Method, system and computer program product for determining causality

By constructing a causal relationship determination method containing expert knowledge constraints, the problem of excessive requirements for expert knowledge in the causal relationship determination process in the existing technology is solved, and more accurate and efficient causal relationship determination is achieved.

CN109598347BActive Publication Date: 2024-12-06NEC CORP
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN201710922464.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2017-09-30
Publication Date
2024-12-06
Estimated Expiration
2037-09-30

AI Technical Summary

Technical Problem

When determining the causal relationship between multiple variables, the prior art relies on expert knowledge to be too demanding and it is difficult to obtain expert knowledge that meets the demanding requirements, resulting in the process of determining the causal relationship complex and inefficient.

Method used

By constructing a problem formula containing expert knowledge constraints, using the fit degree and expert knowledge constraints in the data set and matrix, constraints are given for edges and paths in the directed graph respectively to determine the causal relationship more accurately.

Benefits of technology

It realizes that without the need for complete expert knowledge, it can simply and effectively assist in determining the causal relationship between multiple variables, reduces the requirements for perfection of expert knowledge, avoids interaction with experts, and improves the accuracy of causal relationship determination. and efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN109598347B_ABST
    Figure CN109598347B_ABST
Patent Text Reader

Abstract

The implementation of the present disclosure relates to a method, system and computer program product for determining a causal relationship. A method for determining a causal relationship between multiple variables is provided, comprising: in response to collecting a data set of multiple samples associated with multiple variables, obtaining a matrix describing the causal relationship between the multiple variables, each of the multiple samples including data corresponding to the multiple variables; based on the data set and the matrix, determining the fit and expert knowledge constraints associated with the causal relationship, wherein the expert knowledge constraints include at least one of the edge constraints for the direct causal relationship between two variables in the matrix and the path constraints for the indirect causal relationship between two variables in the matrix; constructing a problem formula describing the causal relationship according to the determined fit and expert knowledge constraints; and solving the constructed problem formula to obtain a candidate result of the matrix. Further, a corresponding system and a computer program product are provided.
Need to check novelty before this filing date? Find Prior Art

Description

Background Art

[0002] The probability model is a graphical network model obtained based on probability reasoning. Here, probability reasoning refers to obtaining the correlation between these variables by analyzing the collected information corresponding to multiple variables. Bayesian network is a probability model proposed to solve the problems of uncertainty and incompleteness. It has been widely used in many fields.

[0003] The Bayesian network can use a directed acyclic graph (DAG) to describe the causal relationship between multiple variables. The DAG can include nodes representing variables and directed edges and paths representing the causal relationship between the variables. For example, a directed edge from a parent node to its child node can indicate that there is a direct causal relationship between the variable represented by the parent node and the variable represented by the child node. For another example, a path from one node to another node can indicate that there is an indirect causal relationship between the variables represented by the two nodes. The Bayesian network is suitable for expressing and analyzing events with uncertainty and probability, and can be determined from the incomplete, imprecise or uncertain information collected corresponding to multiple variables.

[0004] At present, technical solutions have been developed to determine causal relationships based on collected data sets and expert knowledge. However, these technical solutions have too strict requirements on expert knowledge, or may involve interaction with experts in the process of determining causal relationships. Therefore, how to determine the causal relationship between multiple variables in a simple and effective way based on expert knowledge (which may not be perfect) has become a research hotspot. Summary of the invention

[0005] Expert knowledge can improve the accuracy of the causal relationship between multiple variables determined based on a data set to a certain extent. Therefore, it is expected to develop and implement a technical solution that can determine the causal relationship based on expert knowledge in a more accurate and effective way. It is also expected that the technical solution can use expert knowledge to guide the learning process of causal relationships as much as possible.

[0006] According to a first aspect of the present invention, a method for determining a causal relationship between multiple variables is provided. The method includes: in response to collecting a data set of multiple samples associated with multiple variables, obtaining a matrix describing the causal relationship between the multiple variables, each sample in the multiple samples includes data corresponding to the multiple variables; based on the data set and the matrix, determining the fitness and expert knowledge constraints associated with the causal relationship, wherein the expert knowledge constraints include at least one of the edge constraints for the direct causal relationship between two variables in the matrix and the path constraints for the indirect causal relationship between two variables in the matrix; constructing a problem formula describing the causal relationship according to the determined fitness and expert knowledge constraints; and solving the constructed problem formula to obtain a candidate result of the matrix.

[0007] According to a second aspect of the present invention, a system for determining a causal relationship between multiple variables is provided, comprising: one or more processors; a memory coupled to at least one of the one or more processors; computer program instructions stored in the memory, when the computer program instructions are executed by at least one processor, the system executes a method for determining a causal relationship between multiple variables. The method comprises: in response to a data set of multiple samples associated with multiple variables, obtaining a matrix describing the causal relationship between multiple variables, each of the multiple samples including data corresponding to multiple variables; based on the data set and the matrix, determining the fit and expert knowledge constraints associated with the causal relationship, wherein the expert knowledge constraints include at least one of the edge constraints for the direct causal relationship between two variables in the matrix and the path constraints for the indirect causal relationship between two variables in the matrix; constructing a problem formula describing the causal relationship according to the determined fit and expert knowledge constraints; and solving the constructed problem formula to obtain a candidate result of the matrix.

[0008] According to a third aspect of the present invention, a device for determining a causal relationship between multiple variables is provided. The device comprises: an acquisition module configured to acquire a matrix describing the causal relationship between multiple variables in response to a data set of multiple samples associated with multiple variables, wherein each sample in the multiple samples comprises data corresponding to multiple variables; a determination module configured to determine the degree of fit and expert knowledge constraints associated with the causal relationship based on the data set and the matrix, wherein the expert knowledge constraints include at least one of the edge constraints for the direct causal relationship between two variables in the matrix and the path constraints for the indirect causal relationship between two variables in the matrix; a construction module configured to construct a problem formula describing the causal relationship according to the determined degree of fit and expert knowledge constraints; and a solution module configured to solve the constructed problem formula to obtain a candidate result of the matrix.

[0009] According to a fourth aspect of the present invention, there is provided a computer program product comprising a computer-readable storage medium having a computer-readable program stored therein, wherein when the computer-readable program is executed on a computing device, the computing device executes the method for determining a causal relationship between multiple variables as described above.

[0010] By adopting the technical solution for determining the causal relationship between multiple variables of the present invention, corresponding constraints can be given for the edges and paths in the directed graph based on expert knowledge, thereby determining the causal relationship more accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The features, advantages and other aspects of various implementations of the present invention will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings, which illustrate several implementations of the present invention in an exemplary and non-limiting manner. In the accompanying drawings:

[0012] Figure 1 A block diagram schematically illustrates an exemplary computing system suitable for implementing the present invention;

[0013] Figure 2 A block diagram for determining a causal relationship between multiple variables according to an implementation of the present disclosure is schematically shown;

[0014] Figure 3 A flowchart schematically illustrates a method for determining a causal relationship between multiple variables according to an implementation of the present disclosure;

[0015] Figure 4 Schematically illustrates an example of a directed graph associated with a matrix according to one implementation of the present disclosure;

[0016] Figure 5 A block diagram schematically illustrates a solution for determining expert knowledge constraints according to an implementation of the present disclosure;

[0017] Figure 6 A block diagram schematically shows a method for determining an edge knowledge matrix and elements in the path knowledge matrix based on an edge knowledge matrix and a path knowledge matrix according to an implementation of the present disclosure;

[0018] Figure 7 A block diagram schematically illustrates a method for determining a score for reaching a target state from a state according to an implementation of the present disclosure; and

[0019] Figure 8 A block diagram schematically shows a device for determining a causal relationship between multiple variables according to an implementation of the present disclosure. DETAILED DESCRIPTION

[0020] The preferred implementation of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the preferred implementation of the present disclosure is shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the implementation described herein. On the contrary, these implementations are provided to make the present disclosure more thorough and complete, and to fully convey the scope of the present disclosure to those skilled in the art.

