Data analysis device, data analysis method, and program
By detecting the change points and correcting the model, the problem of insufficient data analysis accuracy in the prior art is solved, and high-precision prediction of multiple data is achieved.
Patent Information
- Application Number
- CN202380042670.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2022-06-16
- Filing Date
- 2023-04-25
- Publication Date
- 2024-12-20
AI Technical Summary
In the prior art, when dealing with uncertain elements affecting target variables, it is difficult to analyze multiple data accurately, resulting in inaccurate prediction of future value.
Through the data analysis device, multiple data sets are obtained, multiple regression models representing the relationship between variables and target variables are generated, changing points are detected, the model is corrected to improve accuracy, and the corrected model is generated for prediction.
Improves the accuracy of data analysis and enhances the predictive ability of multiple data, especially in the presence of uncertain elements.
Smart Images
Figure CN119173888A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to a data analysis device that analyzes a plurality of data, a data analysis method, and a program for executing the data analysis method. Background Art
[0002] In the past, data analysis devices that analyze multiple data are known. As an example of such a device, the following data analysis device is described in Patent Document 1: a multiple regression analysis is performed on multiple time series data, and the analysis results are used to predict future values. Specifically, in the data analysis device of Patent Document 1, for explanatory variables of performance data assigned with sequential information such as time series, the multiple regression model of the target variable of the performance data assigned with sequential information such as time series is calculated by adding the items obtained by first differentiating and second differentiating the characteristics of data changes based on sequential information such as time series with time and sequence as new explanatory variables, and predicting the target variable in any date, time, and sequence by calculating the multiple regression model of the target variable of the performance data assigned with sequential information such as time series.
[0003] Prior Art Literature
[0004] Patent Literature
[0005] Patent Document 1: Japanese Patent Application Publication No. 2016-031714 Summary of the invention
[0006] A data analysis device involved in one embodiment of the present invention comprises: a data acquisition unit, which acquires a plurality of explanatory variables and a target variable that can take different values according to time; a first model derivation unit, which uses the plurality of explanatory variables and the target variable to perform a multivariate regression analysis and derive a first model representing the relationship between the plurality of explanatory variables and the target variable; a change point detection unit, which detects a change point, wherein the change point is a time when a given explanatory variable among the plurality of explanatory variables changes beyond a predetermined range; and a second model derivation unit, which corrects the first model based on the given explanatory variable with the change point as a starting point, and derives a second model as a corrected model.
[0007] A data analysis method according to one aspect of the present disclosure includes: a step of obtaining a plurality of explanatory variables and a target variable that can take different values according to time; a step of performing a multivariate regression analysis using the plurality of explanatory variables and the target variable, and deriving a first model representing the relationship between the plurality of explanatory variables and the target variable; a step of detecting a change point, wherein the change point is a time when a given explanatory variable among the plurality of explanatory variables changes beyond a predetermined range; and a step of correcting the first model based on the given explanatory variable with the change point as a starting point, and deriving a second model as the corrected model.
[0008] In addition, these general or specific methods can be implemented by systems, methods, integrated circuits, computer programs, or computer-readable recording media such as CD-ROMs, or by any combination of systems, methods, integrated circuits, computer programs, and recording media. BRIEF DESCRIPTION OF THE DRAWINGS
[0009] Figure 1 This is a diagram showing an example of a data analysis system in the embodiment.
[0010] Figure 2 It is a diagram showing the structure of a data analysis device in an embodiment.
[0011] Figure 3 This is a diagram showing an example of a data set in the embodiment.
[0012] Figure 4 This is a diagram showing an example of explanatory variables and target variables selected from a data set in the embodiment.
[0013] Figure 5 This is a block diagram showing the functional structure of the data analysis device involved in the embodiment.
[0014] Figure 6 The diagram shows first explanatory variables, second explanatory variables, and target variables of a plurality of data acquired by the data analysis device according to the embodiment.
[0015] Figure 7 This is a diagram showing a first prediction error which is a difference between a first prediction value calculated based on the first model and a target variable obtained through actual measurement.
[0016] Figure 8 This is a diagram showing a reference prediction error, which is the difference between a reference prediction value calculated based on a reference model that does not include the second explanatory variable and a target variable obtained through actual measurement.
[0017] Fig. 9 This is a diagram showing an example of the change point score of the second explanatory variable.
[0018] Fig.10 This is a diagram showing another example of the change point score of the second explanatory variable.
[0019] Fig.11 This is a diagram showing the second prediction error which is the difference between the second prediction value calculated based on the second model and the target variable obtained through actual measurement.
[0020] Fig.12 This is a flowchart showing a data analysis method according to an embodiment.
[0021] Fig.13 This is a block diagram showing a functional structure of a data analysis device according to a modification of the embodiment. DETAILED DESCRIPTION
[0022] In the analysis device described in Patent Document 1, for example, when there are uncertain factors affecting the target variable in the performance data, it is difficult to analyze a plurality of data with high accuracy. Therefore, it is difficult to predict future values with high accuracy.
[0023] The present disclosure solves the above-mentioned problems, and an object of the present disclosure is to provide a data analysis device and the like that can analyze a plurality of data with high accuracy.
[0024] Hereinafter, the embodiments and the like will be described with reference to the accompanying drawings. The embodiments and the like described below all represent general or specific examples. The numerical values, shapes, materials, structural elements, the configuration positions of the structural elements, the connection methods, steps, the order of the steps, etc. shown in the following embodiments and the like are examples and are not intended to limit the present disclosure. In addition, among the structural elements in the following embodiments and the like, the structural elements that are not recorded in the independent claims are described as arbitrary structural elements.
[0025] In addition, each figure is a schematic diagram and may not be a strict illustration. In addition, in each figure, the same symbols are given to substantially the same structure, and repeated descriptions are sometimes omitted or simplified. In addition, in each figure, even if the same object is illustrated, the scale is sometimes changed for convenience.
[0026] (Implementation Method)
[0027] [Hardware Structure]
[0028] Figure 1 This is a diagram showing an example of a data analysis system in this embodiment.
