Cause analysis device, method and program

Abstract

To specify a cause of a change to an abnormal state when the change occurs alone.SOLUTION: A cause analysis device comprises an acquisition unit; and a calculation unit comprising a deviation score calculation unit, a relation weight calculation unit, a property weight calculation unit, and a cause score calculation unit. The acquisition unit acquires a reference data group including object variables and explanatory variables, and object data including object variables and explanatory variables. The deviation score calculation unit calculates deviation scores showing deviation degrees of the explanatory variables of the object data to distribution of the explanatory variables of the reference data group. The relation weight calculation unit calculates a relation weight representing strength of a relation between the object variables and the explanatory variables of the reference data group. The property weight calculation unit calculates a property weight representing coincidence degrees of values of the object variables and the explanatory variables of the object data to distribution of the object variables and the explanatory variables of the reference data group. The cause score calculation unit calculates cause scores of the explanatory variables to variation of the object variables of the object data on the basis of the deviation scores, the relation weight and the property weight.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] An embodiment of the present invention relates to a cause analysis device, a cause analysis method, and a program. [Background technology]

[0002] In a logistics warehouse, when the work efficiency, work time, etc. change to a specific state, it is important to identify the cause of the change early. For example, when an abnormality occurs in which the work efficiency changes from a normal state to a decreased state, it is possible to maintain and improve the work efficiency by identifying the cause of the decrease early. The cause can be identified by acquiring various data on warehouse work and closely examining information on the entire warehouse and information on each individual task. For example, when the work efficiency decreases during a certain period, data showing the total amount of goods handled during that period is acquired, and information showing the individual work hours of workers is closely examined. As a result, it is possible to identify the cause of the change as an increase in the total amount of goods handled and the associated extension of work hours.

[0003] In order to identify such causes, in manufacturing plants where mechanization has progressed and data acquisition has become easier, devices or systems that acquire and examine data are being developed. For example, an analysis device that performs analysis to identify the cause of changes to abnormalities in manufacturing plants is known. This analysis device acquires first data on the state corresponding to the objective variable and second data on the manufacturing conditions corresponding to the explanatory variables for a group of products manufactured in a day in the manufacturing plant. The analysis device then obtains index values ​​for the first data and the second data, and changes the display mode of information on the first data and the second data based on the respective index values. It is assumed that multiple abnormalities will occur under the same conditions as a result of repeating a certain task in a manufacturing plant. For this reason, the analysis device focuses on the bias in the number of products determined to be abnormal for the index value of the second data, and identifies the cause of the change in the first data by calculating a p-value within the framework of a statistical test.

[0004] However, unlike manufacturing plants where certain operations are repeated, logistics warehouses have variations in the steady state, and it is expected that anomalies will occur independently for each condition. For example, in logistics warehouses, the types of products and the work of workers are not constant, and there are many factors that warehouse operators cannot control, such as the number of orders from customers and the time of order receipt. For this reason, it is rare for multiple anomalies to occur under the same conditions in logistics warehouses, and anomalies occur independently. Therefore, in logistics warehouses, when a change to an abnormal state occurs independently, it is difficult to identify the cause of the change even if an analysis device that calculates a p-value from the bias of anomalies when multiple anomalies occur is applied. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Patent Publication No. 2021-71896 Summary of the Invention [Problem to be solved by the invention]

[0006] An object of the present invention is to provide a cause analysis device, method, and program that are capable of identifying the cause of a change to an abnormal state when the change occurs alone. [Means for solving the problem]

[0007] A cause analysis device according to an embodiment includes an acquisition unit, a deviation score calculation unit, a relationship weight calculation unit, a property weight calculation unit, and a cause score calculation unit. The acquisition unit acquires a reference data group including an objective variable and an explanatory variable, and target data including the objective variable and the explanatory variable. The deviation score calculation unit calculates a deviation score indicating a deviation degree of the explanatory variable of the target data with respect to a first distribution regarding the explanatory variable of the reference data group. The relationship weight calculation unit calculates a relationship weight indicating a strength of a relationship between the objective variable and the explanatory variable of the reference data group. The property weight calculation unit calculates a property weight indicating a degree of agreement between the objective variable and the explanatory variable of the target data with respect to a second distribution regarding the objective variable and the explanatory variable of the reference data group. The cause score calculation unit calculates a cause score of the explanatory variable with respect to a change in the objective variable of the target data based on the deviation score, the relationship weight, and the property weight. [Brief description of the drawings]

[0008] [Figure 1] FIG. 1 is a block diagram showing an example of a cause analysis device according to a first embodiment. [Diagram 2] 5 is a flowchart for explaining an operation in the first embodiment. [Diagram 3] FIG. 11 is a block diagram showing an example of a cause analysis device according to a second embodiment. [Figure 4] 10 is a flowchart for explaining an operation according to the second embodiment. [Diagram 5] FIG. 11 is a diagram showing an example of tabular data according to the second embodiment. [Figure 6] FIG. 11 is a diagram showing an example of graph data according to the second embodiment. [Figure 7] FIG. 11 is a diagram showing an example of tabular data according to the second embodiment. [Figure 8] FIG. 11 is a diagram showing an example of visualization data according to the second embodiment. [Figure 9] FIG. 13 is a block diagram showing an example of a cause analysis device according to a third embodiment. [Figure 10] 10 is a flowchart for explaining an operation according to a third embodiment. [Figure 11] FIG. 13 is a diagram showing an example of visualization data according to the third embodiment. [Figure 12] FIG. 11 is a schematic diagram for explaining the operation according to the third embodiment. [Figure 13] FIG. 13 is a diagram showing another example of visualization data according to the third embodiment. [Figure 14] FIG. 13 is a diagram showing another example of visualization data according to the third embodiment. [Figure 15] FIG. 13 is a schematic diagram for explaining the operation in a modified example of the third embodiment. [Figure 16] FIG. 13 is a diagram showing an example of visualization data according to another modified example of the third embodiment. [Figure 17] FIG. 13 is a diagram showing an example of a hardware configuration of a cause analysis apparatus according to a fourth embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] Hereinafter, each embodiment will be described with reference to the drawings. In the following description, a cause analysis device that analyzes the cause of a change in a target variable using an explanatory variable will be described as an example. Note that the cause analysis device may be replaced with any name such as a data analysis device or a cause score calculation device.

[0010] (First embodiment) 1 is a block diagram showing an example of a cause analysis device and its peripheral configuration according to the first embodiment. The cause analysis device 1 includes an acquisition unit 10 and a calculation unit 20. The calculation unit 20 includes a deviation score calculation unit 21, a relationship weight calculation unit 22, a property weight calculation unit 23, and a cause score calculation unit 24. The calculation unit 20 is also connected to a display unit 40. Note that the display unit 40 is not essential and may be omitted.

[0011] Here, the acquiring unit 10 acquires a reference data group including a response variable and an explanatory variable, and target data including the response variable and the explanatory variable, and inputs the reference data group and the target data to the calculation unit 20. For example, the acquiring unit 10 may acquire the reference data group and the target data from a database device (not shown) that stores warehouse information in a logistics warehouse via a network. The acquiring unit 10 may be called an input unit or any other arbitrary name.

[0012] The reference data group is a data group used as a standard when performing causal analysis, and includes two or more reference data including one or more objective variables Yi and two or more explanatory variables Xj. To elaborate, the reference data is a set (Yi, X1, ..., Xj) of one or more objective variables Yi and two or more explanatory variables Xj.

[0013] The target data is data that is the subject of causal analysis, and includes one or more objective variables Yi and two or more explanatory variables Xj. Note that the target data may be interpreted as any name such as calculation target data or cause score calculation target data.

[0014] The objective variables {Yi: i=1...N, N is an integer equal to or greater than 1} represent data on the state of a warehouse that has changed due to the occurrence of an abnormality in the warehouse information of the logistics warehouse. Here, N represents the number of items in the objective variables obtained.