[0021] Figure 1 1 is a block diagram of an exemplary computing system 100 suitable for implementing the disclosed implementations. Figure 1 As shown, the computer system 100 may include: a CPU (central processing unit) 101, a RAM (random access memory) 102, a ROM (read only memory) 103, a system bus 104, a hard disk controller 105, a keyboard controller 106, a serial interface controller 107, a parallel interface controller 108, a display controller 109, a hard disk 110, a keyboard 111, a serial external device 112, a parallel external device 113, and a touch screen display 114. Among these devices, the CPU 101, the RAM 102, the ROM 103, the hard disk controller 105, the keyboard controller 106, the serial controller 107, the parallel controller 108, and the display controller 109 are coupled to the system bus 104. The hard disk 110 is coupled to the hard disk controller 105, the keyboard 111 is coupled to the keyboard controller 106, the serial external device 112 is coupled to the serial interface controller 107, the parallel external device 113 is coupled to the parallel interface controller 108, and the touch screen display 114 is coupled to the display controller 109. It should be understood that Figure 1 The structural block diagram shown is only for the purpose of example, and is not intended to limit the scope of the present disclosure. In some cases, some devices may be added or reduced according to specific circumstances.

[0022] Those skilled in the art will appreciate that the present disclosure may be implemented as a system, method, or computer program product. Therefore, the present disclosure may be specifically implemented in the following forms, namely: it may be complete hardware, it may be complete software (including firmware, resident software, microcode, etc.), or it may be a combination of hardware and software, generally referred to herein as a "circuit", "module" or "system". In addition, in some implementations, the present disclosure may also be implemented in the form of a computer program product in one or more computer-readable media, which may contain computer-readable program code.

[0023] Any combination of one or more computer-readable media may be used. A computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, device, or device.

[0024] Computer-readable signal media may include a data signal propagated in baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including, but not limited to, electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0025] Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0026] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect through the Internet).

[0027] The present disclosure will be described below with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products of the implementation methods of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of boxes in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer or other programmable data processing device to produce a machine, and these computer program instructions are executed by a computer or other programmable data processing device to produce a device that implements the functions / operations specified in the boxes in the flowchart and / or block diagram.

[0028] These computer program instructions may also be stored in a computer-readable medium that enables a computer or other programmable data processing device to operate in a specific manner, such that the instructions stored in the computer-readable medium produce a manufacture including an instruction means (instruction means) that implements the functions / operations specified in the blocks in the flowchart and / or block diagram.

[0029] Computer program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operational steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby enabling the instructions executed on the computer or other programmable device to provide a process for implementing the functions / operations specified in the blocks in the flowchart and / or block diagram.

[0030] For the convenience of the following description, an example of a specific application environment of the present disclosure is first introduced. Causal analysis based on Bayesian networks can be applied to a variety of application environments. For example, in an application environment for monitoring whether a control system in a specific area has an abnormality, information corresponding to multiple variables (for example, temperature, humidity, ..., and whether the control system has an abnormality, etc.) can be collected separately (for example, information collected at different time points). The number of variables p is not limited here, but in a simple application environment, the number of variables can be several, and in a complex application environment, the number of variables p can reach dozens or even more.

[0031] The information corresponding to each variable collected at one time point can be stored in a sample (including a vector of p dimensions), and the information collected at n time points can be stored in n samples (the n samples can be referred to as the collected data set). Then, the collected data set can be used as input to determine the causal relationship between the temperature, humidity, etc. at various locations in the area and whether there is an abnormality in the control system.

[0032] For the convenience of the following description, in the context of the present disclosure, only the specific example of determining whether there is an abnormality in the control system will be used to introduce how to determine the causal relationship. According to other implementations of the present disclosure, the technical solution according to the present disclosure can also be adopted in more application environments. For example, in an application environment that determines the causal relationship between the sales volume of a specific commodity (e.g., beer) and various other variables (information such as the price, temperature, time, country, etc. of beer), a data set including sales volume and various other variables can be collected separately, and the causal relationship between the beer sales volume and various other variables can be determined based on the data set. For another example, in an application environment that determines the causal relationship between automobile insurance premiums and various variables (e.g., vehicle brand, model, number of airbags, gender, age of the insured, etc.), a data set including insurance premiums and various other variables can be collected separately, and the causal relationship between insurance premiums and various other variables can be determined based on the data set.

[0033] In the context of the present disclosure, the specific details of the present disclosure will be described using a Bayesian network as a specific example of causal relationships. Here, a Bayesian network is a graphical probabilistic network model defined based on a directed graph. The directed graph can be represented in a matrix manner. Specifically, it is assumed that in an application environment for determining whether a control system is abnormal, there are the following p variables: temperature, humidity, ..., whether the control system is abnormal. Then the data set including n samples can be represented in the form shown in Table 1 below.

[0034] Table 1 Examples of datasets

[0035]

[0036] As shown in Table 1 above, the first column "variable x1 = temperature" means that the first variable among the p variables is "temperature", that is, the temperature value measured at different time points. The second column "variable x2 = humidity" means that the second variable among the p variables is "humidity", that is, the humidity value measured at different time points. The last column (pth column) "variable x p = "existence of abnormality" means that the pth variable among the p variables is "existence of abnormality", that is, whether the control system has abnormality at different time points. The matrix B as shown below can be used to represent the causal relationship between the above p variables.

[0037]

[0038] For example, where matrix B is a p-order matrix including p×p elements, each element indicates whether there is a causal relationship between the two variables corresponding to the position of the element. Specifically, the variable β in matrix B jiIt indicates the causal relationship between the jth variable and the ith variable among the p variables. It should be noted that the different positions of the two variables indicate different causal relationships. Therefore, β ji and β ij They represent different causal relationships. In other words, the directions of the edges in the directed graph represented by the matrix B are different. In addition, the diagonal part in the matrix B represents the causal relationship between each element and itself. Since there is no causal relationship between a specific element and itself, the value of the element at the diagonal part should be set to 0.

[0039] Based on the above description, it can be seen that in a Bayesian network, the problem of determining the causal relationship between p variables based on the collected data set can be converted into a solution process for a matrix describing the causal relationship between multiple elements. At present, a technical solution has been proposed to solve the matrix by constructing a problem formula based on a data set and expert knowledge. However, in the existing technical solutions, the requirements for expert knowledge are too stringent, and it is difficult to obtain expert knowledge that meets such stringent requirements.

[0040] For example, in a technical solution, expert knowledge needs to describe not only whether there is a causal relationship between two variables, but also the probability of the existence of the causal relationship between the two variables. It is understandable that in most existing application environments, expert knowledge is currently limited and incomplete. For example, in the application environment used to determine automobile insurance premiums above, there may be only limited expert knowledge: there is no direct causal relationship between the gender and age of the insured, there is an indirect causal relationship between the number of airbags and automobile insurance premiums, and so on. In existing technical solutions based on expert knowledge, the above-mentioned limited expert knowledge cannot provide all the required information and cannot be used to assist in determining causal relationships.

[0041] In order to solve the defects in the above technical solution, according to an implementation of the present disclosure, a method for determining the causal relationship between multiple variables is proposed. In the technical solution, the concept of expert knowledge constraint is introduced when constructing the problem formula. The expert knowledge constraint includes two constraints: 1) edge constraints for the direct causal relationship between two variables in a matrix describing a directed graph; and 2) path constraints for the indirect causal relationship between two variables in the matrix.

[0042] It should be noted that according to one implementation of the present disclosure, edge constraints and path constraints are determined based on existing (perhaps incomplete) expert knowledge. Thus, according to the present disclosure, expert knowledge can be used to assist in determining the causal relationship between multiple variables in a simple and effective manner without requiring complete expert knowledge between multiple variables.

[0043] Specifically, according to an implementation of the present disclosure, a method for determining a causal relationship between multiple variables is proposed. The method includes: in response to collecting a data set of multiple samples associated with multiple variables, obtaining a matrix describing the causal relationship between the multiple variables, each of the multiple samples includes data corresponding to the multiple variables; based on the data set and the matrix, determining the fit and expert knowledge constraints associated with the causal relationship, wherein the expert knowledge constraints include at least one of the edge constraints for the direct causal relationship between two variables in the matrix and the path constraints for the indirect causal relationship between two variables in the matrix; constructing a problem formula describing the causal relationship according to the determined fit and expert knowledge constraints; and solving the constructed problem formula to obtain a candidate result of the matrix.

