Causal exploration device, causal exploration method, and program

Abstract

A causality search device 10 comprises: a causality information calculation unit 11 that selects different two components from among a plurality of components provided in a subject system and calculates causality information representing a causal relationship between the selected two components; and a causality information modification unit 12 that modifies the causality information on the basis of feature information representing a feature associated with each of the selected two components.

Description

[Technical field]

[0001] The technical field relates to a causal relationship exploration device and a causal relationship exploration method, as well as a program for implementing these. To Regarding. [Background technology]

[0002] In systems such as OT (Operational Technology) and IoT (Internet of Things), when an abnormality is detected in the system, the cause of the abnormality must be inferred and a countermeasure must be devised to restore the system to the desired state. However, inferring the cause of the abnormality and devising a countermeasure requires accurate causal relationships.

[0003] As a related technique, Patent Document 1 discloses a causal relationship learning device that estimates causal relationships without the need for presetting a regression model. According to the causal relationship learning device of Patent Document 1, first, a feature amount of time-series data is calculated using a correct answer label of classification labels classified into three or more categories related to the causal relationships of time-series data and the time-series data corresponding to the correct answer label. Next, a classifier is trained using a pair of the calculated feature amount and the correct answer label so that the output of the classifier for the feature amount becomes the maximum value of the output value of the correct answer label. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2019-185194 A Summary of the Invention [Problem to be solved by the invention]

[0005] Since systems such as OT and IoT are physical systems, spurious correlation may exist in which data output by one component appears to be synchronized with data output by the other component even though there is no correlation between the data actually output by the two components. When spurious correlation exists, if a regression model such as that disclosed in Patent Document 1 is used, regression will be established even though there is no causal relationship between the components.

[0006] However, since it is difficult to eliminate spurious correlations, even if the regression model disclosed in Patent Document 1 is used, it is difficult to accurately estimate causal relationships.

[0007] As one aspect, a causal relationship is accurately estimated by a causal exploration device, a causal exploration method, and program The purpose is to provide. [Means for solving the problem]

[0008] In order to achieve the above object, a causal exploration device in one aspect includes: a causal information calculation unit that selects two different components from a plurality of components provided in a target system and calculates causal information that indicates a causal relationship between the two selected components; a causal information correcting unit that corrects the causal information based on function information that indicates functions associated with each of the two selected components; The present invention is characterized by having the following.

[0009] In order to achieve the above object, a causal exploration method in one aspect includes: a causal information calculation step of selecting two different components from a plurality of components provided in the target system and calculating causal information representing a causal relationship between the two selected components; a causal information modifying step of modifying the causal information based on functional information representing functions associated with each of the two selected components; The present invention is characterized by having the following.

[0010] Furthermore, in order to achieve the above object, M is , a causal information calculation step of selecting two different components from a plurality of components provided in the target system and calculating causal information representing a causal relationship between the two selected components; a causal information modifying step of modifying the causal information based on functional information representing functions associated with each of the two selected components; Run Ruko It is characterized by: Effect of the Invention

[0011] One aspect is that causal relationships can be estimated with high accuracy. [Brief description of the drawings]

[0012] [Figure 1] FIG. 1 is a diagram illustrating an example of a causal exploration device. [Diagram 2] FIG. 2 is a diagram illustrating an example of a data structure of the component information. [Diagram 3] FIG. 3 is a diagram illustrating an example of the causal information. [Figure 4] FIG. 4 is a diagram illustrating an example of the modification determination information. [Diagram 5] FIG. 5 is a diagram illustrating an example of a system having a causal exploration device. [Figure 6] FIG. 6 is a diagram for explaining an example of the operation of the causal exploration device. [Figure 7] FIG. 7 is a diagram for explaining an example of a causal exploration device according to the first modification. [Figure 8] FIG. 8 is a diagram illustrating an example of a causal exploration device according to the second modification. [Figure 9]FIG. 9 is a diagram illustrating an example of a computer that realizes a causal exploration device in the embodiment and the first and second modifications. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] Hereinafter, an embodiment will be described with reference to the drawings. In the drawings described below, elements having the same or corresponding functions are denoted by the same reference numerals, and repeated description thereof may be omitted.

[0014] (Embodiment) The configuration of a causal exploration device in the first embodiment will be described with reference to Fig. 1. Fig. 1 is a diagram for explaining an example of a causal exploration device.

[0015] [Device configuration] The causal exploration device 10 is a device that accurately estimates a causal relationship between two components provided in a target system, based on the functions of each of the two components.

[0016] The target system is a system that uses an OT / IoT network, etc. The target system is, for example, a system used in a power plant, a transportation facility, a factory, an airplane, an automobile, a home appliance, etc. The target system has multiple components.