[0015] As the objective variable Yi, for example, the work efficiency of the warehouse acquired in the unit for which the cause analysis is to be performed or the total amount of handled goods can be appropriately used. The objective variable Yi is typically a statistical amount calculated for a certain unit, but is not limited to this, and may be an actual amount such as the work efficiency of one task of each worker or the amount of handled goods. The objective variable Yi is not limited to the amount, and may be the result of some kind of judgment. For example, the objective variable Yi may be an integer value of a five-level evaluation that judges whether the work time of one task is large or small against multiple thresholds, or may be a binary flag that judges whether it is good or bad against one threshold. In addition, as the objective variable Yi, any warehouse information that the user judges to perform a cause analysis after an abnormality has occurred can be appropriately used.

[0016] The explanatory variables {Xj:j=1...M, M is an integer equal to or greater than 2} represent data in the warehouse information that may be the cause of changes in the objective variable Yi. Here, M represents the number of explanatory variable items obtained.

[0017] As the explanatory variables Xj, for example, the working time of the worker, the number of types of products handled, and the number of processes can be appropriately used. As with the objective variable Yi, the explanatory variables Xj are typically statistics, but are not limited to this and may be actual quantities. In addition, the explanatory variables Xj are not limited to quantities and may be the results of some judgment or category information. For example, when category information such as product names, delivery destination information, and worker names is input as the explanatory variables Xj, a numeric conversion method may be used to convert the category information into a numeric value before inputting it. As the numeric conversion method, for example, one-hot encoding, label encoding, count encoding, and target encoding can be appropriately used. In addition, as the explanatory variables Xj, any warehouse information that the user judges to be a cause of a change in the objective variable Yi can be appropriately used.

[0018] On the other hand, the calculation unit 20 calculates the cause score of the explanatory variable Xj for the change in the objective variable Yi of the target data based on the reference data group and the target data input from the acquisition unit 10. Specifically, the calculation unit 20 calculates the cause score from the reference data group and the target data using the deviation score calculation unit 21, the relationship weight calculation unit 22, the property weight calculation unit 23, and the cause score calculation unit 24.

[0019] The deviation score calculation unit 21 calculates a deviation score based on the reference data group and the target data. For example, the deviation score calculation unit 21 calculates a deviation score indicating the deviation degree of the explanatory variable Xj of the target data with respect to a first distribution regarding the explanatory variable Xj of the reference data group. As the first distribution, for example, a frequency distribution can be appropriately used. The frequency distribution may be called a frequency distribution or a histogram. As the deviation score, for example, a Z score, a score based on Hotelling's T2 method, or a score based on kernel density estimation can be appropriately used.

[0020] The relationship weight calculation unit 22 calculates the relationship weight based on the reference data group. For example, the relationship weight calculation unit 22 calculates the relationship weight representing the strength of the relationship between the objective variable Yi and the explanatory variable Xj of the reference data group. As the relationship weight, for example, the absolute value of the correlation coefficient, the correlation coefficient, the maximum information coefficient, or the cosine similarity can be appropriately used.

[0021] The property weight calculation unit 23 calculates the property weight based on the reference data group and the target data. For example, the property weight calculation unit 23 calculates a property weight representing the degree of agreement between the objective variable Yi and the explanatory variable Xj of the target data for a second distribution related to the objective variable Yi and the explanatory variable Xj of the reference data group. For example, the property weight calculation unit 23 may calculate the degree of agreement as the property weight based on the residual between the theoretical value of the objective variable Yi of the reference data group calculated by the estimation model for the second distribution and the actual measured value of the objective variable Yi of the target data. For example, a scatter diagram can be used as the second distribution. For example, a regression method such as a linear regression model or a nonlinear regression model may be used as the estimation model. The property weight may be called the weight of the residual.

[0022] The cause score calculation unit 24 calculates a cause score of an explanatory variable Xj for a change in a response variable Yi of the target data based on the deviation score, the relationship weight, and the property weight. For example, the cause score calculation unit 24 calculates a cause score for each combination of one response variable Yi and one explanatory variable Xj in the target data. In addition, for example, the cause score calculation unit 24 calculates a cause score by multiplying the deviation score, the relationship weight, and the property weight together.

[0023] The display unit 40 is a display that displays data output from the cause analysis device 1. For example, the display unit 40 displays the deviation score, the relationship weight, the property weight, the cause score, and the like output from the calculation unit 20. In addition, for example, the display unit 40 may display a reference data group or target data acquired by the acquisition unit 10.

[0024] Next, the operation of the cause analysis device configured as above will be described with reference to the flowchart of FIG.

[0025] (Step ST10) The acquisition unit 10 of the cause analysis device 1 acquires a reference data group (Yi, Xj) including a response variable Yi and an explanatory variable Xj, and target data (Yi, Xj) including the response variable Yi and the explanatory variable Xj, and inputs the reference data group and the target data to the calculation unit 20.

[0026] (Step ST20) After step ST10, the calculation unit 20 executes step ST20, which calculates a cause score of each explanatory variable Xj with respect to a change in the objective variable Yi, using the objective variables {Yi: i=1...N, N is an integer equal to or greater than 1} and the explanatory variables {Xj: j=1...M, M is an integer equal to or greater than 1}. Step ST20 includes steps ST21 to ST28.

[0027] In step ST21, the calculation unit 20 sets the item i of one or more objective variables Yi to 1 (i=1), and sets the item j of two or more explanatory variables Xj to 1 (j=1).

[0028] After step ST22, in step ST22, the calculation unit 20 calculates the deviation score So(Yi, Xj), the relationship weight score Wr(Yi, Xj), and the property weight Wp(Yi, Xj) based on the reference data group (Yi, Xj) and the target data (Yi, Xj). Note that the deviation score, the relationship weight, and the property weight are calculated by the deviation score calculation unit 21, the relationship weight calculation unit 22, and the property weight calculation unit 23 in any order.

[0029] For example, the deviation score calculation unit 21 calculates a deviation score So(Yi,Xj) indicating the deviation degree of the explanatory variable Xj of the target data with respect to the first distribution regarding the explanatory variable Xj of the reference data group. The deviation score So(Yi,Xj) is a value indicating the relative position of the explanatory variable Xj of the target data with respect to the distribution of the explanatory variable Xj of the reference data group. As the deviation score So(Yi,Xj), a Z score, which is known as a score indicating the relative position of an individual in a population, is used as an example, but is not limited thereto. For example, the deviation score So(Yi,Xj) may be a score based on Hotelling's T2 method or kernel density estimation. The deviation score So(Yi,Xj) may be calculated using a machine learning method such as a support vector machine, deep learning, or the like. The deviation score So(Yi,Xj) using the Z score is calculated by Equation (1).

[0030]

number

[0031] Here, x is the value of the explanatory variable Xj of the target data. - (The bar directly above μ) is the average value of the explanatory variable Xj in the reference data group. σ is the standard deviation of the explanatory variable Xj in the reference data group. The further the Z-score is from 0, the more it deviates from the distribution of the population. In other words, the Z-score indicates the degree to which the explanatory variable Xj of the target data deviates compared to the explanatory variable Xj of the reference data group, which is the population.

[0032] Next, the relationship weight calculation unit 22 calculates a relationship weight Wr(Yi,Xj) that indicates the strength of the relationship between the objective variable Yi and the explanatory variable Xj of the reference data group. The relationship weight Wr(Yi,Xj) will be described using the absolute value of a correlation coefficient, which is known as a coefficient that indicates the relationship between two variables, as an example, but is not limited to this. In addition, the absolute value of the relationship weight Wr(Yi,Xj) does not need to be used when the positive or negative of the relationship between the objective variable Yi and the explanatory variable Xj is significant. In addition, the relationship weight Wr(Yi,Xj) may be any value that can express the relationship between two variables, such as the maximum information coefficient, cosine similarity, and their derived forms. The relationship weight Wr(Yi,Xj) using the absolute value of the correlation coefficient is calculated from Equation (2).

[0033]

number

[0034] Here, Cov(Yi,Xj) is the covariance between the objective variable Yi and the explanatory variable Xj of the reference data group. σ Yi is the standard deviation of the objective variable Yi of the reference data set. σ Xj is the standard deviation of the explanatory variable Xj of the reference data group. The absolute value of the correlation coefficient ranges from 0 to 1, and the closer the value is to 1, the stronger the relationship between the objective variable Yi and the explanatory variable Xj.