[0029] The data analysis system 900 in this embodiment includes a data analysis device 1 and a manufacturing management device 500 .
[0030] The manufacturing management device 500 is, for example, installed in a manufacturing plant and is a device for managing a manufacturing system for manufacturing products. The manufacturing management device 500 sends a data set Ds obtained from the manufacturing system to the data analysis device 1 via a network such as the Internet. Figure 3 as well as Figure 4 This will be described later.
[0031] The data analysis device 1 is constituted by a personal computer or the like, and receives the data set Ds from the above-mentioned manufacturing management device 500. The data analysis device 1 in this embodiment generates a plurality of models representing the relationship between the data of the explanatory variables and the data of the target variables based on the data set Ds.
[0032] Figure 2 1 is a diagram showing the configuration of a data analysis device 1 according to the present embodiment.
[0033] The data analysis device 1 includes an input unit 101 , a calculation circuit 102 , a memory 103 , an output unit 104 , a storage unit 105 , a database 106 , and a communication unit 107 .
[0034] The communication unit 107 communicates with a device located outside the data analysis device 1. The communication may be wireless communication or wired communication. The wireless communication method may be Wi-Fi (registered trademark), Bluetooth (registered trademark) or ZigBee, or other methods. For example, the communication unit 107 communicates with the manufacturing management device 500 and receives the data set Ds from the manufacturing management device 500.
[0035] The input unit 101 has a function as an HMI (Human Machine Interface) that receives input operations from a user, and includes, for example, a keyboard, a mouse, a touch sensor, a touch pad, and the like.
[0036] The output unit 104 has a display for displaying images or text, and the display is, for example, a liquid crystal display, a plasma display, an organic EL (Electro-Luminescence) display, etc. In addition, the output unit 104 may have a printer for printing images or text, or may have a function of storing data output from the calculation circuit 102 in the storage unit 105 in the form of a file.
[0037] The storage unit 105 stores a program (i.e., a computer program) 105a that describes various commands to the operation circuit 102. In addition, the storage unit 105 may also store various temporary data 105b that are temporarily generated by the processing of the operation circuit 102. In addition, such a storage unit 105 is a non-volatile recording medium, such as a magnetic storage device such as a hard disk, an optical disk, a semiconductor memory, etc. In addition, the program 105a is provided to the data analysis device 1 via a removable medium or a network, for example, and is stored in the storage unit 105. The removable medium is, for example, a CD-ROM (Compact Disc Read Only Memory), a flash memory, etc. Therefore, the communication unit 107 may also have an interface for reading the program 105a of the removable medium.
[0038] The program 105a read out and expanded by the calculation circuit 102 is temporarily stored in the memory 103. Such a memory 103 is, for example, a volatile RAM (Random Access Memory).
[0039] The arithmetic circuit 102 is a circuit for executing the program 105a expanded in the memory 103, and is, for example, a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit). When executing the program 105a, the arithmetic circuit 102 may also use various temporary data 105b stored in the storage unit 105.
[0040] The database 106 is a non-volatile recording medium similar to the storage unit 105 , such as a magnetic storage device such as a hard disk, an optical disk, a semiconductor memory, etc. For example, the arithmetic circuit 102 obtains a data set Ds from the manufacturing management device 500 via the network and the communication unit 107 , and stores the data set Ds in the database 106 .
[0041] In the present embodiment, the storage unit 105 and the database 106 are different recording media, but the storage unit 105 and the database 106 may be configured as a single recording medium including them.
[0042] [Dataset]
[0043] Figure 3 : is a figure which shows an example of the data set Ds in this embodiment.
[0044] The data set Ds is an original data set sent from the manufacturing management device 500, and is a structured data set consisting of a plurality of manufacturing data, for example, representing the physical properties, process conditions, and quality of the product manufactured by the manufacturing process in the manufacturing system. Figure 3 As shown, such a data set Ds represents the variable names of multiple variables and the data of these variables. In addition, the data can be any data as long as it represents at least one of a character and a number. The variable names of multiple variables are arranged in the first row of the data set Ds, and the data of multiple variables are arranged in each row after the second row of the data set Ds.
[0045] Furthermore, the time when production was performed by the manufacturing management device 500 is shown in the leftmost column of the data set Ds.
[0046] like Figure 3As shown, in the first row of the data set Ds, physical property 1, physical property 2, physical property 3, process condition 1, inspection 1, and inspection 2 are configured as the names of the variables. Physical property 1, physical property 2, physical property 3 are appropriately selected from, for example, viscosity, particle size, solid content ratio, etc. Process condition 1 is appropriately selected from, for example, flow rate, pressure, etc. Inspection 1 and inspection 2 are inspection items for products or semi-finished products when manufactured using physical property 1, physical property 2, physical property 3, and process condition 1. Inspection 1 and inspection 2 are appropriately selected from, for example, coating weight, film thickness, coating area, etc. The rows after the second row of the data set Ds contain data of variables identified by these variable names.
[0047] In this embodiment, Figure 3 The physical property 1, physical property 2, physical property 3 and process condition 1 shown are explanatory variables, and inspection 1 and inspection 2 are target variables. In this example, four explanatory variables are shown, and two target variables are shown. In addition, the process condition is not limited to one, and can be two or more.
[0048] Figure 4 : is a diagram showing an example of explanatory variables and target variables selected from the data set Ds. Figure 4 Shows from Figure 3 The state where physical property 1 and inspection 1 are selected from the data set Ds shown and arranged at each production time. Figure 4 In the figure, the data number corresponding to the time is added to each production time. In this figure, property 1 is selected as the explanatory variable and inspection 1 is selected as the target variable.
[0049] In addition, the method of selecting the explanatory variable and the target variable is not limited to this. For example, it is also possible to select physical property 2 as the explanatory variable and to select inspection 2 as the target variable from the data set Ds. In addition, it is also possible to select physical property 1 and physical property 2 as the explanatory variables and to select inspection 1 as the target variable. It is also possible to select physical property 1, physical property 2, and physical property 3 as the explanatory variables and to select inspection 1 as the target variable. That is, it is also possible to select two or more explanatory variables and one target variable respectively.
