A method for handling categorical relationships in dynamic uncertain causal graphs

Abstract

The application provides a computer readable storage medium, characterized in that the storage medium stores a computer program, and the computer program can execute the following processing when executed: a construction and reasoning method of an intelligent system for information of uncertain causality and classification relationship, which adds an expression method of a classification variable C i and a reasoning method thereof on the basis of an existing DUCG technical solution, deletes C i and an input directed arc thereof during a reasoning process, directly connects an output directed arc thereof with V k , that is, takes V k as a parent variable of X n , at this time, F n；i becomes F n；k , and the embedded parameter value is unchanged, then the existing DUCG reasoning method is used for reasoning calculation to obtain a reasoning result of the DUCG. Thus, the existing DUCG cannot express the classification relationship, and the expression error (which may lead to reasoning calculation error) or inconvenience problem is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the intelligent information processing technology, and is a further expansion of the technical solutions described in the granted patents "A construction method of an intelligent system processing uncertain causal relationship information" (Patent No. ZL 2006 80055266.X), "METHOD FOR CONSTRUCTING AN INLLIGENT SYSTEM PROCESSING UNCERTAIN CAUSAL RELATIONSHIP INFORMATION" (Patent No. US 8,255,353 B2), "A method for heuristically detecting abnormal reasons of a system based on a dynamic uncertain causality graph" (Patent No. ZL 2016 10282052.1) and "A construction method of an extended intelligent system processing uncertain causal relationship information" (Patent No. ZL 201710967729.X). Using the technical solutions proposed in the present application, through computer processing, the causal knowledge expression and application capability of the dynamic uncertain causality graph (DUCG) can be further improved, making it more meet the actual needs and more accurately diagnose the reasons for the abnormality of the object system, so as to facilitate people to take effective measures to restore the object system to normal as soon as possible in the case of abnormality. BACKGROUND

[0002] As described in the granted patents "A construction method of an intelligent system processing uncertain causal relationship information", "METHOD FOR CONSTRUCTING AN INLLIGENT SYSTEM PROCESSING UNCERTAIN CAUSAL RELATIONSHIP INFORMATION", "A method for heuristically detecting abnormal reasons of a system based on a dynamic uncertain causality graph" and "A construction method of an extended intelligent system processing uncertain causal relationship information", there are a large number of cause events leading to system abnormalities in industrial systems, social systems and biological systems (referred to as object systems), such as coil short circuit, pump failure, component failure, subsystem failure, conduction path blockage, foreign matter entering, variation, necrosis, pollution, infection, injury or natural failure of certain organizations or organisms, etc., which are referred to as event variables and can be represented by B k or BX k , where k is the event variable label, B kj or BX kj is the j state of the variable B k or BX k . B k , B kj , BXk and BX kj may be represented by the graphic symbols or or or or respectively, B k and BX k are different in that B k is a root cause variable without input, while BX k has input and can be affected by other factors, BX k is B k after being affected. Usually j=0 means B k or BX k is in normal state; j=1, 2, 3... means B k or BX k is in abnormal state.

[0003] If there is only one number in the graphic, the number represents the variable label, and the variable state is unknown. For convenience, kj can be separated by commas (the same below).

[0004] The states of B k and BX k cannot be directly detected in most cases, or are difficult to be directly detected, and are variables that the DUCG intelligent system needs to infer whether they are in abnormal state.

[0005] In addition, there are a large number of variables in the system that have certain or uncertain causal relationships with B k or BX k , such as temperature, pressure, flow, speed, frequency, switch state, various test or physical test results, survey results, imaging results, feelings, symptoms, signs, regions, time, environment, season, religion, skin color, experience, blood relationship, hobby, character, living conditions, working conditions, etc. These are called intermediate or result variables, which can be represented by X y , y=0, 1, 2,.... X yg is the g state of X y , g=0, 1, 2,.... For convenience, g=0 usually means X y is in normal state, g≠0 means X y is in abnormal state with label g. X variable has at least one input (cause) variable, and can have or not have output (result) variable. X y and X yg may be represented by the graphic symbols or or indicates.

[0006] With the DUCG technical solution, one can acquire evidence E by detecting the state of X type variable to deduce the cause B of system abnormality kj or BX kj (j≠0), so as to take effective measures in time to make the system return to normal. E is composed of the known state X yg of at least one X variable, for example, E=X 1,2 X 2,3 X 3,1 X 4,0 X 5,0 .