[0035] Next, the property weight calculation unit 23 calculates a property weight Wp(Yi,Xj) representing the degree of agreement between the objective variable Yi and the explanatory variable Xj of the target data for a second distribution consisting of the objective variable Yi and the explanatory variable Xj of the reference data group. For example, the property weight calculation unit 23 may calculate the degree of agreement as the property weight based on the residual between the theoretical value of the objective variable Yi of the reference data group calculated by the estimation model for the second distribution and the actual measured value of the objective variable Yi of the target data. In this example, a linear regression model is used as the estimation model. However, the estimation model is not limited to a linear regression model, and may be a nonlinear regression model, and may be any model capable of evaluating the distribution of other reference data groups. The property weight Wp(Yi,Xj) using the residual between the theoretical value and the actual measured value is calculated by formula (3).

[0036]

number

[0037] Here, res is the residual between the theoretical value calculated from the estimation model and the actual measured value of the target data. - res (μ res The bar just above σ is the average residual between the theoretical value calculated from the estimation model and each measured value in the reference data set. res is the standard deviation of the theoretical value calculated from the estimation model, each measured value of the reference data group, and the residual. The weight of the residual takes a value between 0 and 1, and the closer the value is to 1, the closer the characteristics of the target data are to the characteristics of the reference data group.

[0038] After calculating the deviation score So(Yi, Xj), the relationship weight score Wr(Yi, Xj), and the property weight Wp(Yi, Xj), the calculation unit 20 ends step ST22.

[0039] In step ST23, the cause score calculation unit 24 calculates the cause score of the explanatory variable Xj for the change in the objective variable Yi of the target data based on the deviation score So(Yi,Xj), the relationship weight score Wr(Yi,Xj), and the property weight Wp(Yi,Xj). For example, the cause score calculation unit 24 calculates a cause score for each combination of one objective variable Yi and one explanatory variable Xj in the target data. The cause score S(Yi,Xj) is a value that evaluates the relative deviation of the target data with respect to the reference data group. Such a cause score S(Yi,Xj) is calculated by multiplying the deviation score So(Yi,Xj), the relationship weight Wr(Yi,Xj), and the property weight Wp(Yi,Xj) together, for example, as shown in formula (4).

[0040]

number

[0041] The further away from 0 the cause score S(Yi,Xj) is from the distribution of the population, the more deviated it is from the distribution of the population. In other words, the further away from 0 the cause score S(Yi,Xj) is from the distribution of the population, the more deviated the explanatory variable Xj of the target data is compared with the explanatory variable Xj of the reference data group of the population, and the more likely it is to be related to the cause of the change in the objective variable Yi. After the cause score S(Yi,Xj) is calculated, step ST23 ends.

[0042] In step ST24, the calculation section 20 judges whether or not the item j of the explanatory variable Xj has reached the maximum value "M", and if not, proceeds to step ST25.

[0043] In step ST25, the calculation unit 20 increments the item j of the explanatory variable Xj by +1 and returns to step ST22. As a result, steps ST22 to ST24 are repeatedly executed for the explanatory variable Xj of the item j that has been incremented by +1, and the deviation score So(Yi, Xj), the relationship weight Wr(Yi, Xj), the property weight Wp(Yi, Xj), and the cause score S(Yi, Xj) are calculated.

[0044] On the other hand, if the result of the determination in step ST24 is that the item j of the explanatory variable Xj has reached the maximum value "M", the calculation section 20 proceeds to step ST26.

[0045] In step ST26, the calculation section 20 determines whether or not the item i of the objective variable Yi has reached the maximum value "N", and if not, proceeds to step ST27.

[0046] In step ST27, the calculation section 20 increments the item i of the objective variable Yi by +1, and then proceeds to step ST28.

[0047] In step ST28, the calculation unit 20 sets the item j of the explanatory variable Xj to the initial value "1" and returns to step ST22. As a result, steps ST22 to ST25 are repeatedly executed for the objective variable Yi of the item i increased by +1 and the explanatory variable Xj of the item j set to the initial value "1".

[0048] On the other hand, if the result of the determination in step ST26 is that the item i of the objective variable Yi reaches the maximum value "N", the calculation unit 20 ends step ST20 including steps ST21 to ST28.

[0049] In the above step ST20, for each combination of the objective variable {Yi:i=1...N} and the explanatory variable {Xj:j=1...M}, the cause score of each explanatory variable Xj for the change of each objective variable Yi is calculated. That is, N x M cause scores are calculated. Note that the calculation of the cause score means estimating the likelihood of the cause of the change of the objective variable {Yi:i=1...N} from the trend of the data of the explanatory variable {Xj:j=1...M}. At this time, the maximum value "M" of the item j of the explanatory variable Xj is at least 2 or more. As a result, two or more cause scores S(Yi,Xj) for each explanatory variable Xj are calculated for one objective variable Yi. In addition, by comparing each calculated cause score S(Yi,Xj), the explanatory variable Xj of the target data used to calculate the maximum cause score S(Yi,Xj) can be identified as the cause.

[0050] After step ST20, the calculation unit 20 appropriately outputs the cause score S(Yi, Xj), the deviation score So(Yi, Xj), the relationship weight Wr(Yi, Xj), and the property weight Wp(Yi, Xj) to the display unit 40. The display unit 40 displays the output cause score S(Yi, Xj), deviation score So(Yi, Xj), relationship weight Wr(Yi, Xj), and property weight Wp(Yi, Xj).

[0051] As described above, according to the first embodiment, the acquisition unit 10 acquires a reference data group including an objective variable and an explanatory variable, and target data including an objective variable and an explanatory variable. The deviation score calculation unit 21 calculates an deviation score indicating the deviation degree of the explanatory variable Xj of the target data with respect to a first distribution regarding the explanatory variable Xj of the reference data group. The relationship weight calculation unit 22 calculates a relationship weight indicating the strength of the relationship between the objective variable Yi and the explanatory variable Xj of the reference data group. The property weight calculation unit 23 calculates a property weight indicating the degree of agreement between the objective variable Yi and the explanatory variable Xj of the target data with respect to a second distribution regarding the objective variable Yi and the explanatory variable Xj of the reference data group. The cause score calculation unit 24 calculates a cause score of the explanatory variable Xj with respect to a change in the objective variable Yi of the target data based on the deviation score, the relationship weight, and the property weight.

[0052] Therefore, when a change to an abnormal state occurs alone, it is possible to identify the cause of the change. To add to that, by calculating the cause score S(Yi,Xj) of the explanatory variable Xj for the change in the objective variable Yi of the target data and quantifying the degree of deviation from the reference data group and the target data, it is possible to numerically evaluate the likelihood that each explanatory variable Xj is the cause of the change in the objective variable Yi. This is expected to allow for early identification of the cause of the change in the objective variable Yi to an abnormal state.

[0053] As a comparative example, in order to deal with an anomaly that occurs singly, such as in a logistics warehouse, a method of evaluation using Z scores can be considered instead of a statistical test that calculates a p-value from the bias of anomalies when multiple anomalies occur. The evaluation method using Z scores can evaluate the extent to which the target data deviates from the frequency distribution from the average value and standard deviation of the explanatory variables Xj of the reference data group. In other words, the evaluation method using Z scores can evaluate the extent to which a single explanatory variable Xj is abnormal compared to other explanatory variables Xj. However, the Z score does not take into account the relationship between the objective variable Yi and the explanatory variables Xj, and it is difficult to identify the cause of the change in the objective variable Yi. For example, in two cases, a first case in which there is a correlation between the objective variable Yi and the explanatory variables Xj and a second case in which there is no correlation, the average value μ of the explanatory variables Xj is calculated as - If the standard deviation σ is the same, the Z-score will be calculated as the same value from formula (1). Therefore, as in the comparative example, the Z-score alone can evaluate the degree of abnormality of a single explanatory variable Xj in the target data, but in the second case where there is no correlation between the objective variable Yi and the explanatory variable Xj, it is difficult to identify the cause of the change in the objective variable Yi.