[0017] Components include, for example, setting devices such as switches and relays installed in the target system, driving devices such as actuators, pumps and robot arms, and measuring devices that output signals or information such as measured pressure, flow rate, temperature, voltage and current.

[0018] The causal exploration device 10 is, for example, an information processing device such as a CPU (Central Processing Unit), a programmable device such as an FPGA (Field-Programmable Gate Array), a GPU (Graphics Processing Unit), or a circuit equipped with any one or more of these, a server computer, a personal computer, or a mobile terminal.

[0019] As shown in FIG. 1, the causal exploration device 10 has a causal information calculation unit 11 and a causal information correction unit 12.

[0020] The causal information calculation unit 11 selects two different components (a set) from a plurality of components provided in the target system, and calculates causal information that indicates a causal relationship between the selected two components (a set).

[0021] The causal information correction unit 12 corrects the causal information based on the function information indicating the functions associated with each of the two selected components.

[0022] In this manner, in the embodiment, the causal relationship can be accurately estimated by correcting the causal information indicating the causal relationship between two components based on the functions associated with each of the components.

[0023] For example, since systems such as IoT / OT are physical systems, there is a lot of sensor data that behaves in a roughly synchronized manner with a delay, making it difficult to eliminate false correlations. However, by using the embodiment, it is possible to eliminate false correlations to a certain extent, and therefore it is possible to accurately estimate causal relationships.

[0024] The causal information calculation unit 11 will be specifically described. The causal information calculation unit 11 first selects two components in a permutation from among a plurality of components based on component information stored in a storage device (not shown).

[0025] 2 is a diagram illustrating an example of a data structure of component information. In each of the component information 21 and 22, an identifier for identifying a component, time-series data acquired from the component, and function information indicating a function of the component are associated with each other.

[0026] The time series data is data acquired in a time series from a component. In the example of the component information 21 and 22 in Fig. 2, the time series data Du is associated with the component U, the time series data Dx is associated with the component X, the time series data Dy is associated with the component Y, and the time series data Dz is associated with the component Z.

[0027] The function information is, for example, information that indicates the functions of a setting device, a driving device, a measuring device, etc. However, there are cases where the function information is not associated with an identifier.

[0028] When the functions of components are known in advance, it is possible to associate information (labels) representing the following functions with each component, as in component information 21 in Fig. 2. The label of a setting device may be, for example, SV (Set Variable: set value (target value)), the label of a driving device may be, for example, MV (Manipulative Variable: manipulated variable (output)), and the label of a measuring device may be, for example, PV (Process Variable: measured value).

[0029] In addition to the above-mentioned labels, the labels may be information indicating a server, a client PC, a network device, an IoT device, a rank in a chain of command, or the like.

[0030] Furthermore, when the function of a component is not clearly understood, the same label may be associated with components that have similar functions, as in the component information 22 in Fig. 2. In the component information 22, since the functions of components X and Z are similar, a label α that indicates the same function is set as the function information.

[0031] The selection of two components will be described. In the component information 21 or 22 in Fig. 2, when three components related to identifiers X, Y, and Z stored in the component information are targeted, the causal information calculation unit 11 selects six pairs of two components in a permutation: (X → Y) (Y → X) (X → Z) (Z → X) (Y → Z) (Z → Y). That is, the causal information calculation unit 11 selects a pair of components with a direction.

[0032] Next, the causal information calculation unit 11 inputs the time series data of the two components for each pair into a causal model for generating causal information representing the causal relationship between the components, and calculates the causal information.

[0033] Causal information is information that indicates the causal relationship between time series data obtained from two components included in a pair. Examples of causal information include (1) and (2) shown below.

[0034] The causal information in (1) is binary information that indicates whether or not there is a causal relationship between the time-series data obtained from the two components. For example, the binary value is "1" if there is a causal relationship and "0" if there is no causal relationship.

[0035] The causal information in (2) is an index that indicates the degree of causal relationship between the time series data obtained from two components. The index is, for example, a directional score (causal score).

[0036] Fig. 3 is a diagram for explaining an example of causal information. In the case of three components related to identifiers X, Y, and Z stored in the component information as in the example of Fig. 2, the causal information (1) is stored in a storage device by associating information (one of two values) indicating whether or not there is a causal relationship with each of six combinations as shown in 31 of Fig. 3.

[0037] Furthermore, the causal information of (2) is stored in a storage device by associating a directional causal score (Score 1 to 6) as causal information with each of the six combinations, as shown by 32 in FIG.

[0038] The causal model may use score-based methods such as Granger Causality, Transfer Entropy, likelihood, and information criteria (AIC: Akaike Information Criterion, BIC: Bayesian information criterion), structural equations, LiNGAM (Linear Non-Gaussian Acyclic Model), logistic regression, PCA (Principal Component Analysis), etc. Also, it may be used in combination with constraint-based methods such as the PC algorithm and TPDA (Three Phase Dependency Analysis) algorithm.

