components in a way that minimizes them. According to the eigenvalue magnitude, all principal components By selecting only principal components with eigenvalues greater than the average of the eigenvalues, the relative quality of the selected principal components can be increased. Additionally, the percentage of cumulative variance explained is known as an absolute evaluation measure that indicates the extent to which the variance values of selected principal components represent the variance values of all principal components. By selecting a principal component that satisfies both the eigenvalue magnitude and the percentage of cumulative variance explained, it is possible to select a principal component that satisfies both relative and absolute quality.

In an embodiment of selecting a portion of the principal components according to the eigenvalue size, the dimensionality reduction module 156 performs Equation 3 below: It can be configured to select d main components from among the main components. According to Equation 3 below, d main components with eigenvalues greater than the average of the eigenvalues of all main components are selected.

here Is is the eigenvalue of the main component, and p is is the number of main components, Is is the average of the eigenvalues of the main components of represents d selected main components.

In the above embodiment, the dimensionality reduction module 156 calculates the percent explained cumulative variance defined by Equation 4 below: Determine the minimum i such that is greater than or equal to a selected percentage, It may be further configured to determine at least the i-th determined principal components among the principal components as at least one observed variable. In implementing Equation 4 below, the cumulative variance percentage is calculated by adding principal components one by one starting from the first principal component, and the principal components added until the first principal component is met to check whether the calculated cumulative variance percent meets the standard set by the user. Only up to this point can be determined as an observed variable. In one embodiment, the threshold for cumulative variance percentage is 70 percent.

here Is represents the variance of the ith principal component of Is represents the variance of the jth principal component of represents the percentage of cumulative variance explained.

According to the embodiments of the device 150 described above, hierarchical clustering and dimensionality reduction are performed on the number of variables specified by the number of sensors connected to the target system 210, so that the number is less than the number of the variables. It is possible to select the main components of and perform observation of the target system 120 by using the selected main components as observation variables. Accordingly, the target system 120 can perform observation of the target system 120 using a smaller number of observation variables, thereby improving observation efficiency.

FIG. 4 is a flowchart illustrating an embodiment of a method for determining observation variables for a target system.

Referring to FIG. 4, one embodiment of a method for determining an observation variable for a target system begins with collecting measurement values from sensors for the target system (S405). In step S410, variables representing measurement values collected from each of the sensors are defined. In this step, the number of variables corresponding to the number of sensors is defined, and the measured values corresponding to each of the defined variables are stored in association with the variable. In step S415, hierarchical clustering is performed on the variables to create at least one cluster. In this step, multiple cluster models for variables are created by applying multiple normalization methods to variables, multiple distance measures between variables, and multiple link functions to cluster variables. That is, multiple combinations are formed by taking one from multiple normalization methods, multiple distance measures, and multiple link functions and combining them, applying each of the multiple combinations to hierarchically cluster the variables, and providing the copenetic distance between the variables. You can create a cluster model that defines them. In this step, the value of the copenetic correlation coefficient associated with each of the plurality of cluster models is also calculated, and the cluster model associated with the largest copenetic correlation coefficient value among the copenetic correlation coefficient values can be determined as the optimal cluster model. . Equation 1 above can be used to calculate the value of the copenetic correlation coefficient associated with each cluster model. In this step, at least one cluster can be created by determining the cutting position in the cluster tree (dendrogram tree) of the determined optimal cluster model and performing cutting. Depending on the cutting location, different numbers of clusters can be created. In step S420, dimensionality reduction is performed on each of the at least one cluster generated in step S415 to determine at least one observation variable for the target system 120. In this step, the variables belonging to each cluster are linearly transformed to provide a smaller number of principal components than the number of corresponding variables, and the provided principal component can be determined as at least one observed variable. First, a principal component matrix composed of principal components defined by Equation 2 above to provide the same number of principal components as the number of variables belonging to the cluster. An eigenvector that sorts the variance values of the principal components in descending order. can be obtained. Eigenvector After obtaining , you can use Equation 3 and Equation 4 above to select a number of principal components that are smaller than the number of variables belonging to the cluster among all principal components. By adopting the selected principal components as observation variables, the target system 120 can perform system observation using a small number of observation variables.

The embodiments described above may be implemented with hardware components, software components, and/or a combination of hardware components and software components. For example, the devices, methods, and components described in the embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, and a field programmable gate (FPGA). It may be implemented using a general-purpose computer or a special-purpose computer, such as an array, programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and software applications running on the operating system. Additionally, a processing device may access, store, manipulate, process, and generate data in response to the execution of software. For ease of understanding, a single processing device may be described as being used; however, those skilled in the art will understand that a processing device includes multiple processing elements and/or multiple types of processing elements. It can be seen that it may include. For example, a processing device may include multiple processors or one processor and one controller. Additionally, other processing configurations, such as parallel processors, are possible.