[0054] On the other hand, according to the first embodiment, unlike the technique of calculating a p-value in the framework of a conventional statistical test, the cause score uses a reference data group as a standard, and even if the target data is independent and there is no bias in the abnormality, the degree of anomaly can be evaluated and the cause can be identified. To supplement, according to the first embodiment, unlike the comparative example using only the Z score, it is possible to consider the strength of the relationship between the objective variable Yi and the explanatory variable Xj of the reference data group and the degree of agreement of the target data from the relationship by further using a relationship weight and a property weight. For example, the strength of the correlation between the objective variable Yi and the explanatory variable Xj of the reference data group can be evaluated by the relationship weight, and the degree of agreement between the reference data group and the target data can be evaluated by the property weight. Therefore, according to the evaluation based on the cause score using all of the outlier score (Z score), the relationship weight (absolute value of the correlation coefficient), and the property weight (degree of agreement), it is possible to identify the cause even for anomalies that occur independently for each condition and have no bias, such as anomalies that occur in a logistics warehouse.

[0055] According to the first embodiment, the reference data group includes two or more reference data including one or more objective variables and two or more explanatory variables. The target data includes one or more objective variables Yi and two or more explanatory variables Xj. The cause score calculation unit 24 calculates a cause score S(Yi,Xj) for each combination of one objective variable Yi and one explanatory variable Xj in the target data. In this way, in addition to the above-mentioned effects, the cause score is calculated based on each of the multiple reference data groups, so that a more likely cause can be identified.

[0056] Moreover, according to the first embodiment, the cause score calculation unit 24 calculates the cause score by multiplying the deviation score, the relationship weight, and the property weight together. In this way, in addition to the above-mentioned effects, the cause score can be calculated as a value proportional to the deviation score, the relationship weight, and the property weight.

[0057] Also, according to the first embodiment, the property weight calculation unit 23 calculates the degree of agreement as the property weight based on the residual between the theoretical value of the objective variable of the reference data group calculated by the estimation model for the second distribution and the actual measured value of the objective variable of the target data. In this way, in addition to the above-mentioned effects, the degree of agreement corresponding to the weight of the residual between the theoretical value and the actual measured value can be calculated as the property weight. For example, when the residual is small, the weight of the residual is large, and the target data has a property similar to that of the reference data group. Also, when the residual is large, the weight of the residual is small, and the target data has a property different from that of the reference data group. Therefore, such a property of the target data can be reflected in the cause score.

[0058] According to the first embodiment, the estimation model is a linear regression model or a nonlinear regression model. In addition to the above-mentioned effects, an estimation model that estimates a response variable of a reference data group can be easily realized by a regression method such as a linear regression model or a nonlinear regression model.

[0059] According to the first embodiment, the outlier score is a Z score, a score based on Hotelling's T2 method, or a score based on kernel density estimation. This allows the outlier score to be realized by various statistical methods in addition to the above-mentioned effects.

[0060] According to the first embodiment, the relationship weight is the absolute value of the correlation coefficient, the correlation coefficient, the maximum information coefficient, or the cosine similarity. In addition to the above-mentioned effects, the relationship weight can be realized by various statistical methods.

[0061] Second embodiment Fig. 3 is a block diagram showing a cause analysis device and its peripheral configuration according to the second embodiment, in which the same parts as those in Fig. 1 are given the same reference numerals and their detailed explanations are omitted, and the following mainly describes the different parts. Note that the same explanations are also omitted for the following embodiments.

[0062] Compared with the configuration shown in FIG. 1, this cause analysis device 1 further includes a generation unit 30 that generates visualization data to be displayed.

[0063] Here, the generating unit 30 generates visualization data in the form of a diagram and a table that allows the result of the cause score to be interpreted in association with the visualization. The generating unit 30 controls the display order and display contents of the visualization data. Specifically, for example, the generating unit 30 generates visualization data that visualizes and represents information including an index including at least a cause score among the deviation score, the relationship weight, the property weight, and the cause score, and an objective variable Yi and an explanatory variable Xj of the target data corresponding to the cause score. The generating unit 30 may generate visualization data including a display mode according to the cause score. The generating unit 30 may generate visualization data including a display mode in which two or more explanatory variables Xj are ranked according to the cause score. This display mode may be a mode in which explanatory variables below a predetermined rank among the ranked explanatory variables Xj are hidden or suppressed. The generating unit 30 may generate visualization data including a display mode that highlights an index that deviates from an acceptable range among the information. Furthermore, the generating unit 30 may generate visualization data including a display mode that highlights the property weight among the indexes when the deviation score and the relationship weight are within the acceptable range and the property weight is small and deviates from the acceptable range.

[0064] In addition to the functions described above, the display unit 40 is controlled by the generation unit 30 and displays the visualization data received from the generation unit 30 .

[0065] The other configuration is similar to that of the first embodiment.

[0066] Next, the operation of the cause analysis device configured as above will be described with reference to the flow chart of FIG. 3 and the schematic diagrams of FIGS.

[0067] Now, as described above, it is assumed that the miss score So(Yi,Xj), the relationship weight Wr(Yi,Xj), the property weight Wp(Yi,Xj), and the cause score S(Yi,Xj) are calculated by executing steps ST10 to ST20. The calculation unit 20 sends the miss score So(Yi,Xj), the relationship weight Wr(Yi,Xj), the property weight Wp(Yi,Xj), and the cause score S(Yi,Xj) to the generation unit 30. In addition, the calculation unit 20 sends information including the objective variable Yi and the explanatory variable Xj of the target data corresponding to the cause score to the generation unit 30. For example, the calculation unit 20 may send information including the reference data group and the target data to the generation unit 30.

[0068] (Step ST30) After step ST20, in step ST30, the generating unit 30 generates visualization data based on the information received from the calculating unit 20. That is, the generating unit 30 generates visualization data based on information including an index including at least a cause score among the deviation score, the relationship weight, the property weight, and the cause score, and the objective variable Yi and the explanatory variable Xj of the target data corresponding to the cause score. At this time, the generating unit 30 may generate visualization data including a display mode according to the cause score, such as a ranking display. The generating unit 30 may also generate visualization data including a display mode that highlights an index that deviates from an acceptable range. Thereafter, the generating unit 30 outputs the visualization data to the display unit 40.

[0069] (Step ST40) After step ST30, in step ST40, the display unit 40 displays the visualization data received from the generation unit 30. This allows the user to visually recognize the indices included in the visualization data being displayed, and the objective variable Yi and explanatory variable Xj of the target data, and confirm the cause of the change in the objective variable Yi.

[0070] Next, the visualization data generated and displayed in steps ST30 to ST40 will be supplementarily described with reference to FIGS.

[0071] The number of cause scores S(Yi,Xj) that can be displayed in the visualization data is equal to the number of combinations of the objective variable Yi and the explanatory variable Xj. The cause scores S(Yi,Xj) may be displayed in the order in which the explanatory variables Xj were obtained, but from the viewpoint of checking the explanatory variables Xj, it is preferable to display the cause scores in a ranking format. For example, a user can check the causes of change in order of importance by checking the explanatory variables Xj in the ranking order of the cause scores.

[0072] Such visualization data is generated with the display order and display contents controlled by the generation unit 30. In addition, the visualization data may include table data, scatter plot data, frequency distribution data, graph data, objective variables Yi, explanatory variables Xj, dates, etc., as appropriate.

[0073] For example, the tabular data d31 may display the cause scores S(Yi, Xj) in a ranked order according to the likelihood that they are related to the cause of the change in the objective variable Yi, as shown in Fig. 5. This tabular data d31 makes it possible to confirm the cause of the change in the objective variable Yi in a display mode in which two or more explanatory variables Xj are ranked according to the ranked cause scores S(Yi, Xj). This allows the user to consider the cause of the change from the numerical data.