[0044] In the following, we will refer to Figure 2 The steps of one implementation according to the present disclosure are summarized. Figure 2 A block diagram 200 for determining a causal relationship between multiple variables according to an implementation of the present disclosure is schematically shown. Figure 2 As shown, a data set 210 of multiple samples (n samples) associated with multiple variables (e.g., the data set shown in Table 1 above) can be collected. And a matrix 220 (e.g., the matrix B shown above) describing the causal relationship between the multiple variables can be obtained (as shown by arrow 252), and each sample in the multiple samples includes data corresponding to the multiple variables. At this time, the values ​​in each element in the constructed matrix are unknown and need to be obtained through the solution process below (as shown by arrow 262).

[0045] As shown by arrows 254 and 256, based on the data set 210 and the matrix 220, the degree of fit 230 and the expert knowledge constraint 232 associated with the causal relationship are determined. The expert knowledge constraint 232 includes at least one of the edge constraint for the direct causal relationship between two variables in the matrix and the path constraint for the indirect causal relationship between two variables in the matrix. As shown by arrows 258 and 260, a problem formula 240 describing the causal relationship can be constructed according to the determined degree of fit 230 and the expert knowledge constraint 232, respectively. Finally, as shown by arrow 262, the constructed problem formula 240 is solved, and the candidate results of the matrix 220 can be obtained. At this time, the candidate results represent the specific values ​​of each element in the matrix 220.

[0046] Figure 3A flowchart of a method 300 for determining a causal relationship between multiple variables according to an implementation of the present disclosure is schematically shown. First, at block 310, in response to collecting a data set 210 of multiple samples associated with multiple variables, a matrix 220 describing the causal relationship between the multiple variables is obtained, each sample in the multiple samples includes data corresponding to the multiple variables.

[0047] Here, the data set 210 is a collection of sample data collected from the application environment. For example, in the application environment for determining whether the control system is abnormal, the data set 210 can be a data set as shown in Table 1 above, where each row of records represents a sample. Each sample includes p variables x1, x2, ..., x p The measured value of p. At this time, a matrix 220 (e.g., the form of the matrix B described above) describing the causal relationship between the p variables can be obtained. It should be noted that the value of each element in the matrix 220 is not known here, but the value of each element can be considered as an unknown number. In the subsequent steps, the candidate results of the matrix (i.e., the value of each element in the matrix) can be obtained by constructing and solving the problem formula. Since the causal relationship between multiple variables is described by the matrix 220, the candidate results of the matrix represent the causal relationship.

[0048] At box 320, based on the data set 210 and the matrix 220, the fitness 230 associated with the causal relationship and the expert knowledge constraints 232 associated with the causal relationship are determined, wherein the expert knowledge constraints 232 include at least any one of the edge constraints for the direct causal relationship between two variables in the matrix and the path constraints for the indirect causal relationship between two variables in the matrix.

[0049] The matrix 220 may be in the form of the matrix B described above, wherein the matrix 220 may include p vectors, each of which is represented by a row in the matrix 220. In the following, for the sake of simplicity of description, only a three-dimensional matrix with p=3 is used as a specific example of the matrix 220 describing the causal relationship to introduce the specific details of an implementation of the present disclosure. At this time, the data set 210 may be represented in the form of Table 2.

[0050] Table 2 Examples of datasets

[0051]

[0052] When p=3, the matrix can be expressed as:

[0053]

[0054] At this point, the vectors in the matrix M are as follows:

[0055] The first vector: β1 = [0 β 1,2 β 1,3 ];

[0056] The second vector: β2 = [β 2,1 0 β 2,3 ];

[0057] The third vector: β3 = [β 3,1 β 3,2 0].

[0058] According to the implementation of the present disclosure, the degree of fit 230 and the expert knowledge constraint 232 can be determined for each of the multiple vectors included in the matrix. The specific formula will be described in detail below. At box 330, a problem formula 240 describing the causal relationship can be constructed based on the degree of fit 230 and the expert knowledge constraint 232. At box 340, the constructed problem formula 240 is solved to obtain the candidate results of the matrix 220. It should be noted that the candidate results include the specific values ​​of each element in the matrix 220. By using the candidate results, the matrix 220 describing the causal relationship between multiple variables can be obtained, thereby obtaining a directed graph of the relationship between the various variables. Since the matrix 220 includes multiple vectors β1, β2, and β3, the degree of fit and the expert knowledge constraint for each vector can be determined separately.

[0059] According to an implementation of the present disclosure, edge constraints and path constraints may be set for the direct causal relationship and the indirect causal relationship between two variables in the matrix. Figure 4 Describe the concepts of direct and indirect causation. Figure 4 An example of a directed graph 400 associated with a matrix according to an implementation of the present disclosure is schematically shown. Specifically, the directed graph 400 is associated with the matrix B described above. Nodes 410, 420, 430, 440, ..., 450 in the directed graph 400 represent variables x1, x2, x3, x4, ..., x50, respectively. p Directed edges 460, 462, 464, 466, 468, and 470 respectively indicate that there is a causal relationship between the variables represented by the two nodes at both ends of the directed edge.

[0060] For example, directed edge 460 from node 410 to node 450 indicates that variable x1 has a p There is a direct causal relationship, in other words, variable x1 is variable x pIn directed graph 400, a path includes a series of edges, and indicates that there is an indirect causal relationship between the two nodes at both ends of the path. Specifically, there are two paths from node 410 to node 450, wherein the first path includes edges 470, 468, and 466, and the second path includes edges 462 and 464. Here, the two paths represent the indirect causal relationship between the variable x1 represented by node 410 and the variable x represented by node 450. p There is an indirect causal relationship between them. Here, the number of edges included in the path can be called the path length. For example, the length of the first path is 3, while the length of the second path is 2.

[0061] In the following, how to construct the problem formula will be described with reference to a specific example. According to an implementation of the present disclosure, the problem formula may include a degree of fit and an expert knowledge constraint associated with a causal relationship. Since the matrix 220 includes multiple vectors β1, β2, and β3, the degree of fit and the expert knowledge constraint for each vector may be determined separately. According to an implementation of the present disclosure, the problem formula may be constructed, for example, in the form of the following formula 1:

[0062]

[0063] In formula 1, β j Represents the current vector of the plurality of vectors in the matrix 220, Fitness j and Constraint j They represent the β for the current vector respectively. j The fitness and expert knowledge constraints of j and Constraint j are respectively associated with the current vector β j Formula 1 can be solved to obtain the value that minimizes the problem formula and use this value as the current vector β j In this way, by solving each of the multiple vectors in the matrix 220, the value of each element in the matrix 220 can be obtained.

[0064] Since the operation for each of the multiple vectors is similar, the following will take one vector as a specific example to describe the details of an implementation according to the present disclosure. Specifically, the current vector β in the multiple vectors in the matrix 220 can be j , to determine the current vector β j The fit and expert knowledge constraints.

[0065] For each vector in the matrix, the fitness represents the vector in the data set that is consistent with the current vector β jThe corresponding fitting relationship between the current column and other columns. According to an implementation of the present disclosure, the β for the current vector can be determined based on the two-norm operation. j Fitness j Specifically, the β for the current vector can be determined based on the following formula 1: j Fitness j :

[0066]

[0067] In formula 1, x j represents the multiple samples in the data set 210 and the current vector β j The associated part, x -j Represents multiple samples with the current vector β j The vector associated with the part other than β j represents the current vector, represents the two-norm operation. For example, when j = 1, x j represents the values ​​T1, T2, ..., Tn in the first column "variable x1 = temperature" in the data set shown in Table 1; and x -j represents the values ​​in columns other than the first column in the data set shown in Table 1 (for example, M1, M2, ..., Mn in the second column "variable x2 = humidity", and the pth column "variable x p = Abnormality exists in "E1, E2, ..., En").

[0068] In formula 2, Represents the β with the current vector j The associated fitness, where the fitness represents the fitting relationship between the current column and other columns in the matrix. The fitness is expressed in the form of a two-norm, which can represent x j With x -j β j If The smaller the value of , the higher the degree of fit between the two; otherwise, the lower the degree of fit between the two.

[0069] In the following, how to determine edge constraints and path constraints will be described with reference to specific examples. According to one implementation of the present disclosure, an edge constraint matrix of the state of the direct causal relationship between two variables in a plurality of variables can be determined based on expert knowledge; and edge constraints can be determined based on the edge constraint matrix and the matrix.