[0039] The causal information correcting unit 12 will be specifically described. The causal information correction unit 12 corrects the causal information calculated by the causal information calculation unit 11 based on the function information indicating the functions associated with each of the two selected components.

[0040] Specifically, the causal information correction unit 12 first refers to the component information for each pair using the identifiers of the two components, and acquires the function information associated with each of the two components.

[0041] 2, for example, if the above-mentioned pair is (X→Y), the label SV associated with the identifier X and the label MV associated with the identifier Y are obtained. For the pair (Y→X) (X→Z) (Z→X) (Y→Z) (Z→Y), the function is obtained in a similar manner.

[0042] Next, the causal information correction unit 12 uses the functions included in the two pieces of acquired functional information to refer to the correction decision information stored in advance in the storage device, and acquires correction information for correcting the causal information.

[0043] 4 is a diagram for explaining an example of modification decision information. The modification decision information is information in which a set of functions associated with two components is associated with modification information for modifying causal information.

[0044] Examples of the correction information include (A), (B), (C), and (D) shown below. The correction information in (A) is binary information that indicates whether or not there is a causal relationship, based on the relationship between a pair of functions (a pair of two functions with a direction). For example, the binary value is "1" if there is a causal relationship, and "0" if there is no causal relationship.

[0045] When the function information is labeled SV, MV, and PV, as in the component information 21 of FIG. 2, the modification information (A) associates information indicating whether or not there is a causal relationship (one of two values) with each of the nine pairs of functions, as shown in the modification decision information 41 of FIG. 4.

[0046] The modification information in (B) is an index that modifies an index that indicates a causal relationship determined based on the relationship between a pair of functions (a pair of two functions with a direction). The modified index is, for example, a directional score between functions (modified score).

[0047] In the modification information of (B), as shown in modification decision information 42 in FIG. 4, modification scores (Modify1 to 9) are associated as modification information with each of the nine sets of functions.

[0048] The modified information in (C) is binary information that indicates whether or not there is a causal relationship, based on the relationship between a pair of functions (a pair of two similar functions).

[0049] As shown in the modification decision information 43 in Fig. 4, the modification information of (C) is set to "0" on the assumption that there is no causal relationship between components having similar functions. Also, the modification information is set to "1" on the assumption that there is a causal relationship between components having dissimilar functions.

[0050] The correction information in (D) is an index (corrected score) that corrects an index representing a causal relationship that is determined based on the relationship between a pair of functions (a pair of two similar functions).

[0051] As shown in modification decision information 44 in Fig. 4, the modification information of (D) assumes that there is no causal relationship between components having similar functions and sets modification scores (Modify10 to 12) as modification information. Also, assumes that there is a causal relationship between components having dissimilar functions and sets a modification score (Modify13) as modification information.

[0052] The relationship between a set of functions and modification information will now be described. In the target system, if it is known that causal relationships are unlikely to occur between pairs of functions such as MV → SV, MV → MV, and PV → PV, information is set to break the causal relationships between the components corresponding to these pairs of functions. That is, as shown in the modification decision information 41 in Fig. 4, "0" is set, which indicates that there is no causal relationship.

[0053] Furthermore, when it is known that a causal relationship is unlikely to occur, information for modifying the degree of the causal relationship between the components corresponding to the pair of functions (modification score), that is, as shown in modification decision information 42 in Fig. 4, the modification information is set to information for modifying the degree of the causal relationship (Modify 1 to 9). Specifically, a modification score to be subtracted from the causal score is set. For example, the value to be added to the causal score (modification score value) is set to -0.5, etc.

[0054] In addition, if it is known that causal relationships are unlikely to occur in pairs of functions such as α → α, β → β, and γ → γ in the target system, information is set to break the causal relationships between the components corresponding to these pairs of functions. That is, as shown in the modification decision information 43 in Fig. 4, "0" is set, which indicates that there is no causal relationship.

[0055] Furthermore, when it is known that causal relationships are unlikely to occur in pairs of functions such as α → α, β → β, and γ → γ in the target system, information (modification score) for modifying the degree of causal relationships between components corresponding to these pairs of functions, that is, modification information as shown in modification decision information 44 in FIG. 4, is set as information for modifying the degree of causal relationships (Modify 10 to 13). Specifically, a modification score to be subtracted from the causal score is set. For example, the modification score to be subtracted from the causal score between components with the same function is set to -0.5 if the functions are the same, and 0 if they are different.

[0056] The acquisition of correction information will now be described. For example, in the case of the above-mentioned pair (X → Y), in the component information 21 of Figure 2, the label associated with the identifier X is SV and the label associated with the identifier Y is MV, so in the case of (A), the modification information "1" associated with (SV → MV) is obtained by referring to the modification decision information 41 of Figure 4.