[0074] For example, the graph data d32 may include scatter diagram data of the reference data group and frequency distribution data of each of the explanatory variables Xj and the objective variables Yi in the reference data group, as shown in FIG. 6. In FIG. 6, the scatter diagram data represents the relationship between the objective variables Y1 and X1 of the reference data group, with the value of Yi on the vertical axis and the value of Xj on the horizontal axis. The scatter diagram data also includes an intersection Dt between the objective variables Y1 and Xj of the target data. Each frequency distribution data is a histogram representing the distribution of the explanatory variable X1 of the reference data group, with the vertical axis representing the frequency and the horizontal axis representing the value of Xj, and a histogram representing the distribution of the objective variable Yi of the reference data group, with the vertical axis representing the value of Yi and the horizontal axis representing the frequency. Such graph data d32 makes it easy to intuitively understand how far the target data deviates from the reference data group, and visually appeals to the user the degree of change.

[0075] Alternatively, as shown in FIG. 7, the tabular data d31 may be generated based on information including indicators including a deviation score, a relationship weight, a property weight, and a cause score, and an objective variable Yi and an explanatory variable Xj of target data corresponding to the cause score. The generating unit 30 may generate the visualized data d30 including a display mode that highlights an indicator that deviates from an acceptable range among the information. For example, the generating unit 30 may obtain additional information by using individual calculation values ​​by the calculating unit 20. The generating unit 30 may generate the visualized data d30 in a display mode that highlights the obtained additional information. For example, as in the row of the explanatory variable X11 ranked 5th in FIG. 7, the deviation score "3.127" and the relationship weight "0.935" are large, and only the property weight "0.182" is small. In this case, the target data of the explanatory variable X11 may show a different tendency from the reference data group. In this way, the generating unit 30 generates visualized data including a display mode that highlights the property weight to prompt confirmation of an unknown abnormality when the deviation score and the relationship weight are within the acceptable range among the indicators and the property weight is small deviating from the acceptable range. The display mode for emphasizing the indicator may be such that the indicator to be emphasized is marked or underlined, or the format of the indicator (bold, color, size, etc.) may be changed. In this case, it becomes possible to further confirm the possibility of an unknown abnormality that is difficult to evaluate using a cause score based on a reference data group.

[0076] Here, the format of the visualization data output by the generation unit 30 will be described. The visualization data representing an image is output in the format of drawing data. Other visualization data representing things other than images is output in a data format that can be displayed on the display unit 40, such as html, xml, json, or csv.

[0077] The generating unit 30 can generate visualization data according to the amount of information and priority of the analysis results based on the cause score S(Yi,Xj) of the explanatory variable Xj related to the change in the objective variable Yi of the target data. The amount of information can be adjusted, for example, by displaying / hiding data with a low cause score in the tabular data d31, or by displaying / hiding the outlier score, the relationship weight, and the property weight in the tabular data d31. The priority can be adjusted, for example, by descending or ascending the cause score in the tabular data d31. In addition, either or both of the amount of information and the priority can be adjusted. When at least the priority is adjusted, the user can monitor the explanatory variables Xj in order from those expected to have a high relationship with the change in the objective variable Yi, so that it is expected that the burden of monitoring and oversights will be reduced.

[0078] Furthermore, the visualization data d30 may include tabular data d31, graph data d32, etc., as shown in Fig. 8. The generation unit 30 may determine the amount of information to be displayed in the visualization data and the manner of displaying it, based on the cause score.

[0079] In Fig. 8, the visualization data d30 includes information on the objective variable Yi, explanatory variables Xj, and cause score S(Yi, Xj) of the target data. Note that the visualization data d30 shown in Fig. 8 is an example in the case where there is one objective variable Yi. In the case where there are two or more objective variables Yi, the generation unit 30 may generate similar visualization data for all objective variables Yi and display each visualization data in an order on the display unit 40. In the case where an index value for each objective variable Yi is calculated, each visualization data may be arranged in an order based on the index value.

[0080] Specifically, for example, the visualization data d30 includes at least one of an objective variable name d33 representing an objective variable Yi, tabular data d31 including one or more explanatory variables and their cause scores, and scatter plot data representing the relationship between the objective variable Yi and the explanatory variable Xj. The visualization data d30 may further include key information d34 indicating the acquisition date of the target data. The key information d34 is a search key for searching the target data from a database device (not shown) of warehouse information. Note that the graph data d32 in the visualization data d30 includes scatter plot data and frequency distribution data, but is not limited thereto, and the frequency distribution data may be omitted.

[0081] Here, the tabular data d31 may include, in addition to the cause score S(Yi, Xj), the deviation score So(Yi, Xj), the relationship weight Wr(Yi, Xj), and the property weight Wp(Yi, Xj) used in calculating the cause score. Alternatively, the tabular data d31 may include only the cause score S(Yi, Xj) as shown in FIG. 5.

[0082] Moreover, the tabular data d31 does not need to include all explanatory variables Xj and cause scores S(Yi, Xj). The tabular data d31 may adjust the number of rows (amount of information) to be displayed based on the cause scores S(Yi, Xj). For example, the tabular data d31 may display rows including the names of explanatory variables Xj whose cause scores are equal to or greater than a predetermined threshold and their cause scores, and may hide (omit) rows including explanatory variables Xj whose cause scores are less than the threshold and their cause scores. In the tabular data d31, rows that are less than the threshold may be completely hidden, or may be suppressed by lightening the text color, for example, as in the row of X23 ranked 6th. The threshold for the cause score may be a value to be compared with each cause score, or may be a value to be compared with the relative ranking among all cause scores. In the latter case, the display mode of the tabular data d31 may be a mode in which explanatory variables Xj that are less than a predetermined rank among the explanatory variables Xj ranked in the same manner as described above are hidden or suppressed. The predetermined rank here is a threshold value of a relative ranking, such as the top 5. Also, in the tabular data d31, a display mode that emphasizes an index may be used, as described above. For example, the row of X11, which is ranked 5th in the tabular data d31, is marked with a mark "!" to emphasize the property weight.

[0083] Furthermore, the tabular data d31 may display elements that are controllable by the warehouse operator separately from elements that are not controllable. For example, as in the row of X2, which is ranked second in FIG. 8, a character string such as "uncontrollable" may be displayed for uncontrollable elements. Alternatively, a character string such as "controllable" may be displayed for controllable elements. Additionally, symbols may be displayed for controllable and uncontrollable elements, and controllable and uncontrollable elements may be displayed in different colors.

[0084] The graph data d32 includes up to M items for one objective variable Yi and one tabular data d31. M items is the number of explanatory variables Xj related to one objective variable Yi. The visualization data d30 may arrange M items of graph data d32 side by side, or may arrange a predetermined number of graph data less than M items. The visualization data d30 may selectively arrange one piece of graph data d32. In this case, when the generation unit 30 receives an operation to select a row including the explanatory variable Xj in the tabular data d31, the visualization data d30 may be updated to switch to the graph data d32 related to the selected explanatory variable Xj.

[0085] In the scatter diagram format data in the graph data d32, when the relationship weight Wr(Yi,Xj) is smaller than a predetermined threshold Thr, the relationship weight Wr(Yi,Xj) may be highlighted. For example, the highlighting may be performed by surrounding the area of ​​the intersection between the objective variable Yi and the explanatory variable Xj related to the relationship weight Wr(Yi,Xj) in red. For the highlighting, the explanatory variable name indicating the explanatory variable Xj related to the relationship weight Wr(Yi,Xj) may be displayed in bold, or a warning or caution mark may be displayed in red. For the property weight Wp(Yi,Xj) is smaller than a predetermined threshold Thr, the property weight Wp(Yi,Xj) may be highlighted in the same manner as described above.

[0086] As described above, according to the second embodiment, the generating unit 30 generates visualization data that visualizes and represents information including indices including at least a cause score among the outlier score, the relationship weight, the property weight, and the cause score, and the objective variable and the explanatory variable of the target data corresponding to the cause score. Therefore, in addition to the above-mentioned effects, it is possible to visualize and represent the result of analyzing the likelihood of a change in the objective variable Yi being a cause.

[0087] Furthermore, according to the second embodiment, the generator 30 may generate visualization data including a display mode according to the cause score. In this case, in addition to the above-mentioned effects, the display mode of the visualization data can be made easier to view according to the cause score.