[0070] In this implementation, expert knowledge may include multiple factors. For example, in an application environment for determining whether a control system is abnormal, expert knowledge may indicate:p There is a direct causal relationship between the variable x2 (humidity) and the variable x p There is an indirect causal relationship between (existence of anomalies). At this time, the edge constraints in the directed graph represented by the matrix can be determined based on the above direct and indirect causal relationships.

[0071] Compared with the technical solutions in the prior art that assist in determining causal relationships based on expert knowledge, in the context of the present disclosure, precise expert knowledge is not required. For example, expert knowledge only needs to generally indicate whether there is a causal relationship between two variables, and there is no need to give more descriptions for the existence of a causal relationship (for example, no specific probability is required). In addition, after the edge constraints and path constraints have been determined based on expert knowledge, no other additional interactions are required. Therefore, according to an implementation of the present disclosure, the requirements for the perfection of expert knowledge can be reduced, and the determination of causal relationships can be assisted in a simple and effective manner without the need for additional interaction with experts.

[0072] Figure 5 A block diagram 500 of a method for determining expert knowledge constraints according to an implementation of the present disclosure is schematically shown. Figure 5 As shown, based on the expert knowledge 510 (e.g., describing which variables have direct causal relationships and which variables have indirect causal relationships), the edge knowledge matrix 520 (e.g., represented as a matrix K) and the path knowledge matrix 522 (e.g., represented as a matrix A) can be determined respectively. Then, based on the edge knowledge matrix 520 and the path knowledge matrix 522, the edge constraint matrix 530 (e.g., represented as a matrix ) and the path constraint matrix 532 (e.g., represented as a matrix ). Further, the expert knowledge constraint 232 may be determined based on the edge constraint matrix 530 and the path constraint matrix 532.

[0073] In the following, we will describe how to determine Figure 5 In one implementation of the present disclosure, in order to introduce expert knowledge into the constraints, the knowledge about the edges in the directed graph can be determined from the expert knowledge. Specifically, the edge knowledge matrix K describing the state of the direct causal relationship between two variables in the plurality of variables can be determined based on the expert knowledge. In one implementation of the present disclosure, each element k in the edge knowledge matrix K can be determined based on the following formula 3: ji .

[0074]

[0075] In Formula 3, if the expert knowledge indicates that the first variable associated with the element (e.g., variable x j ) and a second variable (e.g., variable x i ) have a direct causal relationship, and the element k ji is set to satisfy a first condition (e.g., 1); in response to expert knowledge indicating that there is no direct causal relationship between the first variable and the second variable, the element k ... ji Set to satisfy the second condition; and in response to the expert knowledge not having the knowledge of a direct causal relationship between the first variable and the second variable, the element k ji Set to satisfy the third condition (eg, -1).

[0076] According to an implementation of the present disclosure, a path knowledge matrix describing the state of an indirect causal relationship between two variables among multiple variables may be determined based on expert knowledge. Specifically, each element a in the edge knowledge matrix A may be determined based on the following formula 4: ji .

[0077]

[0078] In Formula 4, if the expert knowledge indicates that the first variable associated with the element (e.g., variable x j ) and a second variable (e.g., variable x i ), the element is set to satisfy a first condition (e.g., 1); if the expert knowledge indicates that there is no indirect causal relationship between the first variable and the second variable, the element is set to satisfy a second condition; and if in response to the expert knowledge that there is no indirect causal relationship between the first variable and the second variable, the element is set to satisfy a third condition (e.g., -1).

[0079] It should be noted that although the element k is changed under different conditions in the above formula ji and a ji Set to 1, 0, -1. In other specific implementations, element k can also be ji It is set to meet other first, second, and third conditions, as long as three different conditions can be distinguished.

[0080] In the above, it has been described in detail how to determine the edge knowledge matrix K and the path knowledge matrix A based on expert knowledge. Subsequently, the edge constraint matrix can be determined based on the edge knowledge matrix K and the path knowledge matrix A. Figure 6 Describe how to determine each element in the edge knowledge matrix and the path knowledge matrix based on the edge knowledge matrix K and the path knowledge matrix A.

[0081] Figure 6A block diagram 600 is schematically shown for determining the edge knowledge matrix and each element in the path knowledge matrix based on the edge knowledge matrix and the path knowledge matrix according to an implementation of the present disclosure. Figure 6 As shown, row 610 in block diagram 600 shows possible values ​​of each element in edge knowledge matrix K (e.g., 0, 1, -1), and column 620 shows possible values ​​of each element in path knowledge matrix A (e.g., 0, 1, -1). Figure 6 The relationship shown can be used to determine the elements in the edge constraint matrix based on the following method.

[0082] As shown in the second column (k ji =0), the edge knowledge matrix K has the same elements as The corresponding element k ji Satisfy the second condition (equal to 0), and set the edge constraint matrix Elements in It will be understood that although Figure 6 The specific example of the larger value is shown as infinity. In other implementations, Set to any other value that gives a large penalty relative to the edge constraint.

[0083] As shown in the third column (k ji =1) as shown in the figure, if the edge knowledge matrix K has the same element The corresponding element k ji To satisfy the first condition (e.g., equal to 1), the edge constraint matrix Elements in Set to zero.

[0084] As shown in the fourth column (k ji =-1) as shown in the figure, if the element corresponding to the element in the edge knowledge matrix satisfies the third condition, the edge constraint matrix Elements in Set to the corresponding element β in matrix B ji Associated.

[0085] According to an implementation of the present disclosure, the edge constraint matrix can be determined based on the following formula 5: The element at position (j, i) in

[0086]

[0087] In formula 5, is a predefined hyperparameter calculated based on the data set, ols represents the least squares operation, || represents the absolute value operation, and γ represents a predetermined parameter. In this implementation, the value of γ can be determined based on the requirements of the specific application environment. Those skilled in the art can determine the value based on the relevant definitions in the prior art. The numerical value will not be repeated in the implementation method according to the content of this disclosure.

[0088] According to one implementation of the present disclosure, according to Figure 6 The relationship shown in FIG. 6A and FIG. 6B can be used to determine the elements in the path constraint matrix based on the following method. ji =0), in response to the path knowledge matrix A and the element The corresponding element a ji Satisfy the second condition (e.g., equal to 0), and set the path constraint matrix Elements in It will be understood that although Figure 6 The specific example of the larger value is shown as infinity. In other implementations, Set to any other value that gives a large penalty relative to the path constraint.

[0089] As shown in the third row (a ji =1) as shown in the figure, if the path knowledge matrix A has the same element The corresponding element a ji Satisfy the first condition (e.g., 1), and transform the path constraint matrix Elements in Set to zero.

[0090] As shown in the fourth row (a ji =-1) as shown in the figure, if the path knowledge matrix A has the same element The corresponding element a ji Satisfy the third condition (e.g., -1), and set the path constraint matrix Elements in Set to 1.

[0091] It should be noted that the symbol "NA" is also shown in the block diagram 600, which indicates that such a situation is not possible. For example, see the second row (a ji =-1) and the third column (k ji =1), which indicates that there is no intersection between the two variables x in the expert knowledge. j and x i The knowledge of the edge relationship between the two variables x j and x iThere is a path between them, which is inconsistent with the definition when constructing the edge knowledge matrix and the path knowledge matrix. Therefore, "NA" is used to indicate that such a situation cannot exist.

[0092] According to one implementation of the present disclosure, in order to determine the constraints for a path in a directed graph represented by a matrix, a path matrix describing a path between two variables among a plurality of variables can be first determined based on the matrix; then, the path constraints can be determined based on the path knowledge matrix, the path matrix, and the matrix.

[0093] In this implementation, during the determination of the constraints for the path in the directed graph represented by the matrix, a process of generating the path based on the multiple edges in the directed graph will be involved. Thus, at this time, the path constraint matrix of the state of the indirect causal relationship between two variables among the multiple variables can be determined based on expert knowledge. Then, based on the path constraint matrix The sum matrix determines the path constraints.

[0094] Specifically, the element β in the matrix ji represents the element at position (j, i) in the matrix, β ji That is, it means variable x among multiple variables j With x i In order to obtain the indirect causal relationship between two variables among multiple variables, the matrix can also be multiplied to obtain the indirect causal relationship between two variables with a path length greater than 1.