[0057] In the case of (B), the modification information "Modify2" associated with (SV→MV) is obtained by referring to the modification decision information 42 in Fig. 4. Note that modification information is also obtained for the groups (Y→X) (X→Z) (Z→X) (Y→Z) (Z→Y) in a similar manner.

[0058] Also, in the case of the above-mentioned pair (X → Y), since in the component information 22 of FIG. 2, the label associated with the identifier X is α and the label associated with the identifier Y is β, in the case of (C), the modification decision information 43 of FIG. 4 is referenced to obtain the modification information “1” corresponding to (α → β).

[0059] In the case of (D), the modification decision information 44 in FIG. 4 is referenced to obtain the modification information "Modify13" corresponding to (α→β).

[0060] Next, the causal information correcting unit 12 corrects the causal information using the obtained correction information. The causal information (1) (causal information 31 in FIG. 3) is corrected using the correction information (A) (correction information 41 in FIG. 4). In this case, if the causal information is "0", there is no causal relationship between the two components of the target pair regardless of the correction information, so the causal information is set to "0". Also, if the causal information is "1" and the correction information is "0", it is determined that there is no causal relationship between the two components of the target pair, and the causal information is corrected to "0".

[0061] Alternatively, the causal information (2) (causal information 32 in Figure 3) is corrected using the correction information (A) (correction information 41 in Figure 4). If the correction information is "0", regardless of the causal score represented by the causal information, there is no causal relationship between the two components of the target pair, so the causal score is lowered. If the correction information is "1", the causal score represented by the causal information is not corrected. After that, if the causal score is equal to or greater than a preset threshold, it is determined that there is a causal relationship between the two components of the target pair.

[0062] Alternatively, the causal information (2) (causal information 32 in FIG. 3) is corrected using the correction information (B) (correction information 42 in FIG. 4). In this case, the causal score is corrected based on the causal score indicated by the causal information and the correction score indicated by the correction information. After that, if the causal score is equal to or greater than a preset threshold, it is determined that there is a causal relationship between the two components of the target pair.

[0063] Alternatively, the causal information (2) (causal information 32 in FIG. 3) is corrected using the correction information (C) (correction information 43 in FIG. 4). In this case, if the correction information is "0", regardless of the causal score represented by the causal information, there is no causal relationship between the two components of the target pair, so the causal score is lowered. If the correction information is "1", the causal score represented by the causal information is not corrected. After that, if the causal score is equal to or greater than a preset threshold, it is determined that there is a causal relationship between the two components of the target pair.

[0064] Alternatively, the causal information (2) (causal information 32 in FIG. 3) is corrected using the correction information (D) (correction information 44 in FIG. 4). In this case, the causal score is corrected based on the causal score indicated by the causal information and the correction score indicated by the correction information. After that, if the causal score is equal to or greater than a preset threshold, it is determined that there is a causal relationship between the two components of the target pair.

[0065] [System Configuration] The configuration of the causal exploration device 10 in the first embodiment will be described more specifically with reference to Fig. 5. Fig. 5 is a diagram for explaining an example of a system having a causal exploration device.

[0066] 5 includes a causal exploration device 10, a storage device 20, and an inference device 30. The causal exploration device 10 includes a causal information calculation unit 11, a causal information correction unit 12, and a causal graph generation unit 13.

[0067] The storage device 20 stores information such as setting information, component information, modification decision information, and causal graphs. Although the storage device 20 is provided outside the causal exploration device 10 in Fig. 5, it may be provided inside the causal exploration device 10. Also, a plurality of storage devices 20 may be provided, and the above-mentioned information may be divided and stored in each of them. The storage device 20 may be, for example, a storage device such as a database or a server device.

[0068] The setting information is information required when generating the causal information. The component information and the modification decision information have already been explained, so the explanation will be omitted. The causal graph is a directed graph that represents the causal relationships between components.

[0069] The inference device 30 is a device that executes an application that analyzes input data using a causal graph. For example, the inference device 30 is a device that, by specifying an event to be analyzed in the input data, traces the variables that influenced the event based on the causal graph and infers the cause of the event.

[0070] The inference device 30 may also be a device that performs a quantitative analysis (such as estimating the extent to which a variable influenced an event) by executing causal effect inference in advance and accumulating the inference results. The inference device 30 may also be a device that estimates the cause of an event and the time when the event occurred with a degree of certainty using a Bayesian model or the like.

[0071] The inference device 30 may also be a device that inputs abnormality data and analyzes the cause of the abnormality. In this case, in addition to estimating the variables and time that cause a specific abnormal event, the device can explain the history of the degree of abnormality (anomaly score) before the abnormal event using a Bayesian model or the like. For example, as a result of an abnormality being transmitted from component X to component Y at time t, i.e., as a result of the anomaly being transmitted through an edge of the causal graph, the degree of abnormality related to component Y has increased.