[0007] The DUCG intelligent reasoning is to solve Pr{H kj |E} = Pr{H kj E} / Pr{E}, wherein H kj is a hypothesis event, usually a variable state combination to be solved in DUCG, for example, H 1,2 =B 1,2 , H 2,1 =BX 2,1 , etc. The subscript k in H kj indicates the variable combination, for example, H1=B1, H2=BX2, etc. The subscript j in H kj indicates the state of the variable in H k , for example, H 1,2 =B 1,2 , H 2,1 =BX 2,1 , etc. The set of all possible hypothesis events H kj under the condition of E is denoted by S H , so H kj ∈S H .

[0008] The following variables are also defined in DUCG:

[0009] A logic gate variable, denoted by G i , has at least two input variables and one output variable. G ij denotes the j state of G i . G i is used to express the logical combination of various states of interest of the input variables, which is described by a logic gate specification table LGS i . For example, G1 is described by LGS1: G 1,1 =B 3,1 X 1,1 , G 11,2 =B 3,1 X 1,2 , G 1,0 =others (remaining states), etc. Gij G ij’ = 0 (empty set, where j ≠ j'), meaning that different states of G are mutually exclusive. As shown in the attached figure of the aforementioned patent document, G... i and G ij Graphic symbols can be used respectively or or Representation. The input variables of a logic gate are represented by a directed arc. Connects to logic gates.

[0010] The default cause variable can be represented by D, and it is one of the cause variables of the corresponding X variable. For example, D4 ​​is the default cause variable of X4. Pr{D i}≡1. D i Available graphics or express.

[0011] Variables B, X, BX, D, and G can be called nodes. The physical meaning of the variables themselves and their various states can be defined according to the object being described. When B, X, BX, D, and G are direct cause variables, they are called parent variables, which can be uniformly expressed by V, where V∈{B, X, BX, D, G}, and the subscripts remain unchanged. For example, V2=X2, V 3,2 =B 3,2 And so on. The result variable can only be either X or BX. A state of a variable is an event. For example, X yg B kj BX kj G ij H kj or V ij These are all events.

[0012] Action variable F n；i It expresses the parent variable V i With sub-variable X n Or BX n The causal relationship matrix can be represented by a directed arc → or other graphical symbols, pointing from cause to effect. The elements in the matrix are called the acting events F. nk；ij Expressing the parent event V ij With sub-event X nk Or BX nk The causal relationship between them. nk；ij ≡(r n；i / r n A nk；ij Where r n；i >0 represents the parent variable V i With sub-variable X n Or BX n The strength of the causal relationship between them Ank；ij For V ij Cause X nk When this random event occurs, a nk；ij ≡Pr{A nk；ij},satisfy f nk；ij ≡Pr{F nk；ij}≡(r n；i / r n )a nk；ij f nk；ij It is V ij For X nk The contribution value of the weighted probability satisfies v ij =Pr{V ij}, V ij and v ij These are the event vectors V i and parameter vector v i The elements. When the cause variable is D. i At that time, F nk；ij ≡F nk；iD At this point, j = D. The expression for other causal variables is similar.

[0013] F nk；ij It can be a conditional event, represented by a dashed directed arc. Indicates that a conditional event expresses its causal event V. ij With outcome event X nk Or BX nk The relationship between them is one of conditional action, that is, based on the conditional event Z. nk；ij Determine F based on whether it meets the requirements. nk；ij Whether it holds true. For example, Z. nk；ij =X 1,2 When X 1,2 When Z is true, nk；ij Satisfy, F nk；ij Established; when X 1,2 When not true, Z nk；ij Not satisfied, F nk；ij This is not true. When the conditional causal relationship between the same pair of parent and child variables is supported by the same conditional event, this conditional event is uniformly represented by Z. n；i To represent, for example, Z n；i =X 1,2 When X 1,2 When Z is true, n；i Satisfy, F n；i Establishment Become →; When X 1,2 When not true, Z n；i Not satisfied, F n；i Not valid It has been deleted.

[0014] For convenience, the full set is denoted as 1 and the empty set is denoted as 0. The above variables and their states can also be represented by other figures or symbols.