[0088] According to the second embodiment, the target data includes one or more objective variables and two or more explanatory variables. The generating unit 30 may generate visualization data including a display form in which two or more explanatory variables are ranked according to the cause score. In this case, the ranked explanatory variables can be visualized and displayed according to the cause score.

[0089] According to the second embodiment, the display mode may be a mode in which explanatory variables that are ranked below a predetermined rank are hidden or suppressed from being displayed. In this case, the user does not need to check explanatory variables that are unlikely to be the cause, thereby reducing the burden on the user.

[0090] According to the second embodiment, the generating unit 30 may generate visualization data including a display mode that emphasizes indicators that deviate from the allowable range among the information. In this case, in addition to the above-mentioned effects, the user can easily visually recognize the emphasized indicators.

[0091] According to the second embodiment, the generating unit 30 may generate visualization data including a display mode that, when the outlier score and the relationship weight of the index are within the allowable range and the property weight is small and deviates from the allowable range, emphasizes the property weight to prompt the user to confirm an unknown anomaly. In this case, in addition to the above-mentioned effect, the user can confirm whether the explanatory variable related to the property weight that is small and deviates from the allowable range is an unknown anomaly.

[0092] (Third embodiment) 9 is a block diagram showing a cause analysis device and its peripheral configuration according to the third embodiment. Compared to the configuration shown in FIG. 3, this cause analysis device 1 further includes an operation unit 50 that accepts user operations.

[0093] Accordingly, the generating unit 30 updates the visualization data in response to the received operation in addition to the above-mentioned functions. The generating unit 30 may update the visualization data to a display form in which part of the information is displayed or hidden. The part of the information may include at least one of tabular data d31 related to the indexes and scatter plot data related to the second distribution and the target data. The part of the information may also include at least one of tabular data d31 related to the indexes and graph data d32. The generating unit 30 may also update the visualization data to a display form in which the order of the indexes is changed. The generating unit 30 may also change the order of the indexes to follow any of the descending or ascending order of the cause score, the descending or ascending order of the deviation score, the descending or ascending order of the relationship weight, and the descending or ascending order of the property weight.

[0094] The other configuration is similar to that of the second embodiment.

[0095] Next, the operation of the cause analysis device configured as above will be described with reference to the flow chart of FIG. 10 and the schematic diagrams of FIGS.

[0096] Now, it is assumed that visualized data is being displayed on the display unit 40 by executing steps ST10 to ST40 in the same manner as described above.

[0097] (Step ST50) After step ST40, the generation unit 30 executes step ST50 of updating the visualization data in response to a user's operation of the operation unit 50. Step ST50 includes steps ST51 to ST53.

[0098] In step ST51, the operation unit 50 accepts an operation from a user. The operation unit 50 inputs, to the generation unit 30, an operation signal corresponding to the accepted operation.

[0099] In step ST52, the generation unit 30 determines whether or not to end the display in response to the received operation, and if the display is to be ended, controls the display unit 40 to end the display of the visualized data. For example, the generation unit 30 determines whether an operation signal corresponding to the received operation is an end command, and controls the display unit 40 in response to the determination result. On the other hand, if the result of this determination is no, the process proceeds to step ST53.

[0100] In step ST53, the generation unit 30 changes the visualization data according to the received operation. For example, the generation unit 30 identifies the contents of an operation signal according to the received operation, and regenerates the visualization data according to the identification result. After completing step ST53, the cause analysis device 1 returns to step ST40 and repeatedly executes the processes of steps ST40 to ST50.

[0101] Next, examples of the visualization data used in step ST53 will be described with reference to FIGS.

[0102] In the example shown in FIG. 11, the visualization data d30 includes a change button Bt for changing the amount of information. For example, in the case of a high amount of information, the visualization data includes table data d31 and graph data d32 related to the objective variable Yi, an objective variable name d33 of the objective variable Yi, and a change button Bt indicating a decrease in the amount of information after the change "-". In addition, in the case of a low amount of information, the visualization data does not include the table data d31 and graph data d32, but includes an objective variable name d33 of the objective variable Yi, and a change button Bt indicating an increase in the amount of information after the change "+". It is to be noted that the visualization data with a low amount of information is not limited to this, and may further include the explanatory variable name of the explanatory variable Xj and at least an index value of the cause score. The change button Bt may omit the notation "-" or "+" indicating the increase or decrease in the amount of information after the change.

[0103] 12, the generation unit 30 changes the visualization data d30 to increase or decrease the amount of information in response to the operation of the change button Bt. Note that the difference between a state S1 in which the amount of information has increased and a state S2 in which the amount of information has decreased is whether the visualization data d30 includes tabular data d31 and graph data d32. In this manner, the visualization data d30 after the amount of information has been increased or decreased is displayed on the display unit 40.

[0104] Next, an example of visualized data d30 shown in Fig. 13 and Fig. 14 will be described. Compared to the example shown in Fig. 11, this visualized data d30 further includes a hide button Bh and a show more button Bs for changing the amount of information, and a radio button Br for changing the order of the information. The hide button Bh is a button for instructing a decrease in the amount of information in the tabular data d31. The show more button Bs is a button for instructing an increase in the amount of information in the tabular data d31. The radio button Br is a button for selecting the order of the information in the tabular data d31. Note that an interface such as a check box or a pull-down menu may be used instead of the radio button Br.

[0105] First, among the non-display button Bh, the expand button Bs, and the radio button Br, the operation of the non-display button Bh and the expand button Bs for changing the amount of information will be described.

[0106] In the case of a large amount of information, as shown in FIG. 13, the display unit 40 displays visualization data d30 including tabular data d31 regarding explanatory variables Xj corresponding to cause scores S(Yi, Xj) of all ranks and a non-display button Bh.

[0107] In the case of a low amount of information, the display unit 40 displays visualization data d30 including tabular data d31 regarding explanatory variables Xj corresponding to the top five cause scores S(Yi, Xj) and an expansion button Bs, as shown in Fig. 14. Note that the top five are merely an example, and may be changed to another order.

[0108] Here, in FIG. 13, the generation unit 30 changes the visualized data d30 to reduce the amount of information in response to the operation of the hide button Bh. As a result, as shown in FIG. 14, visualized data d30 with a low amount of information is displayed on the display unit 40. Also, in FIG. 14, the generation unit 30 changes the visualized data d30 to increase the amount of information in response to the operation of the expand button Bs. As a result, visualized data d30 with a high amount of information is displayed on the display unit 40, as shown in FIG. 13. That is, the generation unit 30 changes the tabular data d31 to increase or decrease the amount of information in the visualized data d30 in response to the operation of the hide button Bh and the expand button Bs. As a result, the visualized data d30 including the tabular data d31 after the amount of information has been increased or decreased and the hide button Bh or the expand button Bs for returning the amount of information is displayed on the display unit 40.

[0109] In addition, without being limited to this, immediately after operating the expand button Bs, the generation unit 30 may display the tabular data d31 in a folded state with information (hereinafter referred to as lower-level information) related to the explanatory variables Xj corresponding to the cause scores S(Yi, Xj) that are less than the top 5.

[0110] Furthermore, the generating unit 30 may generate the tabular data d31 in a state where lower level information is expanded in response to an operation of the expand button Bs.

[0111] Furthermore, in response to an operation of the expand button Bs, the generating unit 30 may generate tabular data d31 including information on the explanatory variables Xj corresponding to the top five cause scores in descending order of the cause scores.

[0112] Furthermore, in response to an operation of an expand button Bs, the generating unit 30 may generate visualization data d30 including tabular data d31 with a high amount of information and a non-display button Bh for returning to a low amount of information.

[0113] Next, the operation of the radio button Br for changing the order of information in the visualization data d30 will be described.

[0114] 13 and 14, the display unit 40 displays visualization data d30 including radio buttons Br for selecting score order or user-specified order and ascending order or descending order. Here, the operation unit 50 accepts an operation to specify the sorting order of the information in response to an operation of selecting score order or user-specified order and then ascending order or descending order.