[0095] How to determine the path matrix associated with paths of different lengths will be described below. According to one implementation of the present disclosure, paths of length K=2, 3, ..., p-1 can be determined step by step. Specifically, a K-length path matrix associated with a path of length K can be determined, wherein the K-length path matrix represents the state of a path of length K between two nodes in a directed graph associated with the matrix, wherein K≤p-1, p represents the number of multiple variables; and a path matrix is ​​determined based on the K-length path matrix.

[0096] Assuming K=2, then the "2-length path matrix" can be determined based on the multiplication of the matrix B as above. Specifically, the 2-length path matrix can be determined based on B×B. According to one implementation of the present disclosure, a K-length path matrix associated with a path length K can be determined based on K-times multiplication of the matrix B. For the convenience of description, the K-length path matrix can be represented as M K , the K-length path matrix can be determined based on the following formula 6 and expressed as M K :

[0097]

[0098] In Formula 6, represents K consecutive multiplications, and abs(B) represents the absolute value of matrix B.

[0099] According to an implementation of the present disclosure, after a K-length path matrix M has been obtained K In the case of K The sum is performed to obtain a path matrix P describing paths of various lengths between two nodes in the directed graph. Specifically, the path matrix P can be determined based on the following formula 7.

[0100]

[0101] The element p at position (j, i) in the matrix P is ji It can be calculated based on the following formula 8:

[0102]

[0103] According to an implementation of the present disclosure, a vector associated with the current vector β can be constructed based on the following formula 9: j The problem formula is:

[0104]

[0105] In Formula 9, β j represents the current vector, represents the part of the edge constraint matrix associated with the current vector, Represents the current vector β j The associated fitness, x j represents the multiple samples that are related to the current vector β j The associated part, x -j represents the multiple samples that are related to the current vector β j The vectors outside the associated part, represents the constraints, where is a predetermined parameter.

[0106] At this time, the expert knowledge constraints include the current vector β in the matrix j The associated edge constraint vector and path constraint vector, where: the edge constraint vector is And the path constraint vector is

[0107] According to the context of the present disclosure, since it is desirable to describe a certain causal relationship in a directed acyclic graph, constraints may also be set to ensure that there are no loops in the matrix obtained by solving. Variables in multiple variables may be considered one by one by constructing a causal sequence, so that different causal sequences may be formed in order to reach a target state.

[0108] Figure 7 A block diagram 700 for determining a score for reaching a target state from a state according to an implementation of the present disclosure is schematically shown. Figure 7 As shown, based on the basic principle of causal inference, assuming that the current causal sequence is Q s , and with the causal sequence Q s The associated state is shown as node 720. s The score of the associated state reaching the target state (eg, as shown in node 730, the state associated with the causal sequence including all variables) can be calculated according to the following formula:

[0109] f(Q s )=g(Q s )+h(Q s ) Formula 10

[0110] In Formula 10, f(Q s ) represents the transition from the current causal sequence Q s The score of the associated state to reach the target state, g(Q s ) represents the transition from the initial state (the empty set shown in node 710) to the current causal sequence Q s The score of the associated state (as shown in node 720), h(Q s ) represents the transition from the current causal sequence Q s The predicted score of the associated state to reach the goal state.

[0111] based on Figure 7 As shown, the target formula can be constructed based on formula 11:

[0112]

[0113] In formula 11, Represents the variable x in the universe j The set of previous variables, Supp(β j ) represents the current vector β j Support, Q s represents a causal sequence associated with at least some of the elements, x j Represents multiple samples with the current vector β j The associated part, x -jRepresents multiple samples with the current vector β j The vectors outside the associated part, represents the two-norm operation, represents the element at position (j, i) in the edge constraint matrix, β ji represents the element at position (j, i) in the matrix, represents the element at position (j, i) in the path constraint matrix, p vi represents the element at position (j, i) in the path matrix, β jv represents the element at position (j, v) in the matrix, and || represents the absolute value operation.

[0114] Furthermore, the above formulas include constraints associated with edges and constraints associated with paths, respectively, and these two constraints can be set to different values. At this time, by setting the constraints associated with paths in formula 11 to different values, the following formula 12 can be obtained.

[0115]

[0116] In this formula 12, Indicates Q s In the variable x i After the node, represents the constraints associated with the edge, and Represents a constraint associated with a path.

[0117] According to an implementation of the present disclosure, a solution can be performed based on the above formula 12 to obtain candidate results for each vector of the matrix. It should be noted that a solution algorithm known in the prior art or to be developed in the future is used here to determine the value of each element in the matrix. In formula 12, edge constraints and path constraints based on expert knowledge have been introduced, so that the existing expert knowledge can be used to assist in determining the causal relationship between multiple variables. It should be noted that although the formula given in the above formula 12 includes edge constraints and path constraints According to an implementation of the present disclosure, the constructed problem formula may include only edge constraints, only path constraints, or both edge constraints and path constraints.

[0118] Figure 8A block diagram of a device 800 for determining a causal relationship between multiple variables according to an implementation of the present disclosure is schematically shown. According to an implementation of the present disclosure, a device 800 for determining a causal relationship between multiple variables is provided. The device includes: an acquisition module 810, configured to acquire a matrix describing the causal relationship between multiple variables in response to a data set of multiple samples associated with multiple variables, each of the multiple samples includes data corresponding to multiple variables; a determination module 820, configured to determine the fitness and expert knowledge constraints associated with the causal relationship based on the data set and the matrix, wherein the expert knowledge constraints include at least one of the edge constraints for the direct causal relationship between two variables in the matrix and the path constraints for the indirect causal relationship between two variables in the matrix; a construction module 830, configured to construct a problem formula describing the causal relationship according to the determined fitness and expert knowledge constraints; and a solution module 840, configured to solve the constructed problem formula to obtain a candidate result of the matrix.

[0119] According to an implementation of the present disclosure, the determination module 820 is further configured to determine an edge constraint matrix of a state of a direct causal relationship between two variables among the plurality of variables based on expert knowledge; and determine an edge constraint based on the edge constraint matrix and the matrix.

[0120] According to one implementation of the present disclosure, the determination module 820 is further configured to determine a path constraint matrix of a state of an indirect causal relationship between two variables among the plurality of variables based on expert knowledge; and determine a path constraint based on the path constraint matrix and the matrix.

[0121] According to an implementation of the present disclosure, the determination module 820 is further configured to determine a path matrix describing a path between two variables among the multiple variables based on the matrix; and determine a path constraint based on the path knowledge matrix, the path matrix, and the matrix.

[0122] According to one implementation of the present disclosure, the determination module 820 is further configured to determine a K-length path matrix associated with a path of length K, wherein the K-length path matrix represents the state of a path of length K between two nodes in a directed graph associated with the matrix, wherein K≤p-1, p represents the number of multiple variables; and determine the path matrix based on the K-length path matrix.

[0123] According to an implementation of the present disclosure, the determination module 820 is further configured to determine a side knowledge matrix describing the state of a direct causal relationship between two variables among the multiple variables based on expert knowledge; and determine a side constraint matrix based on the side knowledge matrix.

[0124] According to one implementation of the present disclosure, the determination module 820 is further configured to, for an element in the edge knowledge matrix, set the element to satisfy a first condition in response to expert knowledge indicating that there is a direct causal relationship between a first variable and a second variable associated with the element; set the element to satisfy a second condition in response to expert knowledge indicating that there is no direct causal relationship between the first variable and the second variable; and set the element to satisfy a third condition in response to expert knowledge that there is no direct causal relationship between the first variable and the second variable.

[0125] According to one implementation of the present disclosure, the determination module 820 is further configured to, for an element in the path knowledge matrix, set the element to satisfy a first condition in response to expert knowledge indicating that there is an indirect causal relationship between a first variable and a second variable associated with the element; set the element to satisfy a second condition in response to expert knowledge indicating that there is no indirect causal relationship between the first variable and the second variable; and set the element to satisfy a third condition in response to expert knowledge that there is no indirect causal relationship between the first variable and the second variable.

[0126] According to one implementation of the present disclosure, the determination module 820 is further configured to, for an element in the edge constraint matrix, set the element in the edge constraint matrix to a larger value in response to an element in the edge knowledge matrix corresponding to the element being equal to 0; set the element in the edge constraint matrix to zero in response to an element in the edge knowledge matrix corresponding to the element satisfying a first condition; and set the element in the edge constraint matrix to be associated with the corresponding element in the matrix in response to an element in the edge knowledge matrix corresponding to the element satisfying a third condition.