Software may include a computer program, code, instructions, or a combination of one or more of these, which may configure a processing unit to operate as desired, or may be processed independently or collectively. You can command the device. Software and/or data may be used on any type of machine, component, physical device, virtual equipment, computer storage medium or device to be interpreted by or to provide instructions or data to a processing device. , or may be permanently or temporarily embodied in a transmitted signal wave. Software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored on a computer-readable recording medium.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. A computer-readable medium may include program instructions, data files, data structures, etc., singly or in combination, and the program instructions recorded on the medium may be specially designed and constructed for the embodiment or may be known and available to those skilled in the art of computer software. It may be possible. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -Includes optical media (magneto-optical media) and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, etc. Examples of program instructions include machine language code, such as that produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc.

The hardware devices described above may be configured to operate as one or multiple software modules to perform the operations of the embodiments, and vice versa.

As described above, although the embodiments have been described with limited drawings, those skilled in the art can apply various technical modifications and variations based on this. For example, the described techniques are performed in a different order than the described method, and/or components of the described system, structure, device, circuit, etc. are combined or combined in a different form than the described method, or other components are used. Alternatively, appropriate results may be achieved even if substituted or substituted by an equivalent.

Therefore, other implementations, other embodiments, and equivalents of the claims also fall within the scope of the claims described below.

120: target system
140: sensor
150: Device for determining observation variables for target system
152: Preprocessing module
154: Clustering module
156: Dimension reduction module

Claims (25)

A method for determining observed variables for a target system, implemented by an observed variable determining device, comprising:
In a preprocessing module of the observation variable determination device, collecting measurement values from sensors for the target system;
In the preprocessing module, defining variables representing measurement values collected from each of the sensors;
In the clustering module of the observed variable determination device, performing hierarchical clustering on the variables to create at least one cluster; and
In the dimension reduction module of the observed variable determination device, performing dimension reduction on each of the at least one cluster to determine at least one observed variable for the target system.
Including,
The generating step is,
A plurality of normalization methods for the variables, a plurality of distance measures between the variables, and a plurality of linkage functions for clustering the variables are applied to obtain a plurality of normalization methods for the variables. generating a cluster model;
calculating a value of a cophenetic correlation coefficient associated with each of the plurality of cluster models; and
Determining the cluster model associated with the largest copper correlation coefficient value among the copper correlation coefficient values as the optimal cluster model.
Method for determining observed variables, including. delete According to paragraph 1,
The step of generating the plurality of cluster models is,
defining a plurality of combinations; and
Applying each of the plurality of combinations to hierarchically cluster the variables and generating a cluster model defining cophenetic distances between the variables.
Including,
Each of the plurality of combinations above is:
Consisting of one of the plurality of normalization methods, one of the plurality of distance measures, and one of the plurality of connection functions,
How to determine observed variables. According to paragraph 3,
The step of calculating the value of the cophenetic correlation coefficient is,
[Equation 1]

A step of calculating the value of the cophenetic correlation coefficient associated with the cluster model using
Including,
The i and j are indices representing variables, and the is a variable calculated using any one normalization method and any one distance measure applied to generate the corresponding cluster model. and variable Indicates the distance between the represents the average of the distances, and is a variable defined by the corresponding cluster model and variable represents the copenetic distance between represents the average of the copenetic distances,
How to determine observed variables. According to paragraph 1,
The generating step is,
Creating the at least one cluster by cutting the cluster tree defined by the determined optimal cluster model at a selected cutting position.
A method for determining observed variables, further comprising: According to paragraph 1,
The plurality of normalization methods are:
Containing at least one of the center method, standard deviation scaling method, Z-score method, and minimum-maximum method,
How to determine observed variables. According to paragraph 1,
The plurality of distance measures are,
Euclidean distance, Mahalanobis distance, Minkowski distance, City block distance, Pearson correlation distance, and Spearman correlation distance ( Containing at least one of Spearman correlation distance),
How to determine observed variables. According to paragraph 1,
The plurality of connection functions are,
At least one of the nearest neighbor linkage function, the farthest neighbor linkage function, the average linkage function, the weighted average linkage function, and the ward linkage function. containing,
How to determine observed variables. According to paragraph 1,
The step of determining the observed variable is,
Linearly transforming the variables belonging to the corresponding cluster to provide a smaller number of principal components than the number of variables, and determining the provided principal component as the at least one observed variable.
Method for determining observed variables, including. According to clause 9,
The step of determining the principal component as the at least one observed variable includes:
[Equation 2]

at Sort the variance values of the main components in descending order. Steps to find
Including,
remind is a matrix composed of variables belonging to the corresponding cluster, and is a principal component matrix composed of principal components, and Is cast An eigenvector that is linearly converted to,
How to determine observed variables. According to clause 10,
The step of determining the principal component as the at least one observed variable includes:
remind Among the main ingredients of
[Equation 3]