[0015] In addition, SX variables and events can also be defined, which can be represented by figures or or SG variables and events can be defined, which can be represented by figures or or RG variables can be defined, which can be represented by figures or or The relevant explanations can be found in documents [1]-

[17] .

[0016] Figure 1 is an example of the existing DUCG.

[0017] However, in practical applications, the above expression method can only express the causal relationship, and cannot express the classification relationship, resulting in inconvenience and even errors in expression. That is, the occurrence of a certain cause event will lead to a number of possible results if a certain experiment or operation is performed. Here, the experiment or operation itself is not the result of the cause event, only the results produced by the experiment or operation are the results of the cause event, and the experiment or operation is only a way or path to obtain the result, and different experiments or operations will produce different categories of results, so the experiment or operation plays a classification role. If we regard the experiment or operation itself as the result of the cause event, an error will be generated; if we directly express the causal relationship between the cause event and various experiment or operation results, the logic and hierarchy will be unclear, and the expression will be difficult to understand.

[0018] Therefore, the present application proposes a method of introducing classification variables into DUCG to express and reason the classification relationship to solve this problem.

[0019] The present application has the following technical references:

[0020] [1] Chinese invention patent: "A construction method of an intelligent system for processing uncertain causal relationship information", patent number: ZL 2006 8 0055266.X; authorized date: April 14, 2010.

[0021] [2] US invention patent: "Method for constructing an intelligent system processing uncertain causal relationship information"; patent number: US 8255353B2; granted date: August 28, 2012.

[0022] [3] Chinese invention patent: "A method for constructing a three-dimensional DUCG intelligent system for dynamic fault diagnosis"; patent number: ZL 2013 10718596.4; granted date: April 15, 2015.

[0023] [4] Chinese invention patent: "A method for heuristically detecting system abnormal reasons based on dynamic uncertain causality graph"; patent number: ZL 2016 10282052.1; granted date: June 5, 2017.

[0024] [5] Chinese invention patent: "A method for constructing an extended intelligent system for processing uncertain causal relationship information"; patent number: ZL 2017 10967729.X; granted date: October 517, 2017.

[0025] [6] Chinese invention patent: "An efficient method for hierarchical recursive reasoning of DUCG"; patent number: ZL 2018 10455192.3; granted date: May 13, 2018.

[0026] [7] Chunling Dong, Qin Zhang, Shichao Geng. A modeling and probabilistic reasoning method of dynamic uncertain causality graph for industrial fault diagnosis. International Journal of Automation and Computing, 11(3) (2014) 288-298.

[0027] [8] Dong Chunling, Zhang Qin, "Research on Weighted Logic Reasoning Algorithm for Uncertain Fault Diagnosis", "Journal of Automation", Vol. 40, No. 12, pp. 2766-2781, 2014.

[0028] [9] Shao-rui Hao, Shi-chao Geng, Lin-xiao Fan, Jia-jia Chen1, Qin Zhang, Lan-juan Li. "Intelligent diagnosis of jaundice with dynamic uncertain causality graph model", Journal of Zhejiang University-SCIENCE B (Biomedicine & Biotechnology), vol. 18, no. 5, pp. 393-401, 2017.

[0029]

[10] Zhenxu Zhou, Qin Zhang. "Model event / fault trees with dynamic uncertain causality graph for better probabilistic safety assessment," IEEE Trans. Reliability, vol. 66, no. 1, pp 178-188, 2017.

[0030]

[11] Q. Zhang & Q. Yao, "Dynamic Uncertain Causality Graph for Knowledge Representation and Reasoning: Utilization of Statistical Data and Domain Knowledge in Complex Cases," IEEE Trans. Neural Networks and Learning Systems, vol. 29, no. 5, pp. 1637-1651, 2018.

[0031]

[12] Q. Zhang, K. Qiu, Z. Zhang, "Calculate joint probability distribution of steady directed cyclic graph with local data and domain casual knowledge," China Communications, pp. 146-155, 2018.

[0032]

[13] C. Dong, Q. Zhang. “The Cubic Dynamic Uncertain Causality Graph: A Methodology for Temporal Process Modeling and Diagnostic Logic Inference,” IEEE Trans. Neural Networks and Learning Systems, vol. 31, no. 10, pp. 4239-4253, 2020.