[0127] According to an implementation of the present disclosure, the determination module 820 is further configured to convert the element at position (j, i) in the edge constraint matrix Set to in is a predefined hyperparameter calculated based on the data set, ols represents a least squares operation, || represents an absolute value operation, and γ represents a predetermined parameter.

[0128] According to an implementation of the present disclosure, the determination module 820 is further configured to determine a path knowledge matrix describing the state of an indirect causal relationship between two variables among the multiple variables based on expert knowledge; and determine a path constraint matrix based on the path knowledge matrix.

[0129] According to one implementation of the present disclosure, the determination module 820 is further configured to, for an element in the path constraint matrix, set the element in the path constraint matrix to a larger value in response to an element in the path knowledge matrix corresponding to the element being equal to 0; set the element in the path constraint matrix to zero in response to an element in the path knowledge matrix corresponding to the element satisfying a first condition; and set the element in the path constraint matrix to 1 in response to an element in the path knowledge matrix corresponding to the element satisfying a third condition.

[0130] According to an implementation of the present disclosure, the edge constraint vector is And the path constraint vector is

[0131] According to an implementation of the present disclosure, the construction module 830 is further configured to: j Construct the problem formula vector:

[0132]

[0133] Where Supp(β j ) represents the current vector β j Support, Q s represents a causal sequence associated with at least some of the elements, x j Represents multiple samples with the current vector β j The associated part, x -j Represents multiple samples with the current vector β j The vectors outside the associated part, represents the two-norm operation, represents the element at position (j, i) in the edge constraint matrix, β ji represents the element at position (j, i) in the matrix, represents the element at position (j, i) in the path constraint matrix, p vi represents the element at position (j, i) in the path matrix, β jv represents the element at the position (j, v) in the matrix, || represents the absolute value operation, Indicates Q s The node after node i in .

[0134] According to an implementation of the present disclosure, the problem formula vector is solved to obtain the current vector β for the matrix j Candidate results.

[0135] According to an implementation of the present disclosure, a system for determining a causal relationship between multiple variables is provided, comprising: one or more processors; a memory coupled to at least one of the one or more processors; computer program instructions stored in the memory, when the computer program instructions are executed by at least one processor, the system executes a method for determining a causal relationship between multiple variables. The method comprises: in response to a data set of multiple samples associated with multiple variables, a matrix describing the causal relationship between the multiple variables is obtained, each of the multiple samples includes data corresponding to the multiple variables; based on the data set and the matrix, the fit and expert knowledge constraints associated with the causal relationship are determined, wherein the expert knowledge constraints include at least one of the edge constraints for the direct causal relationship between two variables in the matrix and the path constraints for the indirect causal relationship between two variables in the matrix; a problem formula describing the causal relationship is constructed according to the determined fit and expert knowledge constraints; and the constructed problem formula is solved to obtain a candidate result of the matrix.

[0136] According to one implementation of the present disclosure, determining expert knowledge constraints includes determining edge constraints, including: determining an edge constraint matrix of a state of a direct causal relationship between two variables among a plurality of variables based on expert knowledge; and determining the edge constraints based on the edge constraint matrix and the matrix.

[0137] According to one implementation of the present disclosure, determining expert knowledge constraints includes determining path constraints, including: determining a path constraint matrix of the state of an indirect causal relationship between two variables among multiple variables based on expert knowledge; and determining the path constraints based on the path constraint matrix and the matrix.

[0138] According to one implementation of the present disclosure, determining the path constraint based on the path constraint matrix and the matrix includes: determining a path matrix describing a path between two variables among a plurality of variables based on the matrix; and determining the path constraint based on the path knowledge matrix, the path matrix and the matrix.

[0139] According to one implementation of the present disclosure, determining a path matrix includes: determining a K-length path matrix associated with a path of length K, wherein the K-length path matrix represents a state of a path of length K between two nodes in a directed graph associated with the matrix, wherein K≤p-1, and p represents the number of multiple variables; and determining the path matrix based on the K-length path matrix.

[0140] According to an implementation of the present disclosure, determining the edge constraint matrix includes: determining an edge knowledge matrix describing the state of a direct causal relationship between two variables among a plurality of variables based on expert knowledge; and determining an edge constraint matrix based on the edge knowledge matrix.

[0141] According to one implementation of the present disclosure, determining the edge knowledge matrix includes: for an element in the edge knowledge matrix, in response to expert knowledge indicating that there is a direct causal relationship between a first variable and a second variable associated with the element, setting the element to satisfy a first condition; in response to expert knowledge indicating that there is no direct causal relationship between the first variable and the second variable, setting the element to satisfy a second condition; and in response to expert knowledge that there is no direct causal relationship between the first variable and the second variable, setting the element to satisfy a third condition.

[0142] According to one implementation of the present disclosure, determining a path knowledge matrix includes: for an element in the path knowledge matrix, in response to expert knowledge indicating that there is an indirect causal relationship between a first variable and a second variable associated with the element, setting the element to satisfy a first condition; in response to expert knowledge indicating that there is no indirect causal relationship between the first variable and the second variable, setting the element to satisfy a second condition; and in response to expert knowledge that there is no indirect causal relationship between the first variable and the second variable, setting the element to satisfy a third condition.

[0143] According to one implementation of the present disclosure, determining an edge constraint matrix based on an edge knowledge matrix includes: for an element in the edge constraint matrix, in response to an element in the edge knowledge matrix corresponding to the element being equal to 0, setting the element in the edge constraint matrix to a larger value; in response to an element in the edge knowledge matrix corresponding to the element satisfying a first condition, setting the element in the edge constraint matrix to zero; and in response to an element in the edge knowledge matrix corresponding to the element satisfying a third condition, setting the element in the edge constraint matrix to be associated with the corresponding element in the matrix.

[0144] According to an implementation of the present disclosure, setting an element in the edge constraint matrix to be associated with a corresponding element in the matrix includes: setting the element at position (j, i) in the edge constraint matrix to be associated with a corresponding element in the matrix. Set to in is a predefined hyperparameter calculated based on the data set, ols represents a least squares operation, || represents an absolute value operation, and γ represents a predetermined parameter.

[0145] According to one implementation of the present disclosure, determining a path constraint matrix includes: determining a path knowledge matrix describing a state of an indirect causal relationship between two variables among a plurality of variables based on expert knowledge; and determining a path constraint matrix based on the path knowledge matrix.

[0146] According to one implementation of the present disclosure, determining a path constraint matrix based on a path knowledge matrix includes: for an element in the path constraint matrix, in response to an element in the path knowledge matrix corresponding to the element being equal to 0, setting the element in the path constraint matrix to a larger value; in response to an element in the path knowledge matrix corresponding to the element satisfying a first condition, setting the element in the path constraint matrix to satisfy a third condition; and in response to an element in the path knowledge matrix corresponding to the element satisfying the third condition, setting the element in the path constraint matrix to 1.

[0147] According to one implementation of the present disclosure, the expert knowledge constraint includes the current vector β in the matrix j The associated edge constraint vector and path constraint vector, where: the edge constraint vector is And the path constraint vector is

[0148] According to an implementation of the present disclosure, constructing a problem formula describing causal relationships includes: for the current vector β j Construct the problem formula vector: Where Supp(β j ) represents the current vector β j Support, Q s represents a causal sequence associated with at least some of the elements, x j Represents multiple samples with the current vector β j The associated part, x -j Represents multiple samples with the current vector β j The vectors outside the associated part, represents the two-norm operation, represents the element at position (j, i) in the edge constraint matrix, β ji represents the element at position (j, i) in the matrix, represents the element at position (j, i) in the path constraint matrix, p vi represents the element at position (j, i) in the path matrix, β jv represents the element at position (j, v) in the matrix, and || represents the absolute value operation.

[0149] According to an implementation of the present disclosure, solving the constructed problem formula to obtain a candidate result of the matrix includes: solving the problem formula vector to obtain the current vector β for the matrix j Candidate results.

[0150] According to one implementation of the present disclosure, a computer program product is provided, which is tangibly stored on a non-transitory computer-readable medium and includes machine-executable instructions, which, when executed, cause a machine to perform the steps of the method described above.