[0072] Furthermore, the inference device 30 may be a device that predicts and classifies events that were not included in the learning data by learning a prediction model according to causal effect inference and causal relationships. For example, this can be achieved by using linear regression, support vector machines, neural networks, branch pruning according to causal graphs, etc. By making predictions and classifications based on causal relationships that eliminate spurious correlations, it is possible to avoid irrational predictions and classifications in which a change in the state of a device affects the state of another device that is not causally related to the device in question.

[0073] The causal exploration device will now be described. The causal information calculation unit 11 and the causal information correction unit 12 have already been described, so their description will be omitted.

[0074] The causal graph generating unit 13 generates a causal graph by using the set of components and the corrected causal information corresponding to the set, and stores the generated causal graph in the storage device 20.

[0075] [Device operation] Next, the operation of the causal exploration device in the first embodiment will be described with reference to FIG. 6. FIG. 6 is a diagram for explaining an example of the operation of the causal exploration device. In the following description, the diagram will be referred to as appropriate. In addition, in the first embodiment, the causal exploration method is implemented by operating the causal exploration device. Therefore, the description of the causal exploration method in the first embodiment will be replaced with the following description of the operation of the causal exploration device.

[0076] The causal information calculation unit 11 first selects two different components (a set) from a plurality of components provided in a target system (step A1). Specifically, in step A1, the causal information calculation unit 11 selects two components in a permutation from among the plurality of components based on component information stored in a storage device (not shown).

[0077] Next, the causal information calculation unit 11 calculates causal information representing the causal relationship between the selected two components (pair) (step A2). Specifically, in step A2, the causal information calculation unit 11 inputs the time series data of the two components for each pair into a causal model for generating causal information representing the causal relationship between the components, and calculates the causal information. The causal information may be, for example, the above-mentioned (1) and (2).

[0078] Next, the causal information correction unit 12 corrects the causal information based on the function information indicating the functions associated with each of the two selected components (step A3).

[0079] Specifically, in step A3, the causal information corrector 12 first refers to the component information for each pair using the identifiers of the two components, and acquires the function information associated with each of the two components.

[0080] Next, in step A3, the causal information correction unit 12 uses the functions included in the two pieces of acquired functional information to refer to the correction decision information stored in advance in the storage device and acquires correction information for correcting the causal information. The correction information may be, for example, the above-mentioned (A), (B), or (C).

[0081] Next, in step A3, the causal information corrector 12 corrects the causal information using the acquired correction information.

[0082] Next, the causal graph generating unit 13 generates a causal graph using the set of components and the corrected causal information corresponding to the set, and stores the generated causal graph in the storage device 20 (step A4).

[0083] [Effects of the embodiment] As described above, according to the embodiment, the causal relationship between two components can be accurately estimated by correcting the causal information indicating the causal relationship between the two components based on the functions associated with the respective components.

[0084] For example, since systems such as IoT / OT are physical systems, there is a lot of sensor data that behaves in a roughly synchronized manner with a delay, making it difficult to eliminate false correlations. However, by using the embodiment, it is possible to eliminate false correlations to a certain extent, and therefore it is possible to accurately estimate causal relationships.

[0085] [program] The program in the embodiment may be a program that causes a computer to execute steps A1 to A4 shown in Fig. 6. By installing and executing this program in a computer, the causal exploration device and the causal exploration method in the embodiment can be realized. In this case, the processor of the computer functions as a causal information calculation unit 11, a causal information correction unit 12, and a causal graph generation unit 13, and performs the processing.

[0086] The program in the embodiment may be executed by a computer system constructed by a plurality of computers. In this case, for example, each computer may function as any one of the causal information calculation unit 11, the causal information correction unit 12, and the causal graph generation unit 13.

[0087] (Variation 1) The configuration of a causal exploration device in the first modification will be described with reference to Fig. 7. Fig. 7 is a diagram for explaining an example of a causal exploration device in the first modification.

[0088] [Device configuration] The causal exploration device 70 is a device that accurately estimates a causal relationship between two components provided in a target system based on the functions of each of the two components. In addition, in the first modification, even if functional information (label) is not associated with a part of an identifier in the component information 21 or 22 shown in FIG. 2, the label can be estimated.

[0089] 7, the causal exploration device 70 has a causal information calculation unit 11, a causal information correction unit 12, and a function estimation unit 71. Note that the causal information calculation unit 11 and the causal information correction unit 12 have already been described, and therefore their description will be omitted.

[0090] The function estimation unit 71 inputs the time series data acquired from a component with no associated functional information into a model that estimates functional information learned using functional information and time series data acquired from a component with which the functional information is associated, and estimates functional information of the component with no associated functional information.