[0033]

[14] Xusong Bu, Lin Lu, Zhan Zhang, Qin Zhang and Yan Zhu. “A General Outpatient Triage System Based on Dynamic Uncertain Causality Graph,” IEEE Access, vol. 8, pp. 93249-93263, 2020.

[0034]

[15] Dongping Ning, Zhan Zhang, Kun Qiu, Lin Lu, Qin Zhang, Yan Zhu & Renzhi Wang. “Efficacy of intelligent diagnosis with a dynamic uncertain causality graph model for rare disorders of sex development,” Frontiers of Medicine, vol. 14, no. 4, pp. 498-505, 2020.

[0035]

[16] Yang Jiao, Zhan Zhang, Ting Zhang, Wen shi, Yan Zhu and Jie Hu. “Development of an artificial intelligence diagnostic model based on dynamic uncertain causality graph for the differential diagnosis of dyspnea,” Frontiers of Medicine, vol. 14, pp. 488-497, 2020.

[0036]

[17] Q. Zhang, X, Bu, Z. Zhang, M. Zhang and J. Hu,“Dynamic uncertain causality graph for computer-aided general clinical diagnoses with nasal obstruction as an illustration,” Artificial Intelligence Review, vol. 54, pp. 27-61, 2021.

[0037]

[18] Hao Nie & Qin Zhang. “A New Inference Algorithm of Dynamic Uncertain Causality Graph Based on Conditional Sampling Method for Complex Cases,” IEEE Access, DOI: 10.1109 / ACCESS.2021.3093205, 2021. SUMMARY

[0038] The present invention discloses a technical solution which further extends the technical solution disclosed in the granted Chinese invention patents ZL 2006 80055266.X, ZL 2016 10282052.1, ZL 2017 10967729.X and the US invention patent US 8255353 B2, so that DUCG can also handle categorical relationships.

[0039] The existing DUCG technical solution on which the present invention is based is:

[0040] Root cause event variable is denoted by B k or BX k , k is event variable label, B kj or BX kj is variable B k or BX k 's j state. As shown in Figure 1 B k , B kj , BX k and BX kj may be represented by graphs or or or or , Bk and BX k The difference lies in B k The root cause variable has no input, while BX... k Input is available, but it can be affected by other factors, BX k B after being affected k Usually, j=0 indicates B. k Or BX k In normal state; j = 1, 2, 3... represents B k Or BX k The variable is in various abnormal states. If there is only one number in the graph, that number represents the variable label, and the variable state is unknown. If there are two numbers in the graph, the first number represents the variable label, and the second number represents the variable state. For convenience, the two numbers can be separated by a comma (the same applies below).

[0041] B k and BX k The state of these variables cannot be directly detected in most cases, or is difficult to detect directly. These variables are the ones that DUCG needs to infer whether they are in an abnormal state.

[0042] In addition, there are a large number of B-related issues in the system. k Or BX k Variables with definite or indefinite causal relationships, such as temperature, pressure, flow rate, speed, frequency, on / off status, results of various laboratory or physical tests, survey results, imaging results, sensations, symptoms, signs, region, time, environment, season, religion, skin color, experience, blood relationship, hobbies, personality, living conditions, working conditions, etc., are called intermediate or outcome variables, and can be represented by X. y This indicates that y = 0, 1, 2, ... X yg For X y The g-state, g = 0, 1, 2, ... . Usually g = 0 represents X. y In the normal state, g≠0 indicates X y The variable X is in an abnormal state labeled g. X has at least one input (cause) variable and may or may not have an output (result) variable. y and X yg Graphic symbols can be used respectively or or express.

[0043] Using the DUCG technique, one can obtain evidence E by detecting the state of a variable of type X, in order to infer the cause B of the system anomaly. kj Or BX kj (j≠0), thus enabling timely and effective measures to restore the system to normal. E is determined by the known state X of at least one variable X. ygFor example, E = X 1,2 X 2,3 X 3,1 X 4,0 X 5,0 .