[0050] In addition, Figure 4 Although the time every 10 minutes is selected in the data set Ds, the method of selecting the time is not limited to this. For example, 9:00, 9:20, and 9:40 may be selected every 20 minutes from 9:00 to 9:59, 10:00, 10:20, and 10:40 may be selected every 20 minutes from 10:00 to 10:59, and 110:00, 11:20, and 11:40 may be selected every 20 minutes from 11:00 to 11:40.
[0051] The data analysis device of this embodiment performs data analysis on the data set Ds as exemplified above. In order to facilitate understanding of the present invention, the above-mentioned explanatory variables and target variables are further simplified and described below.
[0052] [Structure of data analysis device]
[0053] Reference Figure 5 to Figure 11 The structure of the data analysis device involved in the embodiment is described.
[0054] Figure 5 It is a block diagram showing the functional structure of the data analysis device 1 according to the embodiment.
[0055] like Figure 5 As shown, the data analysis device 1 includes: a data acquisition unit 10 that acquires data; a data editing unit 20 that accumulates data; a first model derivation unit 30 that derives a first model M1 based on the data; and a first prediction value calculation unit 40 that predicts data based on the first model M1. In addition, the data analysis device 1 includes: a change point detection unit 50 that detects a change point of the data; a second model derivation unit 60 that corrects the first model M1 and derives a second model M2; a second prediction value calculation unit 70 that predicts data based on the second model M2; and a model selection unit 80 that selects a model to be used in the future. The functional structure of the data analysis device 1 is realized by executing a program stored in the storage unit 105.
[0056] The data acquisition unit 10 acquires a plurality of data from the outside. For example, the data acquisition unit 10 acquires a plurality of data through operation input by a user using the data analysis device 1 or data input by an external device.
[0057] Figure 6 1 is a diagram showing a first explanatory variable X, a second explanatory variable E, and a target variable Y as a plurality of data acquired by the data analysis device 1 .
[0058] Figure 6 The graphs (a) to (d) show a plurality of data organized in time series. Figure 6 The table (e) shows the data that form the basis of each graph. Figure 6 It is shown in Figure 3 Different datasets.
[0059] The plurality of data are, for example, actual measurement data such as manufacturing conditions and manufacturing performance data. The plurality of data are composed of the first explanatory variable X and the second explanatory variable E, which are the data of the cause and the result, and the target variable Y, which is the data of the result. The first explanatory variable X, the second explanatory variable E, and the target variable Y are respectively expressed as physical quantities of SI basic units such as length, mass, current, temperature, and time. The first explanatory variable X, the second explanatory variable E, and the target variable Y can take different values depending on the time.
[0060] The data editing unit 20 associates the first explanatory variable X, the second explanatory variable E, and the target variable Y acquired by the data acquisition unit 10 with the production time and edits them. For example, the data editing unit 20 sorts the first explanatory variable X, the second explanatory variable E, and the target variable Y in ascending order of time and sets them as time series data. When using manufacturing performance data, the measured data can be sorted based on the manufacturing order, for example, the manufacturing performance data can also be edited in ascending order of manufacturing time.
[0061] In this way, the first explanatory variable X, the second explanatory variable E, and the target variable Y are edited according to time, thereby becoming time series data representing the time change of the physical quantity. The data editing unit 20 stores the edited time series data in the database 106 .
[0062] Here, each variable that will become the time series data is represented as the first explanatory variable X k (t), the second explanatory variable E(t), and the target variable Y(t). t is a data number corresponding to a time, and k is a number indicating a type of data. In addition, t may be simply referred to as "time" hereinafter.
[0063] First, explain the variable X k (t) is a variable that contributes most to the target variable Y(t). k (t) is composed of one or more explanatory variables such as an explanatory variable X1(t) related to the manufacturing process conditions and an explanatory variable X2(t) related to the equipment component state (for example, the number of component shots).
[0064] The second explanatory variable E(t) is, for example, an explanatory variable such as a material property value. The second explanatory variable E(t) is also a variable that contributes more to the target variable Y(t), but it has uncertain elements, and the contribution to the target variable Y(t) sometimes changes. The so-called uncertain elements are, for example, external factors that are difficult to control, such as failure stop, batch change of materials, and measurement environment. Therefore, the second explanatory variable E(t) is more likely to be used than the first explanatory variable X. k(t) The coefficient of variation is large. The coefficient of variation is a value obtained by dividing the standard deviation indicating the degree of dispersion of data by the average value. The data set of the second explanatory variable E(t) may also be composed of a plurality of populations.
[0065] The target variable Y(t) is a numerical value reflecting the quality of the manufactured product, such as an inspection value of an intermediate product or a manufactured product. Alternatively, the target variable Y(t) may be a value indicating a result of determining whether the product is acceptable or not.
[0066] like Figure 6 As shown in FIG. 1 , the first explanatory variable X1(t), X2(t), the second explanatory variable E(t), and the target variable Y(t) change according to the time. In addition, in the figure, these variables and time are represented by simplified numerical values. Information related to these multiple data and time is stored in the database 106 and output to the first model derivation unit 30 and the second model derivation unit 60 described later.
[0067] The first model deriving unit 30 performs a multiple regression analysis using a plurality of explanatory variables X1(t), X2(t), E(t) and a target variable Y(t) to derive a first model M1 representing the relationship between the plurality of explanatory variables X1(t), X2(t), E(t) and the target variable Y(t). The first model M1 is a model for predicting future values of the plurality of explanatory variables X1(t), X2(t), E(t) and the target variable Y(t). The first model M1 is defined by the following (Formula 1).
[0068] [Mathematical formula 1]
[0069]
[0070] In (Equation 1), Y is the target variable, X k is the first explanatory variable, E is the second explanatory variable. K is the number of the first explanatory variable. β0 is the regression constant, β k and β k＋1 is the multiple regression coefficient. Figure 6 The data shown is input into (Formula 1), and a multivariate regression analysis is performed to derive the first model M1. The first model M1 derived by the first model deriving unit 30 is stored in the database 106.