Step of selecting d main components according to
It further includes,
remind Is is the eigenvalue of the main component, and p is is the number of main components of Is is the average of the eigenvalues of the main components of represents the d main components selected,
How to determine observed variables. According to clause 11,
The step of determining the principal component as the at least one observed variable includes:
[Equation 4]

Percentage of cumulative variance explained, defined by determining the minimum i such that is greater than or equal to a selected percentage; and
remind Determining the determined minimum i-th main components among the main components as the at least one observed variable.
It further includes,
remind Is represents the variance of the ith principal component of Is represents the variance of the jth principal component of represents the cumulative percentage of variance explained,
How to determine observed variables. A device for determining observation variables for a target system, comprising:
a preprocessing module that collects measurement values from sensors for the target system and defines variables representing measurement values collected from each of the sensors;
a clustering module that generates at least one cluster by performing hierarchical clustering on the variables; and
A dimensionality reduction module configured to determine at least one observed variable for the target system by performing dimensionality reduction on each of the at least one cluster.
Including,
The clustering module is,
Applying a plurality of normalization methods for the variables, a plurality of distance measures between the variables, and a plurality of link functions for clustering the variables to generate a plurality of cluster models for the variables, and generating the plurality of clusters Calculating the value of the copenetic correlation coefficient associated with each of the models, and determining the cluster model associated with the largest copenetic correlation coefficient value among the copenetic correlation coefficient values as the optimal cluster model,
Observed variable determination device. delete According to clause 13,
The clustering module is,
Define a plurality of combinations, apply each of the plurality of combinations to hierarchically cluster the variables, and create a clustering model that defines cophenetic distances between the variables,
Each of the plurality of combinations above is:
Consisting of one of the plurality of normalization methods, one of the plurality of distance measures, and one of the plurality of connection functions,
Observed variable determination device. According to clause 15,
The clustering module is,
[Equation 1]

Calculate the value of the copenetic correlation coefficient associated with the cluster model using
The i and j are indices representing variables, and the is a variable calculated using any one normalization method and any one distance measure applied to generate the corresponding cluster model. and variable Indicates the distance between the represents the average of the distances, and is a variable defined by the corresponding cluster model above and variable represents the copenetic distance between the represents the average of the copenetic distances,
Observed variable determination device. According to clause 13,
The clustering module is,
Generating the at least one cluster by cutting the cluster tree defined by the determined optimal cluster model at a selected cutting position,
Observed variable determination device. According to clause 13,
The plurality of normalization methods are:
Including at least one of a centering method, a standard deviation scaling method, a Z-score method, and a minimum-maximum method,
Observed variable determination device. According to clause 13,
The plurality of distance measures are,
Containing at least one of Euclidean distance, Mahalanobis distance, Minkowski distance, city block distance, Pearson correlation distance, and Spearman correlation distance,
Observed variable determination device. According to clause 13,
The plurality of connection functions are,
Comprising at least one of a shortest link function, a longest link function, an average link function, a weighted average link function, and a Ward link function,
Observed variable determination device. According to clause 13,
The dimension reduction module,
Linearly transforming the variables belonging to the corresponding cluster to provide a smaller number of principal components than the number of variables, and determining the provided principal components as the at least one observed variable,
Observed variable determination device. According to clause 21,
The dimension reduction module,
[Equation 2]

at Sort the variance values of the principal components in descending order. Looking for,
remind is a matrix composed of variables belonging to the corresponding cluster, and is a principal component matrix composed of principal components, and Is cast An eigenvector that is linearly converted to,
Observed variable determination device. According to clause 22,
The dimension reduction module,
remind Among the main ingredients of
[Equation 3]

Select d main components according to
remind Is is the eigenvalue of the main component, and p is is the number of main components of Is is the average of the eigenvalues of the main components of represents the d main components selected,
Observed variable determination device. According to clause 23,
The dimension reduction module,
[Equation 4]

Percentage of cumulative variance explained, defined by Determine the minimum i such that is greater than or equal to a selected percentage, and Among the main components of , determine the determined minimum i-th main components as the at least one observed variable,
remind Is represents the variance of the ith principal component of Is represents the variance of the j-th principal component of represents the cumulative percentage of variance explained,
Observed variable determination device. A computer program stored in a computer-readable recording medium for executing any one of the methods of claims 1, 3 to 12.