[0044] DUCG inference computation is to solve Pr{H kj |E} = Pr{H kj E} / Pr{E}, where H kj is a hypothesis event, usually a variable state combination to be solved in DUCG, such as H 1,2 = B 1,2 , H 2,1 = BX 2,1 , etc. The subscript k in H kj identifies the variable, such as H1= B1, H2= BX2, etc. The subscript j in H kj identifies the state of the variable in H k , such as H 1,2 = B 1,2 , H 2,1 = BX 2,1 , etc. The set of all possible hypothesis events H kj under the condition of E is denoted by S H , then H kj ∈ S H .

[0045] In DUCG, the following variables are also defined:

[0046] A logic gate variable, denoted by G i , has at least two input variables and one output variable. G ij denotes the j state of G i . G i is used to express the logical combination of various states of interest of the input variables, which is described by a logic gate specification table LGS i . For example, G1 is described by LGS1: G 1,1 = B 3,1 X 1,1 , G 11,2 = B 3,1 X 1,2 , G 1,0 = others (remaining states), etc. G ij G ij’ = 0 (empty set, where j≠j’), that is, the different states of G are mutually exclusive. As shown in the above patent document drawings, G i and G ij can be represented by graphical symbols or or Directed arc The input variable is connected with G i .

[0047] The default cause variable, denoted by D, is one of the cause variables of the corresponding X variable. For example, D4 is the default cause variable of X4. Pr{D i}≡1. D i may be represented by a graph or .

[0048] The B, X, BX, D and G variables can be called nodes, and the physical meaning of the variable itself and its various states can be defined according to the described object. When B, X, BX, D and G are direct cause variables, they are called parent variables, which can be expressed uniformly as V, V∈{B, X, BX, D, G}, and the subscript is unchanged. For example, V2=X2, V 3,2 =B 3,2 , and so on. The result variable can only be an X or BX variable. A state of a variable is an event. For example, X yg , B kj , BX kj , G ij , H kj or V ij , etc. are all events.

[0049] The action variable F n；i is a matrix expressing the causal relationship between the parent variable V i and the child variable X n or BX n , which can be represented by a directed arc→or other graphical symbols, with the cause pointing to the result. The element in the matrix is called an action event F nk；ij , which expresses the causal relationship between the parent event V ij and the child event X nk or BX nk , F nk；ij ≡(r n；i / r n )A nk；ij . Where r n；i >0 is the causal relationship strength between the parent variable V i and the child variable X n , A nk；ij is the random event caused by V ij to X nk or BX nk , a nk；ij ≡Pr{A nk；ij}, and satisfies f nk；ij ≡Pr{F nk；ij}≡(r n；i / rn )a nk；ij f nk；ij It is V ij For X nk The contribution value of the weighted probability satisfies v ij =Pr{V ij}, V ij and v ij These are the event vectors V i and parameter vector v i The elements. When the cause variable is D. i At that time, F nk；ij ≡F nk；iD At this point, j = D. The expression for other causal variables is similar.

[0050] F nk；ij It can be a conditional event, represented by a dashed directed arc. Indicates that a conditional event expresses its causal event V. ij With outcome event X nk Or BX nk The relationship between them is one of conditional action, that is, based on the conditional event Z. nk；ij Determine F based on whether it meets the requirements. nk；ij Whether it holds true. For example, Z. nk；ij =X 1,2 When X 1,2 When Z is true, nk；ij Satisfy, F nk；ij Established; when X 1,2 When not true, Z nk；ij Not satisfied, F nk；ij This is not true. When the conditional causal relationship between the same pair of parent and child variables is supported by the same conditional event, this conditional event is uniformly represented by Z. n；i To represent, for example, Z n；i =X 1,2 When X 1,2 When Z is true, n；i Satisfy, F n；i Establishment Become →; When X 1,2 When not true, Z n；i Not satisfied, F n；i Not valid It has been deleted.

[0051] For convenience, the universal set is denoted as 1, and the empty set is denoted as 0. The above variables and their states can also be represented by other graphics or symbols.