[0091] Specifically, the function estimation unit 71 learns using labels (labels such as SV, MV, PV that represent functions, or labels such as α, β, γ that represent only functional similarity) and time-series data acquired from components associated with those labels, and generates a label estimation model that estimates labels.

[0092] Labels that indicate functional similarity are information that allows one to determine whether the functions of each component are identical or not by assigning the same label to identical or similar functions, although the specific function that each label represents is unknown.

[0093] The label estimation model may be, for example, a decision tree, a linear regression, a support vector machine, a neural network, clustering such as k-means, a Bayesian model, etc. Distance learning may also be used in combination.

[0094] The function estimation unit 71 inputs time-series data acquired from a component with no associated label to a label estimation model to estimate the label of the component with no associated label. After that, the function estimation unit 71 associates the estimated label with the identifier of the component with no associated label in the component information.

[0095] In addition, in the first modification, the causal information correction unit 12 may correct the causal information (causal score) by using the judgment amount used by the function estimation unit 71 when estimating the label.

[0096] Alternatively, when the function estimation unit 71 estimates the label, it is assumed that the judgment amounts are SV: 0.7, MV: 0.2, and PV: 0.1. In this case, the function estimation unit 71 estimates the SV with the highest value among the judgment amounts as the label.

[0097] Therefore, the causal score is modified using the judgment amounts of the two components. For example, consider a case where it is difficult to reflect the causal relationships MV → SV and SV → SV. If the judgment amounts of one component are SV: 0.7, MV: 0.2, PV: 0.1, and the judgment amounts of the other component are SV: 0.4, MV: 0.3, PV: 0.3, then the modified score of -0.2 x 0.4 - 0.7 x 0.4 = -0.36 may be added to the causal score for modification.

[0098] [Effects of Modification 1] As described above, according to the first modification, even if functional information (label) is not associated with a part of an identifier, the label can be estimated. Therefore, by correcting the causal information representing the causal relationship between two components based on the functions associated with each of the components, the causal relationship can be estimated with high accuracy.

[0099] (Variation 2) The configuration of a causal exploration device in the second modification will be described with reference to Fig. 8. Fig. 8 is a diagram for explaining an example of a causal exploration device in the second modification.

[0100] [Device configuration] The causal exploration device 80 is a device that accurately estimates the causal relationship between two components provided in a target system. In addition, in the second modification, even if, for example, function information (labels) is not associated with all identifiers in the component information 21, 22 shown in Fig. 2, modification information can be estimated. Since no labels are associated with all identifiers in the second modification, a high degree of similarity indicates similar functions.

[0101] 8, the causal search device 80 has a causal information calculation unit 11, a correction information estimation unit 81, and a causal information correction unit 82. Note that the causal information calculation unit 11 has already been described, so a description thereof will be omitted.

[0102] The correction information estimation unit 81 estimates a similarity using time-series data acquired from two components whose function information is not associated with each other, and estimates correction information based on the estimated similarity.

[0103] Specifically, the modification information estimation unit 81 first acquires time series data from each of the two components. Next, the modification information estimation unit 81 estimates the similarity of the time series data acquired from the two selected components using, for example, the Kullback-Leibler distance, the Jensen-Shannon distance, the f-divergence, the Hellinger distance, the Wasserstein distance, etc. Next, the modification information estimation unit 81 estimates modification information based on the estimated similarity.

[0104] The causal information correction unit 82 first obtains the causal information (causal score) calculated by the causal information calculation unit 11 and the correction information (correction score) estimated by the correction information estimation unit 81. Next, the causal information correction unit 82 corrects the causal information (causal score) by using the correction information (correction score).

[0105] For example, if the similarity of the time series data of two components X and Y is 0.3, adding -0.3 to the causal score for X → Y and the causal score for Y → X indicates that a causal relationship is unlikely to occur between variables with similar functions.

[0106] [Effects of Modification 2] As described above, according to the second modification, even if function information (labels) are not associated with all identifiers, if the similarity is high, the labels can be estimated as the functions are similar. Therefore, by correcting the causal information representing the causal relationship between two components based on the functions associated with each component, the causal relationship can be estimated with high accuracy.

[0107] [Physical configuration] Here, a computer that realizes the causal exploration device by executing the programs in the embodiment and the first and second modifications will be described with reference to Fig. 9. Fig. 9 is a diagram for explaining an example of a computer that realizes the causal exploration device in the embodiment and the first and second modifications.

[0108] 9, the computer 110 includes a CPU (Central Processing Unit) 111, a main memory 112, a storage device 113, an input interface 114, a display controller 115, a data reader / writer 116, and a communication interface 117. These components are connected to each other via a bus 121 so as to be able to communicate data with each other. Note that the computer 110 may include a GPU or an FPGA in addition to the CPU 111 or instead of the CPU 111.