[0115] The generating unit 30 generates visualization data d30 in which the order of information in the tabular data d31 is changed in response to an operation of the radio button Br. As a result, the visualization data d30 including the tabular data d31 in the specified order is displayed on the display unit 40. In this way, the display priority can be changed by selecting ascending order or descending order. In addition, the information on which the order is based can be changed by selecting score order or user-specified order. In addition, in FIG. 13 and FIG. 14, the score order is the order of explanatory variables Xj based on the minimum or maximum score of each cause score. In the descending order, the score order is based on the maximum score, and in the ascending order, the score order is based on the minimum score. In addition, the user-specified order is the order of explanatory variables Xj based on the minimum or maximum value of an index specified by the user among indexes other than the cause score. In the descending order, the user-specified order is based on the maximum value, and in the ascending order, the minimum value. The generation unit 30 generates visualization data d30 including tabular data d31 in which the sort order is changed to either descending order by score, ascending order by score, descending order by user specified order, or ascending order by user specified order, depending on the operation of the radio button Br.

[0116] As described above, according to the third embodiment, the operation unit 50 accepts a user's operation. The generation unit 30 updates the visualization data in response to the accepted operation. Therefore, in addition to the effect of the second embodiment, it is possible to update the visualization data in response to a user's operation.

[0117] Furthermore, according to the third embodiment, the generation unit 30 updates the visualization data to a display form in which part of the information included in the visualization data is displayed or hidden. Therefore, in addition to the above-mentioned effects, the amount of information in the visualization data can be increased or decreased in response to a user operation.

[0118] Furthermore, according to the third embodiment, the part of the information includes at least one of tabular data on the index and scatter plot data on the second distribution and the target data. Therefore, in addition to the above-mentioned effects, the amount of information of the part of the tabular data and the scatter plot data included in the visualization data can be increased or decreased according to a user's operation.

[0119] Moreover, according to the third embodiment, the generating unit 30 updates the visualization data to a display form in which the order of the indices is changed. Therefore, in addition to the above-mentioned effects, the order of the indices included in the visualization data can be changed, so that the priority of the indices that prompt confirmation can be changed. Furthermore, the user can not only monitor the explanatory variables Xj that are expected to be highly related to the change in the objective variable Yi with priority, but also check the analysis results from more diverse perspectives. For example, in daily monitoring work, it is possible to realize a usage in which the items that are expected to be highly related to the anomaly that are displayed first are checked, and in situations where detailed monitoring is required, items that are relatively less related to the anomaly are also monitored.

[0120] Furthermore, according to the third embodiment, the generation unit 30 changes the order of the indices to follow any one of the following: descending or ascending order of the cause score, descending or ascending order of the deviation score, descending or ascending order of the relationship weight, or descending or ascending order of the property weight. Therefore, in addition to the above-mentioned effects, when the user checks the cause, the cause score and the indices used for calculating the cause score can be rearranged.

[0121] (Modification of the third embodiment) In the third embodiment, the amount of information of the tabular data d31 is increased or decreased by the hide button Bh and the expand button Bs, but the present invention is not limited to this. For example, as shown in FIG. 15, the generation unit 30 may generate the visualization data d30 so that the information consisting of the objective variable name d33 and the change button Bt related to the objective variable Yi is switched between hidden and displayed by the hide button Bh and the expand button Bs. The difference between the state S2 in which the amount of information is reduced and the state S3 in which the amount of information is further reduced is whether the visualization data d30 includes the information consisting of the objective variable name d33 and the change button Bt. In this way, the visualization data d30 after the amount of information is increased or decreased is displayed on the display unit 40. According to this modification, in addition to the effect of the third embodiment, visualization data in the state S3 in which the amount of information is further reduced can be generated.

[0122] In the third embodiment, the display of the details of the graph data d32 is not described, but the present invention is not limited to this. For example, as shown in FIG. 16, when the cursor cs is placed on the graph data d32, the generating unit 30 generates visualization data d30 including details of the data indicated by the cursor cs. The display unit 40 displays the generated visualization data d30. For example, when the cursor cs is placed on an element of the scatter diagram format data of the graph data d32 in response to a user's operation, the generating unit 30 generates visualization data d30 including the value of the objective variable Yi and the value of the explanatory variable Xj of the element. The display unit 40 displays the generated visualization data d30. According to this modification, in addition to the effect of the third embodiment, visualization data including detailed numerical values ​​of the graph data d32 can be generated in response to a user's operation.

[0123] In the third embodiment, visualization data including both controllable elements and uncontrollable elements is generated, but the present invention is not limited to this. For example, the generation unit 30 may generate visualization data including only one of the controllable elements and the uncontrollable elements in response to a user's operation. The uncontrollable elements are explanatory variables X2 to which "uncontrollable" is added and an index related to the explanatory variables X2. The controllable elements are explanatory variables Xj other than the uncontrollable elements and their indexes. According to this modified example, in addition to the effect of the third embodiment, visualization data including only one of the controllable elements and the uncontrollable elements can be generated in response to a user's operation.

[0124] In the third embodiment, the threshold for the cause score is not changed, but the present invention is not limited to this. For example, the generating unit 30 may change the threshold for the cause score in response to a user operation, and change the display target included in the tabular data d31 based on the changed threshold. According to such a modified example, in addition to the effect of the third embodiment, visualization data in which the display target is changed in response to a user operation can be generated.

[0125] In the third embodiment, the order of the cause scores is exemplified as descending or ascending order of the score order and descending or ascending order of the user-specified order, but is not limited thereto. For example, the generating unit 30 may change the order of the cause scores to an arbitrary order in response to a user's operation. For example, the order of the explanatory variables Xj may be used as the arbitrary order. Alternatively, the order of the difference between the multiplication value of the deviation score and the relationship weight and the property weight may be used as the arbitrary order. That is, the generating unit 30 may rearrange the cause scores in an order based on a value not included in the tabular data d31. According to such a modified example, in addition to the effect of the third embodiment, visualization data in which the order of the cause scores is changed to an arbitrary order in response to a user's operation can be generated.

[0126] In the third embodiment, visualization data is generated in which the state of the amount of information and the order of information related to the objective variable Y1 are changed, but the present invention is not limited to this, and the objective variable Yi may be specified in response to a user operation. That is, the generation unit 30 may generate visualization data in which the state of the amount of information and the order of information related to the specified objective variable Yi are changed. According to such a modification, in addition to the effect of the third embodiment, by specifying the objective variable Yi in response to a user operation, visualization data related to the specified objective variable Yi can be generated.

[0127] (Fourth embodiment) 17 is a block diagram illustrating a hardware configuration of a cause analysis device according to a fourth embodiment. The fourth embodiment is a specific example of the first to third embodiments, and is a form in which the cause analysis device 1 is realized by a computer.

[0128] The cause analysis device 1 includes, as hardware, a CPU (Central Processing Unit) 201, a RAM (Random Access Memory) 202, a program memory 203, an auxiliary storage device 204, and an input / output interface 205. The CPU 201 communicates with the RAM 202, the program memory 203, the auxiliary storage device 204, and the input / output interface 205 via a bus. That is, the cause analysis device 1 of this embodiment is realized by a computer having such a hardware configuration.

[0129] The CPU 201 is an example of a general-purpose processor. The RAM 202 is used by the CPU 201 as a working memory. The RAM 202 includes a volatile memory such as a Synchronous Dynamic Random Access Memory (SDRAM). The program memory 203 stores a cause analysis program for implementing each unit according to each embodiment. The cause analysis program may be, for example, a program for causing a computer to realize each function of the acquisition unit 10, the calculation unit 20, and the generation unit 30. In addition, for example, a Read-Only Memory (ROM), a part of the auxiliary storage device 204, or a combination thereof is used as the program memory 203. The auxiliary storage device 204 stores data non-temporarily. The auxiliary storage device 204 includes a non-volatile memory such as a hard disc drive (HDD) or a solid state drive (SSD).

[0130] The input / output interface 205 is an interface for connecting to other devices, and is used to connect, for example, an operation unit 50 such as a keyboard and a mouse, a database device (not shown) that stores reference group data and subject data, and a display unit 40 such as a display.