[0052] In addition, SX type variables and events are defined, which can be used graphically respectively. or or This indicates that SG variables and events are defined, and can be represented graphically. or or This indicates that the RG variable is defined, and can be represented graphically. or or This invention adds a method for processing classification relationships to the above technical solution. Specifically,

[0053] 1. This invention provides a computer-readable storage medium storing a computer program. When executed, the computer program performs a method for constructing and reasoning an intelligent system that processes information involving uncertain causal and classification relationships. This method, based on the existing DUCG technology solution, adds a classification variable C. i The methods of expression and reasoning, wherein the characteristics of the methods of expression are: (1) C i The formal parent variable is the cause variable V. k V∈{B、BX、X、G},C i With V k The number of states is the same for C i There is only one parent variable; (2)C i The formal sub-variables are several X n SX n , and / or SG n (X will be used only below) n (representative), these X n Actually, it's V k The sub-variables, but in form C i Sub-variables; (3)V k With C i They are connected by directed arcs, by V k Pointing to C i (4)C i With X n The directed arc F between them n；i Connected, by C i Point to X n The parameter a embedded within it nj；iy It expresses X nj With V ky Uncertain causal relationship between them, parameter r n；i It expresses X n With V k The strength of the causal relationship between them; the characteristic of the reasoning method is: (5) in the reasoning process, C i The directed arcs input to it are deleted, and its directed arcs output are directly connected to V. kConnected, i.e. k As the parent variable of X n , F n；i becomes F n；k , and the parameter values embedded in it remain unchanged, and then the inference calculation is carried out according to the existing DUCG inference method to obtain the inference result of DUCG.

[0054] 2. In 1(3), the directed arc between V k and C i is defined as F i；k = I i；k , I i；k is a unit matrix whose row number and column number are equal to the state number of V k and C i , and instead of the inference method of 1(5), C i is regarded as a causal variable, I k；i is regarded as a causal action variable F i；k , and the inference calculation is carried out as the same as other causal variables and causal action variables.

[0055] 3. As described in 1, when there are multiple C i that share a parent variable V k and a child variable X n , these C i , i∈S C , are merged into one C g , and where S C is the subscript set of these C i .

[0056] The technical solution of the present application solves the existing DUCG problem of easy expression error (may cause inference calculation error) or inconvenience due to the inability to express classification relationship. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 is an example of the existing DUCG;

[0058] Figure 2 is the DUCG expression of insulation oil oxidation and dampness (disease 1 and disease 2) and observation of extracted insulation oil (blood routine and urine routine examination) and its results;

[0059] Figure 3 represents the expression after C i and its input directed arc are deleted;

[0060] Figure 4 represents an example of the repeated path of C variable. DETAILED DESCRIPTION

[0061] This invention provides a computer-readable storage medium storing a computer program. When executed, the computer program performs a method for constructing and reasoning an intelligent system that processes information involving uncertain causal and classification relationships. This method, based on the existing DUCG technology, adds a classification variable C. i The methods of expression and reasoning.

[0062] The following description, with reference to the accompanying drawings, explains the additional classification variable C in this invention. i The specific implementation methods of the expression and reasoning methods.

[0063] Example 1

[0064] Figure 2 DUCG expression was observed in insulating oil oxidation and moisture absorption (disease 1 and disease 2) and the results of extracted insulating oil (complete blood and urine tests).

[0065] like Figure 2 As shown, let variable B1 represent whether the transformer insulating oil is oxidized (or whether it suffers from disease 1), and let B1 represent the absence of oxidation (no disease 1). 1,0 Oxidation (with disease 1) is B. 1,1 Its probability of occurrence is b 1,1 =0.003; Variable B2 represents whether the transformer is damp (or whether it has disease 2), and B is not damp (no disease 2). 2,0 If it is damp (or has disease 2), it is classified as B. 2,1 b 2,1 = 0.002. Here, B1 and B2 are the causal variables V. k That is, V = B, k = 1, 2. The method for detecting whether transformer insulating oil is oxidized and damp (whether it suffers from diseases 1 and 2) is to extract the insulating oil for laboratory testing (complete blood count and urinalysis). When oxidation (disease 1) is present, there is a 60% probability that the insulating oil will appear cloudy (abnormal white blood cells), which can be detected using X-ray dilution. 3,1 It means (X) 3,0 This indicates no turbidity (normal white blood cell count); there is an 80% probability that the insulating oil will appear light brown (abnormal uric acid), indicated by X-ray. 4,1 It means (X) 4,0 This indicates normal color (normal uric acid). When damp (Disease 2), X can be observed with a 90% probability. 3,1 X has a 40% probability of being observed. 4,1 Clearly, sampling insulating oil for observation (performing blood and urine tests) is not B. 1,1 and B 2,1 The result is not to obtain B. 1,1 and B 2,1 Result X 3,1 and X4,1 The method or path.