[0071] The first prediction value calculation unit 40 calculates a first prediction value P1(t) corresponding to the target variable Y(t) based on the first model M1. Specifically, the first prediction value calculation unit 40 calculates a first prediction value P1(t) corresponding to the target variable Y(t) by substituting the first explanatory variable X k (t) and the second explanatory variable E(t) are substituted into the first model M1 to calculate the first predicted value P1(t) corresponding to the target variable Y(t).
[0072] Here, in order to verify the accuracy of the first model M1 , the difference between the first predicted value P1 (t) calculated based on the first model M1 and the target variable Y (t) obtained by actual measurement is checked.
[0073] Figure 7 This is a diagram showing a first prediction error which is a difference between a first prediction value P1 (t) calculated based on the first model M1 and a target variable Y (t) obtained by actual measurement.
[0074] Figure 7 The target variable Y(t) obtained by actual measurement is shown in the graph (a). Figure 6 The same as (d). Figure 7 The graph of (b) shows the first predicted value P1 (t) calculated based on the first model M1. Figure 7 The first prediction error is shown in the graph of (c). The first prediction error is a value obtained by subtracting the first prediction value from the target variable, which is an actual measurement value, and is calculated by a formula of Y(t)-P1(t). Figure 7 The table (d) shows the data that form the basis of each graph.
[0075] As shown in the figure, the first prediction error increases or decreases with respect to 0. That is, the first prediction value P1(t) calculated based on the first model M1 does not match the target variable Y(t) which is the actual measurement value. Figure 6 If a model is generated using the data shown as it is, that is, using data containing uncertain elements, the accuracy of the model may decrease.
[0076] Therefore, a case where the second explanatory variable E(t) including an uncertain element is removed from the first model M1 is studied.
[0077] Figure 8 This is a diagram showing a reference prediction error which is a difference between a reference prediction value P1a(t) calculated based on a reference model M1a that does not include the second explanatory variable E(t) and a target variable Y(t) obtained by actual measurement.
[0078] Figure 8 The target variable Y(t) obtained by actual measurement is shown in the graph (a). Figure 6 The same as (d).
[0079] Figure 8 The graph of (b) shows the reference prediction value P1a(t) calculated based on the reference model M1a. The reference model M1a is obtained by removing the third term (β) of the first model M1 from the first model M1. k＋1 The model obtained by E) is represented by the following (Formula 2).
[0080] [Mathematical formula 2]
[0081]
[0082] The reference predicted value P1a(t) is obtained by taking the first explanatory variable X k (t) is substituted into the reference model M1a and calculated.
[0083] Figure 8 The graph of (c) shows the reference prediction error, which is the difference between the reference prediction value P1a(t) and the target variable Y(t). The reference prediction error is a value obtained by subtracting the reference prediction value from the target variable, and is calculated using the formula Y(t)-P1a(t). Figure 8 The table (d) shows the data that form the basis of each graph.
[0084] The reference prediction error shown in the figure is a small value from time 0 to time 5, but increases sharply after time 6. That is, the reference prediction value P1a(t) calculated based on the reference model M1a is roughly consistent with the target variable Y(t) until time 5, but is inconsistent after time 6. The reference prediction value P1a(t) is inconsistent with the target variable Y(t) because the component of the second explanatory variable E(t) that contributes to the target variable Y(t) has undergone a large change from time 5 to time 6, which has a large impact on the target variable Y(t). For this reason, it is believed that by identifying the time when the second explanatory variable E(t) changes significantly and generating a model corresponding to the change, the accuracy of the model can be improved. Therefore, the data analysis device 1 involved in the present embodiment has a change point detection unit 50 that detects the change point of the data.
[0085] The change point detection unit 50 detects a change point Tc, which is a time when a given explanatory variable among the plurality of explanatory variables X1, X2, and E changes beyond a predetermined range. In this embodiment, the case where the explanatory variable to be detected as the change point Tc is the second explanatory variable E(t) is described.
[0086] The change point detection unit 50 uses the change point score Sc representing the degree of change of the second explanatory variable E(t) to determine whether the second explanatory variable E(t) exceeds a predetermined range. The change point score Sc is calculated using a change point detection algorithm. An example of a change point detection algorithm is "ChangeFinder (registered trademark)". In addition, the change point Tc can also be derived by the k-nearest neighbor method, AR (autoregressive) model, ARMA (autoregressive moving average) model or RuLSIF (Relative unconstrained Least-Squares Importance Fitting).
[0087] Fig. 9 FIG. 1 is a diagram showing an example of the change point score Sc of the second explanatory variable E. FIG.
[0088] The horizontal axis of the figure represents time, and the vertical axis represents the data of the second explanatory variable E and the change point score Sc. The change point detection unit 50 calculates the change point score Sc of the second explanatory variable E based on the time series data of the second explanatory variable E. For example, when the change point score Sc at the second moment t2, which is the next moment of the first moment t1, is more than twice the change point score Sc at the first moment t1, the change point detection unit 50 determines that the second explanatory variable E exceeds a predetermined range. In the figure, it is determined that there is a change point Tc at the moment 3 indicated by the dotted line. In addition, the second explanatory variable E can also be used to judge the change point Tc, but in order to be less susceptible to sudden interference, noise, etc., it is desirable to use the change point score Sc to judge the change point Tc.
[0089] Fig.10 FIG. 1 is a diagram showing another example of the change point score Sc of the second explanatory variable E. FIG.
[0090] The horizontal axis of the figure represents time, and the vertical axis represents the data of the second explanatory variable E and the change point score Sc. The change point detection unit 50 calculates the change point score Sc of the second explanatory variable E based on the time series data of the second explanatory variable E. For example, when the change point score Sc at the second time t2 is greater than the change point score Sc at the first time t1 by 20 or more, the change point detection unit 50 determines that the second explanatory variable E exceeds the predetermined range. In the figure, it is determined that the change point Tc exists at the time 5 indicated by the dotted line.