[0109] The CPU 111 loads the programs (codes) in the embodiment stored in the storage device 113 into the main memory 112 and executes them in a predetermined order to perform various calculations. The main memory 112 is typically a volatile storage device such as a DRAM (Dynamic Random Access Memory). The programs in the embodiment are provided in a state stored in a computer-readable recording medium 120. The programs in the embodiment may be distributed on the Internet connected via the communication interface 117. The recording medium 120 is a non-volatile recording medium.

[0110] Specific examples of the storage device 113 include a hard disk drive and a semiconductor storage device such as a flash memory. The input interface 114 mediates data transmission between the CPU 111 and input devices 118 such as a keyboard and a mouse. The display controller 115 is connected to a display device 119 and controls the display on the display device 119.

[0111] Data reader / writer 116 mediates data transmission between CPU 111 and recording medium 120, reads programs from recording medium 120, and writes processing results in computer 110 to recording medium 120. Communication interface 117 mediates data transmission between CPU 111 and other computers.

[0112] Specific examples of the recording medium 120 include general-purpose semiconductor storage devices such as CF (Compact Flash (registered trademark)) and SD (Secure Digital), magnetic recording media such as a flexible disk, or optical recording media such as a CD-ROM (Compact Disk Read Only Memory).

[0113] The causal exploration device in the embodiment and modifications 1 and 2 can be realized by using hardware corresponding to each part, rather than a computer with a program installed. Furthermore, the causal exploration device may be realized in part by a program and the remaining part by hardware.

[0114] [Note] The following supplementary notes are further disclosed regarding the above-described embodiments. A part or all of the above-described embodiments can be expressed by (Supplementary Note 1) to (Supplementary Note 15) described below, but are not limited to the following descriptions.

[0115] (Appendix 1) a causal information calculation unit that selects two different components from a plurality of components provided in a target system and calculates causal information that indicates a causal relationship between the two selected components; a causal information correcting unit that corrects the causal information based on function information representing functions associated with each of the two selected components; A causal exploration device having the above structure.

[0116] (Appendix 2) A causal exploration device according to claim 1, The causal information correction unit determines that there is no causal relationship between the two selected components when the function information of each of the two selected components indicates the same function. Causal search device.

[0117] (Appendix 3) A causal exploration device according to claim 1, The causal information correction unit determines correction information used to constrain spurious correlation based on the functional information, and corrects the causal information based on the correction information. Causal search device.

[0118] (Appendix 4) A causal exploration device according to any one of claims 1 to 3, a function estimation unit that inputs time-series data acquired from a component with which the functional information is not associated into a model that estimates the functional information and that is trained using the functional information and time-series data acquired from the component with which the functional information is associated, and estimates a function of the component with which the functional information is not associated; Causal search device.

[0119] (Appendix 5) A causal exploration device according to claim 3, a correction information estimation unit that estimates a similarity using time-series data acquired from the two components whose functional information is not associated with each other, and estimates the correction information based on the estimated similarity; Causal search device.

[0120] (Appendix 6) a causal information calculation step of selecting two different components from a plurality of components provided in the target system and calculating causal information representing a causal relationship between the two selected components; a causal information modifying step of modifying the causal information based on functional information representing functions associated with each of the two selected components; A causal search method having the following structure.

[0121] (Appendix 7) A causal exploration method according to claim 6, The causal information correction step determines that there is no causal relationship between the two selected components when the function information of each of the two selected components indicates the same function. Causal search method.

[0122] (Appendix 8) A causal exploration method according to claim 6, The causal information modifying step includes determining modification information used to constrain spurious correlation based on the functional information, and modifying the causal information based on the modification information. Causal search method.

[0123] (Appendix 9) A causal exploration method according to any one of Supplementary Notes 6 to 8, A function estimation step of inputting time series data acquired from a component to which the functional information is not associated into a model for estimating the functional information, the model having been trained using the functional information and time series data acquired from the component to which the functional information is associated, and estimating a function of the component to which the functional information is not associated. Causal search method.

[0124] (Appendix 10) A causal exploration method according to claim 8, comprising: A correction information estimation step of estimating a similarity using time-series data acquired from the two components whose functional information is not associated with each other, and estimating the correction information based on the estimated similarity. Causal search method.

[0125] (Appendix 11) On the computer, a causal information calculation step of selecting two different components from a plurality of components provided in the target system and calculating causal information representing a causal relationship between the two selected components; a causal information modifying step of modifying the causal information based on function information representing functions associated with each of the two selected components; A program that contains instructions to execute Hmm.

[0126] (Appendix 12) As described in Appendix 11 program And, The causal information correction step determines that there is no causal relationship between the two selected components when the function information of each of the two selected components indicates the same function. program .