[0151] The various implementations of the method of the present invention have been described above with reference to the accompanying drawings. It can be understood by those skilled in the art that the above method can be implemented in software, hardware, or a combination of software and hardware. In addition, it can be understood by those skilled in the art that by implementing the various steps in the above method in software, hardware, or a combination of software and hardware, a device based on the same inventive concept can be provided. Even if the device is the same as a general processing device in hardware structure, due to the role of the software contained therein, the device exhibits characteristics different from the general processing device, thereby forming a device of various implementations of the present invention. The device described in the present invention includes several devices or modules, and the devices or modules are configured to perform corresponding steps. The technicians in this field can understand how to write a program to implement the actions performed by the device or module by reading this specification. Since the device and the method are based on the same inventive concept, the same or corresponding implementation details therein are also applicable to the device or module corresponding to the above method, and since it has been described in detail and completely above, it may not be repeated in the following.

[0152] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to multiple implementation modes of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a part of a module, program segment or code, and the part of the module, program segment or code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0153] The various implementations of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed implementations. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described implementations. The choice of terms used herein is intended to best explain the principles of the implementations, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the implementations disclosed herein.

Claims

1. A method for determining data causing an abnormality in a control system, comprising: receiving an input data set, wherein data in the data set includes temperature, humidity, and information about whether the control system is abnormal; In response to receiving the data set, obtaining a matrix describing a causal relationship between a temperature variable, a humidity variable, and a variable related to whether the control system is abnormal; Based on the data set, the element values ​​in the matrix of the causal relationship are determined by solving a problem formula, wherein the problem formula is constructed in the following manner: β1 , ..., β p Multiple vectors in the matrix representing the causal relationship, Fitness j and Constraint j They respectively represent the fitness of the current vector and the expert knowledge constraint for the multiple vectors in the matrix, and the fitness is determined in the following manner: x j Represents the multiple samples in the data set that are related to the current vector β j The associated part, x -j represents the multiple samples that are related to the current vector β j The vector associated with the part other than β j represents the current vector, represents a two-norm operation, the expert knowledge constraint includes at least one of an edge constraint or a path constraint, the edge constraint is a constraint on a direct causal relationship between two variables among the temperature variable, the humidity variable, and the variable regarding whether the control system is abnormal, is the element at position (j,i) in the edge constraint matrix, β ji The element at position (j,i) in the matrix representing the causal relationship, the path constraint is a constraint on the indirect causal relationship between the two variables. is the element at position (j,i) in the path constraint matrix used to determine the path constraint, p υi represents the element at position (v,i) in the path matrix describing the path between the two variables, β jυ The element at position (j,v) in the matrix representing the causal relationship, Q s represents a causal sequence associated with at least some of the multiple elements, Indicates Q s The node after node i in ; Based on the determined element values ​​in the matrix of the cause-effect relationship, data causing the abnormality of the control system is determined.

2. The method according to claim 1, further comprising: Determine, based on expert knowledge, a side constraint matrix of a state of a direct causal relationship between two variables among the temperature variable, the humidity variable, and a variable regarding whether the control system is abnormal; as well as The edge constraint is determined based on the edge constraint matrix and the matrix.

3. The method according to claim 2, further comprising: determining a path constraint matrix of a state of an indirect causal relationship between two variables among the temperature variable, the humidity variable, and a variable regarding whether the control system is abnormal based on the expert knowledge; as well as The path constraint is determined based on the path constraint matrix and the matrix.

4. The method according to claim 3, wherein the determining the path constraint based on the path constraint matrix and the matrix comprises: determining, based on the matrix, a path matrix describing a path between two variables among the temperature variable, the humidity variable, and a variable regarding whether the control system is abnormal; as well as The path constraints are determined based on a path knowledge matrix, the path matrix, and the matrix.

5. The method of claim 4, wherein said determining said path matrix comprises: Determining a K-length path matrix associated with the path of length K, wherein the K-length path matrix represents the state of a path of length K between two nodes in a directed graph associated with the matrix, wherein K≤p-1, and p represents the number of the temperature variable, the humidity variable, and the variable regarding whether the control system is abnormal; and The path matrix is ​​determined based on the K-length path matrix.

6. The method of claim 4, wherein determining the edge constraint matrix comprises: determining a side knowledge matrix describing a state of a direct causal relationship between two variables among the temperature variable, the humidity variable, and a variable regarding whether the control system is abnormal based on the expert knowledge; as well as The edge constraint matrix is ​​determined based on the edge knowledge matrix.

7. The method according to claim 6, wherein the determining the edge knowledge matrix comprises: For the elements in the edge knowledge matrix, In response to the expert knowledge indicating that there is a direct causal relationship between a first variable and a second variable associated with an element in the side knowledge matrix, setting the element in the side knowledge matrix to satisfy a first condition; In response to the expert knowledge indicating that there is no direct causal relationship between the first variable and the second variable, setting an element in the edge knowledge matrix to satisfy a second condition; as well as In response to the expert knowledge not having knowledge of a direct causal relationship between the first variable and the second variable, an element in the side knowledge matrix is ​​set to satisfy a third condition.

8. The method according to claim 6, wherein the determining the path knowledge matrix comprises: For the elements in the path knowledge matrix, In response to the expert knowledge indicating that there is an indirect causal relationship between a first variable and a second variable associated with an element in the path knowledge matrix, setting the element in the path knowledge matrix to satisfy a first condition; In response to the expert knowledge indicating that there is no indirect causal relationship between the first variable and the second variable, setting an element in the path knowledge matrix to satisfy a second condition; as well as In response to the expert knowledge not having knowledge of an indirect causal relationship between the first variable and the second variable, an element in the path knowledge matrix is ​​set to satisfy a third condition.

9. The method according to claim 6, wherein the determining the edge constraint matrix based on the edge knowledge matrix comprises: For the elements in the edge constraint matrix, In response to an element in the edge knowledge matrix corresponding to an element in the edge constraint matrix being equal to 0, setting the element in the edge constraint matrix to a larger value; In response to an element in the edge knowledge matrix corresponding to an element in the edge constraint matrix satisfying a first condition, setting the element in the edge constraint matrix to zero; as well as In response to an element in the edge knowledge matrix corresponding to an element in the edge constraint matrix satisfying a third condition, the element in the edge constraint matrix is ​​set to be associated with a predefined hyperparameter.

10. The method of claim 9, wherein setting the elements in the edge constraint matrix to be associated with predefined hyperparameters comprises: Set the element at position (j, i) in the edge constraint matrix Set to in is a predefined hyperparameter calculated based on the data set, ols represents a least squares operation, || represents an absolute value operation, and γ represents a predetermined parameter.

11. The method of claim 6, wherein determining the path constraint matrix comprises: determining a path knowledge matrix describing a state of an indirect causal relationship between two variables among the temperature variable, the humidity variable, and a variable regarding whether the control system is abnormal based on the expert knowledge; as well as The path constraint matrix is ​​determined based on the path knowledge matrix.

12. The method according to claim 11, wherein the determining the path constraint matrix based on the path knowledge matrix comprises: For the elements in the path constraint matrix, In response to an element in the path knowledge matrix corresponding to an element in the path constraint matrix being equal to 0, setting the element in the path constraint matrix to a larger value; In response to an element in the path knowledge matrix corresponding to an element in the path constraint matrix satisfying a first condition, setting the element in the path constraint matrix to zero; as well as In response to an element in the path knowledge matrix corresponding to an element in the path constraint matrix satisfying a third condition, setting the element in the path constraint matrix to 1.

13. The method according to claim 12, wherein the expert knowledge constraint comprises a constraint corresponding to the current vector β in the matrix j The associated edge constraint vector and path constraint vector, where: The edge constraint vector is And the path constraint vector is 14. The method according to claim 13, wherein constructing the problem formulation describing the causal relationship comprises: For the current vector β j Construct the problem formula vector: Where Supp(β j ) represents the current vector β j Support, Q s represents a causal sequence associated with at least some of the elements, x j Represents multiple samples with the current vector β j The associated part, x -j Represents multiple samples with the current vector β j The part of the vector associated with the represents the two-norm operation, represents the element at position (j, i) in the edge constraint matrix, β ji represents the element at position (j, i) in the matrix, represents the element at position (j, i) in the path constraint matrix, p υi represents the element at position (j, i) in the path matrix, β jυ represents the element at the position (j, v) in the matrix, || represents the absolute value operation, Indicates Q s The node after node i in .