[0091] In the present embodiment, the first model M1 is corrected using a plurality of data starting from the change point Tc detected by the change point detection unit 50. An example of a method of correcting the first model M1 will be described below.
[0092] The second model deriving unit 60 corrects the first model M1 based on the second explanatory variable E(t) starting from the change point Tc, and derives the second model M2 as the corrected model. The second model M2 is also a model for predicting future values of a plurality of explanatory variables X1(t), X2(t), E(t) and the target variable Y(t).
[0093] For example, the second model derivation unit 60 corrects the first model M1 by comparing the second explanatory variable E(t) at the change point Tc with the second explanatory variable E(t) at a time before the change point Tc. Specifically, when there is a change point Tc at time t, the second model derivation unit 60 calculates the moving average of the second explanatory variable E at time (t-1), time (t-2), ∙∙∙, and time (tn), and calculates the difference ΔE between the moving average and the second explanatory variable E(t) at time t. The difference ΔE is represented by the following (Formula 3). In addition, n is an integer greater than 1.
[0094] [Mathematical formula 3]
[0095]
[0096] The second model derivation unit 60 multiplies the difference ΔE of (Formula 3) by the multivariate regression coefficient β of (Formula 1) k＋1 , calculate the change ΔY E . Change ΔY E It is expressed by the following (Formula 4).
[0097] [Formula 4]
[0098]
[0099] Then, the second model derivation unit 60 replaces the third term of the first model M1 shown in (Equation 1) with the change amount ΔY E , to correct the first model M1. The second model M2, which is a corrected model of the first model M1, is defined by the following (Formula 5).
[0100] [Formula 5]
[0101]
[0102] Thus, the second model deriving unit 60 obtains the difference between the second explanatory variable E at the change point Tc and the moving average of the second explanatory variable E at a time before the change point Tc, and corrects the first model M1 based on the difference. The second model M2 derived by the second model deriving unit 60 is stored in the database 106.
[0103] The second prediction value calculation unit 70 calculates the second prediction value P2(t) corresponding to the target variable Y(t) based on the second model M2. Specifically, the second prediction value calculation unit 70 calculates the second prediction value P2(t) corresponding to the target variable Y(t) by substituting the first explanatory variable X k (t) and the second explanatory variable E(t) are substituted into the second model M2 to calculate the second predicted value P2(t) corresponding to the target variable Y(t).
[0104] Fig.11This is a diagram showing a second prediction error which is a difference between a second prediction value P2 (t) calculated based on the second model M2 and a target variable Y (t) obtained by actual measurement.
[0105] Fig.11 The target variable Y(t) obtained by actual measurement is shown in the graph (a). Figure 6 The same as (d). Fig.11 The graph of (b) shows the second predicted value P2 (t) calculated based on the second model M2. Fig.11 The second prediction error is shown in the graph of (c). The second prediction error is a value obtained by subtracting the second prediction value from the target variable, which is an actual measurement value, and is calculated by a formula of Y(t)-P2(t). Fig.11 The table (d) shows the data that form the basis of each graph.
[0106] As shown in the figure, the second prediction error is roughly consistent with the reference 0. That is, the second prediction value P2(t) calculated based on the second model M2 is roughly consistent with the target variable Y(t) which is the actual measured value. In this way, by using the second model M2 that corrects the first model M1, the prediction accuracy of the data is improved.
[0107] The model selection unit 80 selects a model to be used in the future from the first model M1 and the second model M2 based on the first prediction value P1 and the second prediction value P2. Specifically, the model selection unit 80 compares the first prediction error and the second prediction error, and selects the model with the smaller error as the model to be used in the future. The so-called future refers to a time later than when the second model M2 is derived. Figure 7 as well as Fig.11 In the example shown, the second model M2 is selected as the model to be used in the future. Alternatively, when the first prediction value P1 has a smaller error than the second prediction value P2, the first model M1 may be selected. The model selected by the model selection unit 80 is stored in the database 106.
[0108] Output unit 104 (see Figure 2 ) is a display device such as a liquid crystal panel, which displays the explanatory variables X, E or the target variable Y predicted based on the first model M1 or the second model M2. In addition, the output unit 104 can also display the first model M1 and the second model M2.
[0109] Thus, the data analysis device 1 comprises: a data acquisition unit 10, which acquires a plurality of explanatory variables X, E and a target variable Y that can take different values according to time; a first model derivation unit 30, which uses the plurality of explanatory variables X, E and the target variable Y to perform a multivariate regression analysis, and derives a first model M1 representing the relationship between the plurality of explanatory variables X, E and the target variable Y; a change point detection unit 50, which detects a change point Tc, which is a time when a given explanatory variable (for example, E) among the plurality of explanatory variables X, E changes beyond a predetermined range; and a second model derivation unit 60, which corrects the first model M1 based on a given explanatory variable E starting from the change point Tc, and derives a second model M2 as a corrected model.
[0110] In this way, by detecting the change point Tc of the given explanatory variable E and correcting the first model M1 based on the given explanatory variable E starting from the change point Tc, the accuracy of the second model M2 as the corrected model can be improved. Thus, a data analysis device 1 capable of analyzing a plurality of data with good accuracy can be provided.
[0111] [Data analysis method]
[0112] Reference Fig.12 An example of a data analysis method according to an embodiment of the present invention will be described.
[0113] Fig.12 This is a flowchart showing a data analysis method according to an embodiment.
[0114] First, the data acquisition unit 10 of the data analysis device 1 acquires Figure 6 The plurality of data are, for example, manufacturing performance data, and are composed of a first explanatory variable X, a second explanatory variable E, and a target variable Y.
[0115] Next, the data editing unit 20 sorts the plurality of data in time series (step S21). Specifically, the data editing unit 20 sorts the plurality of data in ascending order of manufacturing time. For example, the plurality of data sorted in time series is sequentially assigned data numbers corresponding to the time. The data edited by the data editing unit 20 is stored in the database 106.