[0127] (Appendix 13) As described in Appendix 11 program And, The causal information modifying step includes determining modification information used to constrain spurious correlation based on the functional information, and modifying the causal information based on the modification information. program .

[0128] (Appendix 14) Any one of appendix 11 to 13 program And, The program causes the computer to a function estimation step of inputting time series data acquired from a component to which the functional information is not associated into a model for estimating the functional information, the model having been trained using the functional information and time series data acquired from the component to which the functional information is associated, and estimating a function of the component to which the functional information is not associated; Contains instructions to execute program .

[0129] (Appendix 15) As described in Appendix 13 program And, The program causes the computer to a correction information estimation step of estimating a similarity using time series data acquired from the two components whose functional information is not associated with each other, and estimating the correction information based on the estimated similarity; Contains instructions to execute program .

[0130] Although the present invention has been described above with reference to the embodiment, the present invention is not limited to the above embodiment. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention. [Industrial Applicability]

[0131] INDUSTRIAL APPLICABILITY As described above, according to the present invention, causal relationships can be estimated with high accuracy, and the present invention is useful in fields in which causal relationships are used. [Explanation of symbols]

[0132] 10, 70, 80 Causal search device 11 Causal information calculation unit 12 Causal information correction section 13 Causal graph generation unit 20 Storage device 30 Reasoning device 71 Function Prediction Section 81 Modified information estimation section 82 Causal information correction part 110 Computer 111 CPU 112 Main memory 113 Storage device 114 Input Interface 115 Display Controller 116 Data Reader / Writer 117 Communication Interface 118 Input Devices 119 Display device 120 Recording media 121 Bus

Claims

1. a causal information calculation means for selecting two different components from a plurality of components provided in a target system and calculating causal information representing a causal relationship between the two selected components; a causal information correcting means for correcting the causal information based on functional information representing functions associated with each of the two selected components; A causal exploration device having the above structure.

2. The causal exploration device according to claim 1 , The causal information correction means determines that there is no causal relationship between the two selected components when the function information of each of the two selected components indicates the same function. Causal search device.

3. The causal exploration device according to claim 1 , The causal information correction means determines correction information used to constrain spurious correlation based on the functional information, and corrects the causal information based on the correction information. Causal search device.

4. 4. A causal exploration device according to claim 1, a function estimation means for inputting time series data acquired from a component to which the functional information is not associated into a model for estimating the functional information, the model having been trained using the functional information and time series data acquired from the component to which the functional information is associated, and estimating a function of the component to which the functional information is not associated; Causal search device.

5. The causal exploration device according to claim 3, a correction information estimation means for estimating a similarity using time-series data acquired from the two components whose functional information is not associated with each other, and estimating the correction information based on the estimated similarity; Causal search device.

6. A computer comprising: Selecting two different components from a plurality of components provided in the target system, and calculating causal information representing a causal relationship between the two selected components; modifying the causal information based on functional information representative of functions associated with each of the two selected components; Causal search method.

7. A causal exploration method according to claim 6, comprising: If the function information of each of the two selected components indicates the same function, the computer determines that there is no causal relationship between the two selected components. Causal search method.

8. A causal exploration method according to claim 6, comprising: The computer determines modification information used to constrain spurious correlations based on the functional information, and modifies the causal information based on the modification information. Causal search method.

9. A causal exploration method according to any one of claims 6 to 8, comprising: The computer inputs the time series data acquired from the component to which the functional information is not associated into a model trained using the time series data acquired from the component to which the functional information is associated, and estimates a function of the component to which the functional information is not associated. Causal search method.

10. A causal exploration method according to claim 8, comprising: The computer estimates a similarity using time-series data acquired from the two components whose functional information is not associated with each other, and estimates the correction information based on the estimated similarity. Causal search method.

11. On the computer, selecting two different components from a plurality of components provided in the target system, and calculating causal information representing a causal relationship between the two selected components; modifying the causal information based on functional information representative of functions associated with each of the two selected components; A program containing instructions.

12. The program according to claim 11, If the function information of each of the two selected components indicates the same function, it is determined that there is no causal relationship between the two selected components. program.

13. The program according to claim 11, The program causes the computer to determining modification information used to constrain spurious correlations based on said functional information; and modifying said causal information based on said modification information. program.

14. 14. The program according to claim 11, The program causes the computer to inputting the time series data acquired from the component to which the functional information is not associated into a model trained using the time series data acquired from the component to which the functional information is associated, thereby estimating the function of the component to which the functional information is not associated; A program containing instructions.

15. The program according to claim 13, The program causes the computer to a similarity is estimated using time-series data acquired from the two components whose functional information is not associated with each other, and the correction information is estimated based on the estimated similarity; A program containing instructions.

Images