15. The method of claim 14, wherein determining the element value in the matrix of causal relationships comprises: Solve the problem formula vector to obtain the current vector β for the matrix j Candidate results.

16. A system for determining data causing an abnormality in a control system, comprising: One or more processors; a memory coupled to at least one of the one or more processors; The computer program instructions stored in the memory, when executed by the at least one processor, cause the system to perform a method for determining data causing an abnormality in a control system, the method comprising: receiving an input data set, wherein data in the data set includes temperature, humidity, and information about whether the control system is abnormal; In response to receiving the data set, obtaining a matrix describing a causal relationship between a temperature variable, a humidity variable, and a variable related to whether the control system is abnormal; Based on the data set, the element values ​​in the matrix of the causal relationship are determined by solving a problem formula, wherein the problem formula is constructed in the following manner: β1, ..., β p Multiple vectors in the matrix representing the causal relationship, Fitness j and Constraint j They respectively represent the fitness of the current vector and the expert knowledge constraint for the multiple vectors in the matrix, and the fitness is determined in the following manner: x j Represents the multiple samples in the data set that are related to the current vector β j The associated part, x -j represents the multiple samples that are related to the current vector β j The vector associated with the part other than β j represents the current vector, represents a two-norm operation, the expert knowledge constraint includes at least one of an edge constraint or a path constraint, the edge constraint is a constraint on a direct causal relationship between two variables among the temperature variable, the humidity variable, and the variable regarding whether the control system is abnormal, is the element at position (j,i) in the edge constraint matrix, β ji The element at position (j,i) in the matrix representing the causal relationship, the path constraint is a constraint on the indirect causal relationship between the two variables. is the element at position (j,i) in the path constraint matrix used to determine the path constraint, p υi represents the element at position (v,i) in the path matrix describing the path between the two variables, β jυ The element at position (j,v) in the matrix representing the causal relationship, Q s represents a causal sequence associated with at least some of the multiple elements, Indicates Q s The node after node i in ; Based on the determined element values ​​in the matrix of the cause-effect relationship, data causing the abnormality of the control system is determined.

17. The system of claim 16, wherein the system is further configured to perform: Determine, based on expert knowledge, a side constraint matrix of a state of a direct causal relationship between two variables among the temperature variable, the humidity variable, and a variable regarding whether the control system is abnormal; and The edge constraint is determined based on the edge constraint matrix and the matrix.

18. The system of claim 17, wherein the system is further configured to perform: Determine a path constraint matrix of a state of an indirect causal relationship between two variables among the temperature variable, the humidity variable, and a variable regarding whether the control system is abnormal based on the expert knowledge; and The path constraint is determined based on the path constraint matrix and the matrix.

19. The system of claim 18, wherein the determining the path constraint based on the path constraint matrix and the matrix comprises: determining, based on the matrix, a path matrix describing a path between two variables among the temperature variable, the humidity variable, and a variable regarding whether the control system is abnormal; as well as The path constraints are determined based on a path knowledge matrix, the path matrix, and the matrix.

20. The system of claim 19, wherein said determining said path matrix comprises: Determining a K-length path matrix associated with the path of length K, wherein the K-length path matrix represents the state of a path of length K between two nodes in a directed graph associated with the matrix, wherein K≤p-1, and p represents the number of the temperature variable, the humidity variable, and the variable regarding whether the control system is abnormal; and The path matrix is ​​determined based on the K-length path matrix.

21. The system of claim 19, wherein determining the edge constraint matrix comprises: determining a side knowledge matrix describing a state of a direct causal relationship between two variables among the temperature variable, the humidity variable, and a variable regarding whether the control system is abnormal based on the expert knowledge; as well as The edge constraint matrix is ​​determined based on the edge knowledge matrix.

22. The system of claim 21, wherein said determining said edge knowledge matrix comprises: For the elements in the edge knowledge matrix, In response to the expert knowledge indicating that there is a direct causal relationship between a first variable and a second variable associated with an element in the side knowledge matrix, setting the element in the side knowledge matrix to satisfy a first condition; In response to the expert knowledge indicating that there is no direct causal relationship between the first variable and the second variable, setting an element in the edge knowledge matrix to satisfy a second condition; as well as In response to the expert knowledge not having knowledge of a direct causal relationship between the first variable and the second variable, an element in the side knowledge matrix is ​​set to satisfy a third condition.

23. The system of claim 21, wherein said determining said path knowledge matrix comprises: For the elements in the path knowledge matrix, In response to the expert knowledge indicating that there is an indirect causal relationship between a first variable and a second variable associated with an element in the path knowledge matrix, setting the element in the path knowledge matrix to satisfy a first condition; In response to the expert knowledge indicating that there is no indirect causal relationship between the first variable and the second variable, setting an element in the path knowledge matrix to satisfy a second condition; as well as In response to the expert knowledge not having knowledge of an indirect causal relationship between the first variable and the second variable, an element in the path knowledge matrix is ​​set to satisfy a third condition.

24. The system of claim 21, wherein the determining the edge constraint matrix based on the edge knowledge matrix comprises: For the elements in the edge constraint matrix, In response to an element in the edge knowledge matrix corresponding to an element in the edge constraint matrix being equal to 0, setting the element in the edge constraint matrix to a larger value; In response to an element in the edge knowledge matrix corresponding to an element in the edge constraint matrix satisfying a first condition, setting the element in the edge constraint matrix to zero; as well as In response to an element in the edge knowledge matrix corresponding to an element in the edge constraint matrix satisfying a third condition, the element in the edge constraint matrix is ​​set to be associated with a predefined hyperparameter.

25. The system of claim 24, wherein setting the elements in the edge constraint matrix to be associated with predefined hyperparameters comprises: Set the element at position (j, i) in the edge constraint matrix Set to in is a predefined hyperparameter calculated based on the data set, ols represents a least squares operation, || represents an absolute value operation, and γ represents a predetermined parameter.

26. The system of claim 21, wherein determining the path constraint matrix comprises: determining a path knowledge matrix describing a state of an indirect causal relationship between two variables among the temperature variable, the humidity variable, and a variable regarding whether the control system is abnormal based on the expert knowledge; as well as The path constraint matrix is ​​determined based on the path knowledge matrix.

27. The system of claim 26, wherein the determining the path constraint matrix based on the path knowledge matrix comprises: For the elements in the path constraint matrix, In response to an element in the path knowledge matrix corresponding to an element in the path constraint matrix being equal to 0, setting the element in the path constraint matrix to a larger value; In response to an element in the path knowledge matrix corresponding to an element in the path constraint matrix satisfying a first condition, setting the element in the path constraint matrix to zero; as well as In response to an element in the path knowledge matrix corresponding to an element in the path constraint matrix satisfying a third condition, setting the element in the path constraint matrix to 1.

28. The system of claim 27, wherein the expert knowledge constraint comprises a constraint corresponding to the current vector β in the matrix j The associated edge constraint vector and path constraint vector, where: The edge constraint vector is And the path constraint vector is 29. The system of claim 28, wherein constructing the problem formulation describing the cause-effect relationship comprises: For the current vector β j Construct the problem formula vector: Where Supp(β j ) represents the current vector β j Support, Q s represents a causal sequence associated with at least some of the elements, x j Represents multiple samples with the current vector β j The associated part, x -j Represents multiple samples with the current vector β j The vectors outside the associated part, represents the two-norm operation, represents the element at position (j, i) in the edge constraint matrix, β ji represents the element at position (j, i) in the matrix, represents the element at position (j, i) in the path constraint matrix, p υi represents the element at position (j, i) in the path matrix, β jυ represents the element at the position (j, v) in the matrix, || represents the absolute value operation, Indicates Q s The node after node i in .

30. The system of claim 29, wherein determining the element value in the matrix of the causal relationship comprises: Solve the problem formula vector to obtain the current vector β for the matrix j Candidate results.

31. A computer program product tangibly stored on a non-transitory computer readable medium and comprising machine executable instructions which, when executed, cause a machine to perform the steps of the method according to any one of claims 1 to 15.