[0116] In addition, when the data organized in time series is initially input to the data acquisition unit 10, step S21 may be omitted. In addition, when the first explanatory variable X, the second explanatory variable E, and the target variable Y are not pre-set in the plurality of data input to the data acquisition unit 10, the first explanatory variable X, the second explanatory variable E, and the target variable Y may be set by the data acquisition unit 10 or the data editing unit 20.
[0117] Next, the first model deriving unit 30 performs a multiple regression analysis using the first explanatory variable X, the second explanatory variable E, and the target variable Y, and derives a first model M1 representing the relationship among the first explanatory variable X, the second explanatory variable E, and the target variable Y (step S31 ).
[0118] The first predicted value calculation unit 40 calculates the first predicted value P1 corresponding to the target variable Y based on the first model M1 (step S41 ). Step S41 may be executed between steps S61 and S71 described later, or between steps S71 and S81 .
[0119] Next, the change point detection unit 50 detects a change point Tc, which is a time when the second explanatory variable E changes beyond a predetermined range (step S51). The change point detection unit 50 uses a change point score Sc indicating the degree of change of the second explanatory variable E(t) to determine whether the second explanatory variable E(t) exceeds the predetermined range. If the change point Tc is not detected, the first model M1 is used as a model for predicting future data. If the change point Tc is detected, the process proceeds to the next step of deriving the second model M2.
[0120] The second model deriving unit 60 corrects the first model M1 based on the second explanatory variable E(t) starting from the change point Tc, and derives the second model M2 as the corrected model (step S61). For example, the second model deriving unit 60 corrects the first model M1 by comparing the second explanatory variable E(t) at the change point Tc with the second explanatory variable E(t) at a time before the change point Tc.
[0121] The second predicted value calculation unit 70 calculates a second predicted value P2 corresponding to the target variable Y based on the second model M2 (step S71 ).
[0122] Next, the model selection unit 80 selects a model to be used from the first model M1 and the second model M2 based on the first prediction value P1 and the second prediction value P2 (step S81). Specifically, the model selection unit 80 compares the first prediction error and the second prediction error, and selects the model with the smaller error as the model to be used.
[0123] The data analysis device 1 performs future data analysis based on the model selected by the model selection unit 80. Through these steps S11 to S81, a plurality of data can be analyzed with high accuracy.
[0124] [Modification of the embodiment]
[0125] Reference Fig.13The configuration of a data analysis device 1A according to a modification of the embodiment will be described. In the modification, an example in which the second model M2 derived by the second model deriving unit 60 is used as a model to be used thereafter will be described.
[0126] Fig.13 It is a block diagram showing a functional structure of a data analysis device 1A according to a modification of the embodiment.
[0127] like Fig.13 As shown, the data analysis device 1A includes a data acquisition unit 10, a data editing unit 20, a first model derivation unit 30, a change point detection unit 50, a second model derivation unit 60, and a model selection unit 80A. The data acquisition unit 10, the data editing unit 20, the first model derivation unit 30, the change point detection unit 50, and the second model derivation unit 60 have the same structure as the embodiment.
[0128] The model selection unit 80A according to the modification selects the second model M2 as a model to be used from now on, between the first model M1 and the second model M2.
[0129] The data analysis device 1A of the modified example also detects the change point Tc of a given explanatory variable (e.g., E) and corrects the first model M1 based on the given explanatory variable E starting from the change point Tc, thereby improving the accuracy of the second model as the corrected model. Thus, a data analysis device 1A capable of analyzing a plurality of data with good accuracy can be provided.
[0130] (Summarize)
[0131] The data analysis device 1 involved in this embodiment includes: a data acquisition unit 10, which acquires a plurality of explanatory variables X, E and a target variable Y that can take different values according to time; a first model derivation unit 30, which uses the plurality of explanatory variables X, E and the target variable Y to perform a multivariate regression analysis, and derives a first model M1 that represents the relationship between the plurality of explanatory variables X, E and the target variable Y; a change point detection unit 50, which detects a change point Tc, which is a time when a given explanatory variable (for example, E) among the plurality of explanatory variables X, E changes beyond a predetermined range; and a second model derivation unit 60, which corrects the first model M1 based on the given explanatory variable E with the change point Tc as the starting point, and derives a second model M2 as the corrected model.
[0132] In this way, by detecting the change point Tc of the given explanatory variable E and correcting the first model M1 based on the given explanatory variable E starting from the change point Tc, the accuracy of the second model M2 as the corrected model can be improved. Thus, a data analysis device 1 capable of analyzing a plurality of data with good accuracy can be provided.
[0133] Furthermore, the change point detection unit 50 may use the change point score Sc indicating the degree of change of the given explanatory variable E to determine whether the given explanatory variable E exceeds a predetermined range.
[0134] In this way, by using the change point score Sc to determine whether a given explanatory variable E exceeds a predetermined range, the change point Tc can be accurately detected. Therefore, the first model M1 can be corrected based on the given explanatory variable E starting from the change point Tc, and the accuracy of the second model M2 can be improved. Thus, a data analysis device 1 capable of analyzing a plurality of data with good accuracy can be provided.
[0135] In addition, the change point detection unit 50 may determine that the given explanatory variable E exceeds a predetermined range in at least one of the following cases: the change point score Sc at the second moment which is the next moment after the first moment is more than twice the change point score Sc at the first moment, and more than 20.
[0136] Thus, the change point Tc can be accurately detected. Therefore, the first model M1 can be corrected based on a given explanatory variable E starting from the change point Tc, and the accuracy of the second model M2 can be improved. Thus, a data analysis device 1 capable of analyzing a plurality of data with good accuracy can be provided.
[0137] Alternatively, the change point score Sc may be calculated using a change point detection algorithm.
[0138] Thus, the change point Tc can be accurately detected. Therefore, the first model M1 can be corrected based on a given explanatory variable E starting from the change point Tc, and the accuracy of the second model M2 can be improved. Thus, a data analysis device 1 capable of analyzing a plurality of data with good accuracy can be provided.