[0066] The classification variable C added according to this invention i The expression method and its reasoning method, namely, (1)C i The formal parent variable is the cause variable V. k V∈{B、BX、X、G},C i With V k The number of states is the same for C i There is only one parent variable; (2)C i The formal sub-variables are several X n SX n , and / or SG n (X will be used only below) n (representative), these X n Actually, it's V k The sub-variables, but in form C i Sub-variables; (3)V k With C i They are connected by directed arcs, by V k Pointing to C i (4)C i With X n The directed arc F between them n；i Connected, by C i Point to X n The parameter a embedded within it nj；iy It expresses X nj With V ky Uncertain causal relationship between them, parameter r n；i It expresses X n With V k The strength of the causal relationship between them; (5) In the reasoning process, C i The directed arcs input to it are deleted, and its directed arcs output are directly connected to V. k Connected, that is, V k As X n The parent variable, F at this time n；i Become F n；k The embedded parameter values ​​remain unchanged, allowing the construction of DUCGs containing categorical variables, such as... Figure 2 As shown in the figure, C1 and C2 are categorical variables for whether insulating oil was extracted for observation (blood routine and urine routine examination), which are represented graphically. and graphics This indicates that the descriptions of C1 and C2 may be the same or different.

[0067] According to C i The formal parent variable is the cause variable V. k V∈{B、BX、X、G},Ci With V k The number of states is the same for C i In the reverse form with only one parent variable, C1 and B1 have the same number of states, and C2 and B2 have the same number of states, both being states 0 and 1. For the transformer, C... 1,1 and C 2,1 Both indicate that insulating oil was extracted for observation, C 1,0 and C 2,0 Both indicate that the insulating oil is not extracted for observation. Note that although both C1 and C2 indicate whether the insulating oil is extracted for observation, according to 1(1), each C variable can only have one parent variable, so two C variables must be used to express this.

[0068] According to C i With X n The directed arc F between them n；i Connected, by C i Point to X n The parameter a embedded within it nj；iy It expresses X nj With V ky Uncertain causal relationship between them, parameter r n；i It expresses X n With V k The strength of the causal relationship between them, we have a 3,1；1,1 =0.6, a 3,1；2,1 =0.9, a 4,1；1,1 =0.8, a 4,1；2,1 =0.4. Since the causal relationship between B1, B2 and X3 and X4 is certain, we have r 3；1 =r 4；1 =r 3；2 =r 4；2 =1. By definition, r3 = r 3；1 +r 3；2 =2, r4=r 4；1 +r 4；2 =2, r 3；1 / r3=r 3；2 / r3=r 4；1 / r4=r 4；2 / r4 = 1 / 2.

[0069] Suppose the observed evidence is E=X 3,1 X 4,1 Find B 1,1 and B 2,1 The probability of occurrence.

[0070] According to the reasoning method of this invention, that is, C i The directed arcs input to it are deleted, and its directed arcs output are directly connected to V. kConnected, that is, V k As X n The parent variable, F at this time n；i Become F n；k The embedded parameter values ​​remain unchanged. Figure 2 become Figure 3 .

[0071] exist Figure 3 middle, Figure 2 In the code, C1 and C2 and their input directed arcs are removed, and the output directed arcs are directly connected to the parent variables B1 and B2 of C1 and C2.

[0072] According to the reasoning method of the present invention, C i The directed arcs input to it are deleted, and its directed arcs output are directly connected to V. k Connected, that is, V k As X n The parent variable, F 3,1；1,1 F 4,1；1,1 F 3,1；2,1 and F 4,1；2,1 The parameters remain unchanged. Based on Figure 3 We perform the following reasoning and calculations according to DUCG's existing methods:

[0073]

[0074] in,

[0075]

[0076]

[0077] Obtain Pr{B 1,1 |E} is the inference of the probability of oxidation (disease 1) under the known evidence E; obtaining Pr{B 2,1 |E} infers the probability of dampness (disease 2) under the known evidence E. This inference is obtained by adding categorical variables C1 and C2 to DUCG, avoiding the expression errors (potentially leading to inference calculation errors) or inconveniences that were prone to occur in the previous DUCG due to its inability to express categorical relationships.