[0131] The cause analysis program stored in the program memory 203 includes computer executable instructions. When the data analysis program (computer executable instructions) is executed by the CPU 201, which is a processing circuit, the data analysis program causes the CPU 201 to execute a predetermined process. For example, when the cause analysis program is executed by the CPU 201, the CPU 201 executes a series of processes described with respect to each unit in FIG. 1, FIG. 3, or FIG. 9. For example, when the computer executable instructions included in the cause analysis program are executed by the CPU 201, the CPU 201 executes a cause analysis method. The cause analysis method may include each operation corresponding to each function of the acquisition unit 10, the calculation unit 20, and the generation unit 30 described above. In addition, the cause analysis method may include each operation shown in FIG. 2, FIG. 4, FIG. 10, FIG. 12, or FIG. 15 as appropriate.

[0132] The cause analysis program may be provided to the cause analysis device 1, which is a computer, in a state where it is stored in a computer-readable storage medium. In this case, for example, the cause analysis device 1 further includes a drive (not shown) for reading data from the storage medium, and acquires the cause analysis program from the storage medium. As the storage medium, for example, a magnetic disk, an optical disk (CD-ROM, CD-R, DVD-ROM, DVD-R, etc.), a magneto-optical disk (MO, etc.), a semiconductor memory, etc. can be appropriately used. The storage medium may be called a non-transitory computer readable storage medium. Also, the cause analysis program may be stored in a server on a communication network, and the cause analysis device 1 may download the cause analysis program from the server using the input / output interface 205.

[0133] The processing circuit that executes the cause analysis program is not limited to a general-purpose hardware processor such as the CPU 201, but may be a dedicated hardware processor such as an ASIC (Application Specific Integrated Circuit). The term processing circuit (processing unit) includes at least one general-purpose hardware processor, at least one dedicated hardware processor, or a combination of at least one general-purpose hardware processor and at least one dedicated hardware processor. In the example shown in FIG. 17, the CPU 201, the RAM 202, and the program memory 203 correspond to the processing circuit.

[0134] According to at least one of the embodiments described above, when a change to an abnormal state occurs alone, it becomes possible to identify the cause of the change.

[0135] Although some embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included in the scope of the invention and its equivalents described in the claims, as well as in the scope and spirit of the invention. [Explanation of symbols]

[0136] 1...cause analysis device, 10...acquisition unit, 20...calculation unit, 21...outlier score calculation unit, 22...relationship weight calculation unit, 23...property weight calculation unit, 24...cause score calculation unit, 30...generation unit, 40...display unit, 50...operation unit, d30...visualized data, d31...tabular data, d32...graph data, d33...objective variable name, d34...key information, Bt...change button, Br...radio button, Bh...hide button, Bs...expand button.

Claims

1. an acquisition unit that acquires a reference data group including a response variable and an explanatory variable, and target data including the response variable and the explanatory variable; a deviation score calculation unit that calculates a deviation score indicating a deviation degree of the explanatory variable of the target data with respect to a first distribution regarding the explanatory variable of the reference data group; a relationship weight calculation unit that calculates a relationship weight representing the strength of a relationship between an objective variable and an explanatory variable of the reference data group; a property weight calculation unit that calculates a property weight representing a degree of agreement between values ​​of the objective variable and the explanatory variable of the target data for a second distribution regarding the objective variable and the explanatory variable of the reference data group; a cause score calculation unit that calculates a cause score of an explanatory variable for a change in the objective variable of the target data based on the deviation score, the relationship weight, and the property weight; A cause analysis device comprising:

2. the reference data group includes two or more pieces of reference data each including one or more objective variables and two or more explanatory variables; The target data includes one or more objective variables and two or more explanatory variables, The cause analysis device according to claim 1 , wherein the cause score calculation unit calculates the cause score for each combination of one objective variable and one explanatory variable in the target data.

3. The cause analysis device according to claim 1 , wherein the cause score calculation unit calculates the cause score by multiplying the deviation score, the relationship weight, and the property weight together.

4. 2. The cause analysis device according to claim 1, further comprising: a generation unit that generates visualization data that visualizes and represents information including an index including at least the cause score among the deviance score, the relationship weight, the property weight, and the cause score, and a response variable and an explanatory variable of the target data corresponding to the cause score.

5. The cause analysis device according to claim 4 , wherein the generation unit generates the visualization data including a display form according to the cause score.

6. The target data includes one or more objective variables and two or more explanatory variables, The cause analysis device according to claim 5 , wherein the generation unit generates the visualization data including a display form in which the two or more explanatory variables are ranked according to the cause score.

7. 7. The cause analysis device according to claim 6, wherein the display mode is a mode in which explanatory variables of the ranked explanatory variables below a predetermined rank are hidden or displayed in a suppressed manner.

8. The cause analysis device according to claim 4 , wherein the generation unit generates the visualization data including a display form that highlights indicators that deviate from an allowable range, out of the information.

9. 9. The cause analysis device according to claim 8, wherein, when the outlier score and the relationship weight are within an acceptable range and the property weight is small and deviates from the acceptable range, the generation unit generates the visualization data including a display mode that emphasizes the property weight to prompt confirmation of an unknown anomaly.

10. 2. The cause analysis device according to claim 1, wherein the property weight calculation unit calculates the degree of agreement as the property weight based on a residual between a theoretical value of the dependent variable of the reference data group calculated by an estimation model for the second distribution and an actual measured value of the dependent variable of the target data.

11. The cause analysis device according to claim 10 , wherein the estimation model is a linear regression model or a nonlinear regression model.

12. The causal analysis device according to claim 1 , wherein the outlier score is a Z-score, a score based on Hotelling's T2 method, or a score based on kernel density estimation.

13. The cause analysis device according to claim 1 , wherein the relationship weight is an absolute value of a correlation coefficient, a correlation coefficient, a maximum information coefficient, or a cosine similarity.

14. An operation unit that accepts an operation by a user, The cause analysis device according to claim 4 , wherein the generation unit updates the visualization data in response to the received operation.

15. The cause analysis device according to claim 14 , wherein the generation unit updates the visualization data to a display form in which a part of the information is displayed or hidden.

16. The cause analysis device according to claim 15 , wherein the portion of information includes at least one of tabular data regarding the indicator and scatter plot data regarding the second distribution and the target data.

17. The cause analysis device according to claim 14 , wherein the generation unit updates the visualization data to a display form in which an order of the indicators is changed.

18. The cause analysis device according to claim 17 , wherein the generation unit changes the order of the indexes to follow any one of a descending or ascending order of the cause score, a descending or ascending order of the deviation score, a descending or ascending order of the relationship weight, or a descending or ascending order of the property weight.

19. Obtaining a reference data group including a response variable and an explanatory variable, and target data including the response variable and the explanatory variable; Calculating a deviation score indicating a deviation degree of the explanatory variable of the target data with respect to a first distribution regarding the explanatory variable of the reference data group; calculating a relationship weight representing the strength of a relationship between the objective variable and the explanatory variable of the reference data group; calculating a property weight representing a degree of agreement between values ​​of the objective variable and the explanatory variable of the target data for a second distribution regarding the objective variable and the explanatory variable of the reference data group; Calculating a causal score of an explanatory variable for a change in a dependent variable in the target data based on the deviation score, the relationship weight, and the property weight; A cause analysis method that includes:

20. Computer, an acquisition unit that acquires a reference data group including a response variable and an explanatory variable, and target data including the response variable and the explanatory variable; a deviation score calculation unit that calculates a deviation score indicating a degree of deviation of the explanatory variable of the target data with respect to a first distribution regarding the explanatory variable of the reference data group; a relationship weight calculation unit that calculates a relationship weight that indicates the strength of a relationship between an objective variable and an explanatory variable of the reference data group; a property weight calculation unit that calculates a property weight representing a degree of agreement between values ​​of the objective variable and the explanatory variable of the target data for a second distribution regarding the objective variable and the explanatory variable of the reference data group; a cause score calculation unit that calculates a cause score of an explanatory variable for a change in a target variable in the target data based on the deviation score, the relationship weight, and the property weight; A cause analysis program to function as.

Images