[0139] Furthermore, the second model deriving unit 60 may correct the first model M1 by comparing the given explanatory variable E at the change point Tc with the given explanatory variable E at a time before the change point Tc.
[0140] This makes it possible to appropriately correct the first model M1 and improve the accuracy of the second model M2. Thus, it is possible to provide a data analysis device 1 that can analyze a plurality of data with high accuracy.
[0141] Alternatively, the second model deriving unit 60 may obtain a difference between a given explanatory variable E at the change point Tc and a moving average of the given explanatory variable E at a time before the change point Tc, and correct the first model M1 based on the difference.
[0142] This makes it possible to appropriately correct the first model M1 and improve the accuracy of the second model M2. Thus, it is possible to provide a data analysis device 1 that can analyze a plurality of data with high accuracy.
[0143] Alternatively, the plurality of explanatory variables may include a first explanatory variable X and a second explanatory variable E, and the given explanatory variable may be the second explanatory variable E, and the coefficient of variation may be greater than that of the first explanatory variable X.
[0144] According to this, even when a plurality of data has a second explanatory variable E with a large coefficient of variation, a model can be appropriately generated. Thus, it is possible to provide a data analysis device 1 capable of analyzing a plurality of data with high accuracy.
[0145] In addition, the data analysis device 1 may further include a data editing unit 20 for editing the plurality of explanatory variables X and E and the target variable Y acquired by the data acquisition unit 10 by associating them with time.
[0146] In this way, by editing by associating multiple explanatory variables X, E and target variable Y with time, the change point Tc can be accurately detected. Therefore, the first model M1 can be corrected based on the given explanatory variable E starting from the change point Tc, and the accuracy of the second model M2 can be improved. Thus, a data analysis device 1 capable of analyzing multiple data with good accuracy can be provided.
[0147] In addition, the data analysis device 1 may also include: a first prediction value calculation unit 40, which calculates the first prediction value P1 corresponding to the target variable Y based on the first model M1; a second prediction value calculation unit 70, which calculates the second prediction value P2 corresponding to the target variable Y based on the second model M2; and a model selection unit 80, which selects a model to be used in the future from the first model M1 and the second model M2 based on the first prediction value P1 and the second prediction value P2.
[0148] This makes it possible to appropriately select a model to be used in the future. Thus, it is possible to provide a data analysis device 1 that can analyze a plurality of data with high accuracy.
[0149] Furthermore, the data analysis device 1A may further include a model selection unit 80 that selects the second model M2 as a model to be used from among the first model M1 and the second model M2.
[0150] This makes it possible to easily and appropriately select a model to be used in the future. Thus, it is possible to provide a data analysis device 1A that can analyze a plurality of data with high accuracy.
[0151] In addition, the data analysis method involved in the present embodiment includes: a step of obtaining a plurality of explanatory variables X, E and a target variable Y that can take different values according to time; a step of performing a multivariate regression analysis using the plurality of explanatory variables X, E and the target variable Y, and deriving a first model M1 representing the relationship between the plurality of explanatory variables X, E and the target variable Y; a step of detecting a change point Tc, which is a moment when a given explanatory variable (for example, E) among the plurality of explanatory variables X, E changes beyond a predetermined range; and a step of correcting the first model M1 based on the given explanatory variable E with the change point Tc as a starting point, and deriving a second model M2 as the corrected model.
[0152] Thus, by detecting the change point Tc of the given explanatory variable E and correcting the first model M1 based on the given explanatory variable E starting from the change point Tc, the accuracy of the second model M2 as the corrected model can be improved. Thus, a plurality of data can be analyzed with high accuracy.
[0153] The program according to the present embodiment is a program for causing a computer to execute the above-mentioned data analysis method.
[0154] By executing this program, a plurality of data can be analyzed with high accuracy.
[0155] (Other Implementations)
[0156] The data analysis device and the like involved in the present disclosure are described above based on various embodiments, but the present disclosure is not limited to these embodiments. As long as it does not deviate from the main purpose of the present disclosure, various modifications that can be thought of by those skilled in the art are applied to the schemes of the various embodiments, and other methods constructed by combining some of the structural elements in the various embodiments are also included in the scope of the present disclosure.
[0157] For example, in the embodiment, an example is shown in which a plurality of data is constituted by two types of first explanatory variables X, one type of second explanatory variable E, and one type of target variable Y. However, the first explanatory variable X may be one type or three or more types.
[0158] In addition, in the embodiment, an example is shown in which a plurality of data includes one target variable Y. However, if a plurality of data has two or more target variables, for example, data analysis based on the present embodiment may be performed on each of the two or more target variables.
[0159] In the embodiment, data with data numbers corresponding to the time are sequentially arranged, but data with data numbers may be data arranged at equal time intervals or data arranged at different time intervals. Data with data numbers may also be interval-removed data that is not sequentially arranged but spaced apart from data that is finely arranged in time series.
[0160] For example, specifically, the above-mentioned data analysis device can also be composed of a computer system, which is composed of a microprocessor, ROM, RAM, hard disk drive, display unit, keyboard, mouse, etc. A data analysis program is stored in the RAM or hard disk drive. The microprocessor acts according to the data analysis program, so that the data analysis device realizes its function. Here, the data analysis program is a program composed of a plurality of command codes representing instructions to the computer in order to realize a given function.
[0161] Furthermore, part or all of the structural elements constituting the above-mentioned data analysis device may also be constituted by a system LSI (Large Scale Integration). System LSI is a super-multifunctional LSI manufactured by integrating multiple structural parts on a single chip, and specifically, is a computer system composed of a microprocessor, ROM, RAM, etc. Computer programs are stored in RAM. The microprocessor operates according to the computer program, so that the system LSI realizes its functions.
[0162] Furthermore, in addition, part or all of the structural elements constituting the above-mentioned data analysis device may also be constituted by an IC card or a single module that can be detached from a computer. The IC card or module is a computer system composed of a microprocessor, ROM, and RAM. The IC card or module may also include the above-mentioned super-multifunctional LSI. The IC card or module realizes its function by operating according to a computer program through a microprocessor. The IC card or module may also be tamper-proof.