[0078] Example 2

[0079] This example represents another aspect of the invention, in which, in the example of insulating oil shown in Example 1, V... k With C i The directed arc between them is defined as F i；k =I i；k I i；k It is an identity matrix, whose number of rows and columns is equal to V. k With C iThe number of states, but instead of using the reasoning method described above, is C. i Treat I as a causal variable k；i Treat F as a causal variable i；k It is treated the same as other causal variables and causal effect variables in reasoning and calculation, and the input directed arcs of C1 and C2 are given as matrix I with 2 rows and 2 columns, that is... Other parameters remain the same as in Example 1.

[0080] based on Figure 2 And according to the reasoning algorithm that treats all variables as causal variables, we have:

[0081] By comparison, we can see that (8) is exactly the same as (3). Therefore, we can still get the same result as (6) and (7).

[0082] In this example, the reasoning method of causal relationship is retained in form, that is, the variable C is also treated as a causal variable. Given the I matrix, the same result can be obtained.

[0083] Example 3

[0084] In this example, for the insulating oil shown in Example 1, the observation of extracted insulating oil is subdivided into two types: visual observation and microscopic observation. C5 represents visual observation of extracted insulating oil, C6 represents microscopic observation of extracted insulating oil, and X3 still indicates whether the insulating oil is cloudy. That is, both observations can determine whether the insulating oil is cloudy, but the probabilities are different. Let the former be... The latter is Furthermore, since the causal relationship is established, r 3；5 =r 3；6 =1. Everything else is the same as in Example 1.

[0085] Figure 4 An example of a C variable repeating a path.

[0086] According to the existence of multiple C i There is one parent variable V k and a child variable X n At that time, these C i , i∈S C merged into one C g ,and Where S C For these C i The method for processing the subscript set is as follows: C5 and C6 are merged into C1, such as... Figure 2 As shown, Figure 2 F in 3；1 equal:

[0087] F3；1 = F 3；5 + F 3；6 (9)

[0089] i.e.

[0090] f 3；1 = f 3；5 + f 3；6 (10)

[0092] According to the given parameters, we have:

[0093]

[0094] i.e. Figure 2 r 3；1 = 1, based on Figure 2 , we can make the same reasoning and calculation as in Example 1 or Example 2.

[0095] When C5 and C6 share multiple sub-variables, similar operations are performed on each sub-variable as in this example, and then the reasoning and calculation are performed based on the DUCG graph after merging the C variables in the same way as described above.

Claims

1. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed, can execute the construction and reasoning method of the intelligent system for processing uncertain cause-effect relationship and classification relationship information, which adds the expression method of classification variable C i and the reasoning method thereof on the basis of the existing DUCG technical solution. wherein the method of expressing comprises: (1) C i The formal parent variable of V is C k , V ∈ {B, BX, X, G}, C i has the same number of states as V k ; C i has only one parent variable; (2) C i The formal sub-variables are several X n SX n , and / or SG n The following uses only X n Representatives, these X n Actually, it's V k The sub-variables, but in form C i Sub-variables; (3) V k with C i are connected by a directed arc from V k to C i ; (4) C i is connected with X n by a directed arc F n;i , with C i pointing to X n , the parameter a nj;iy embedded in it represents the uncertain causal relationship between X nj and V ky , and the parameter r n;i represents the causal relationship strength between X n and V k ; wherein the method of reasoning comprises: (5) In the reasoning process, C i and its input directed arc are deleted, and its output directed arc is directly connected with V k , i.e. V k is the parent variable of X n , at this time F n;i becomes F n;k , the embedded parameter value is unchanged, and then the reasoning calculation is carried out according to the existing DUCG reasoning method to obtain the reasoning result of DUCG.

2. The computer-readable storage medium of claim 1, wherein: In the above (3), the directed arc between V k and C i is defined as F i;k = I i;k , I i;k is an identity matrix whose row number and column number are equal to the state number of V k and C i , instead of the inference method of the above (5), C i is regarded as a causal variable, I k;i is regarded as a causal action variable F i;k , and inference calculation is performed as if it is the same as other causal variables and causal action variables.

3. The computer-readable storage medium of claim 1, wherein: When there are multiple C i There is one parent variable V k and a child variable X n At that time, these C i , i∈S C merged into one C g ,and S C For these C i The set of subscripts.

Images