[0163] In addition, the present disclosure may also be a data analysis method executed by the above-mentioned data analysis device. In addition, the data analysis method may be implemented by a computer executing a data analysis program, or may be implemented by a digital signal composed of the data analysis program.
[0164] Furthermore, the present disclosure may also be constituted by a non-temporary recording medium that can read the data analysis program or the above-mentioned digital signal. Examples of the recording medium include a floppy disk, a hard disk, a CD-ROM, an MO, a DVD, a DVD-ROM, a DVD-RAM, a BD (Blu-ray (registered trademark) Disc), a semiconductor memory, and the like. In addition, the data analysis program may also be constituted by the above-mentioned digital signal recorded in a non-temporary recording medium.
[0165] In addition, the present disclosure may be configured by transmitting the data analysis program or the digital signal via an electric communication line, a wireless or wired communication line, a network represented by the Internet, or data broadcasting.
[0166] Furthermore, the present disclosure may be a computer system including a microprocessor and a memory, wherein the memory stores a data analysis program and the microprocessor operates according to the data analysis program.
[0167] Furthermore, the data analysis program or the digital signal may be recorded in the non-temporary recording medium and transmitted, or the data analysis program or the digital signal may be transmitted via the network or the like, so that the program may be implemented by another independent computer system.
[0168] Furthermore, the data analysis system may be composed of a server and a terminal held by a user connected to the server via a network.
[0169] According to the data analysis device and the like of the present disclosure, it is possible to analyze a plurality of data with high accuracy.
[0170] Industrial Applicability
[0171] The data analysis device disclosed in the present invention can be applied to data analysis for predicting target variables with high accuracy. In addition, since the conditions for satisfying the target value of the target variable can be predicted and calculated with high accuracy, it can be applied to data analysis for calculating and indicating the optimal manufacturing conditions by utilizing data, for example. In addition, it can be applied to assisting manufacturing operations, for example.
[0172] -Explanation of symbols-
[0173] 1.1A Data analysis device
[0174] 10 Data Acquisition Department
[0175] 20 Data Editing Department
[0176] 30 Model 1 export section
[0177] 40 1st prediction value calculation unit
[0178] 50 Change point detection unit
[0179] 60 Second model export section
[0180] 70 Second predicted value calculation unit
[0181] 80, 80A Model Selection Department
[0182] 101 Input section
[0183] 102 Operational Circuit
[0184] 103 Memory
[0185] 104 Output
[0186] 105 Storage
[0187] 105a Procedure
[0188] 105b Temporary Data
[0189] 106 Database
[0190] 107 Ministry of Communications
[0191] 500 Manufacturing Management Device
[0192] 900 Data Analysis System
[0193] Ds dataset
[0194] E The second explanatory variable (given explanatory variable)
[0195] M1 1st Model
[0196] M2 Model 2
[0197] P1 1st predicted value
[0198] P2 2nd predicted value
[0199] Sc Change point score
[0200] Tc change point
[0201] X 1st explanatory variable
[0202] Y Target variable.
Claims
1. A data analysis device comprising: A data acquisition unit that acquires a plurality of explanatory variables and a target variable that can take different values according to time; a first model deriving unit that performs a multivariate regression analysis using the plurality of explanatory variables and the target variable, and derives a first model that represents a relationship between the plurality of explanatory variables and the target variable; a change point detection unit that detects a change point, the change point being a time when a given explanatory variable among the plurality of explanatory variables changes beyond a predetermined range; and The second model deriving unit corrects the first model based on the given explanatory variables starting from the change point, and derives a second model as the corrected model.
2. The data analysis device according to claim 1, wherein: The change point detection unit determines whether the given explanatory variable exceeds a predetermined range by using a change point score indicating a degree of change of the given explanatory variable.
3. The data analysis device according to claim 2, wherein: In at least one of the cases where the change point score at the second moment, which is the next moment after the first moment, is more than twice the change point score at the first moment and is greater than 20, the change point detection unit determines that the given explanatory variable exceeds a predetermined range.
4. The data analysis device according to claim 3, wherein: The change point scores are calculated using a change point detection algorithm.
5. The data analysis device according to claim 1, wherein: The second model derivation unit corrects the first model by comparing the given explanatory variable at the change point with the given explanatory variable at a time before the change point.
6. The data analysis device according to claim 5, wherein: The second model derivation unit obtains a difference between the predetermined explanatory variable at the change point and a moving average of the predetermined explanatory variable at a time before the change point, and corrects the first model based on the difference.
7. The data analysis device according to claim 1, wherein: The plurality of explanatory variables include a first explanatory variable and a second explanatory variable, The given explanatory variable is the second explanatory variable, and has a larger coefficient of variation than the first explanatory variable.
8. The data analysis device according to claim 1, wherein: The data analysis device further includes a data editing unit configured to edit the plurality of explanatory variables and the target variable acquired by the data acquisition unit by associating them with time.
9. The data analysis device according to any one of claims 1 to 8, wherein: The data analysis device further comprises: a first predicted value calculation unit that calculates a first predicted value corresponding to the target variable based on the first model; a second predicted value calculation unit that calculates a second predicted value corresponding to the target variable based on the second model; and The model selection unit selects a model to be used subsequently from among the first model and the second model based on the first predicted value and the second predicted value.
10. The data analysis device according to any one of claims 1 to 8, wherein: The data analysis device further includes a model selection unit that selects the second model as a model to be used in the future, between the first model and the second model.
11. A data analysis method, comprising: The step of obtaining a plurality of explanatory variables and a target variable that can take different values according to time; A step of performing a multivariate regression analysis using the plurality of explanatory variables and the target variable to derive a first model representing a relationship between the plurality of explanatory variables and the target variable; a step of detecting a change point, wherein the change point is a time when a given explanatory variable among the plurality of explanatory variables changes beyond a predetermined range; and A step of correcting the first model based on the given explanatory variables starting from the change point, and deriving a second model as the corrected model.
12. A program for causing a computer to execute the data analysis method according to claim 11.