WO2017185449A1 - 一种基于动态不确定因果图的启发式检测系统异常原因的方法 - Google Patents

一种基于动态不确定因果图的启发式检测系统异常原因的方法 Download PDF

Info

Publication number
WO2017185449A1
WO2017185449A1 PCT/CN2016/083906 CN2016083906W WO2017185449A1 WO 2017185449 A1 WO2017185449 A1 WO 2017185449A1 CN 2016083906 W CN2016083906 W CN 2016083906W WO 2017185449 A1 WO2017185449 A1 WO 2017185449A1
Authority
WO
WIPO (PCT)
Prior art keywords
variable
state
ducg
detected
determining
Prior art date
Application number
PCT/CN2016/083906
Other languages
English (en)
French (fr)
Inventor
张勤
Original Assignee
张湛
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 张湛 filed Critical 张湛
Priority to US16/096,675 priority Critical patent/US11397741B2/en
Publication of WO2017185449A1 publication Critical patent/WO2017185449A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24Querying
    • G06F16/245Query processing
    • G06F16/2457Query processing with adaptation to user needs
    • G06F16/24578Query processing with adaptation to user needs using ranking
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00Computing arrangements based on specific mathematical models
    • G06N7/02Computing arrangements based on specific mathematical models using fuzzy logic
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90Details of database functions independent of the retrieved data types
    • G06F16/901Indexing; Data structures therefor; Storage structures
    • G06F16/9024Graphs; Linked lists
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24Querying
    • G06F16/245Query processing
    • G06F16/2458Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2462Approximate or statistical queries
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10Complex mathematical operations
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10Complex mathematical operations
    • G06F17/11Complex mathematical operations for solving equations, e.g. nonlinear equations, general mathematical optimization problems
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00Computing arrangements based on specific mathematical models
    • G06N7/01Probabilistic graphical models, e.g. probabilistic networks

Definitions

  • the invention relates to an intelligent information processing technology, and the processing object is the information of the uncertain causal relationship expressed in the dynamic uncertainty causal map DUCG, and uses the technical solution proposed by the invention to optimize the sorting of the variables to be detected in the DUCG by computer processing.
  • the detection is performed serially or in parallel to determine the state of these variables.
  • the purpose is to diagnose the cause of the abnormality of the target system as soon as possible, and take effective measures to restore the target system to normal.
  • B n or BX n there are a large number of cause events that cause system anomalies in industrial systems, social systems, and biological systems, such as coil shorts, pump failures, component failures, subsystem failures, blocked conduction paths, foreign matter entry, contamination of a tissue or body, infection , damage or natural failure, etc.
  • B nk or BX nk is the k state of the variable B n or BX n .
  • B n and BX n is that B n is the root cause variable, there is no input, and BX n has input, which can be affected by other factors.
  • Some feature quantities increase or decrease the likelihood of BX nk (k ⁇ 0) occurring, such as geographic location, time, environment, season, religion, skin color, experience, kinship, hobbies, personality, living conditions, working conditions, and so on.
  • These feature quantities can all be represented by the event variable X i , where i is the variable label and X ij is the j state of X i .
  • the dynamic uncertainty causal map is an intelligent technical solution to complete this diagnosis.
  • DUCG is a method of expressing an indeterminate causal relationship between the above-mentioned cause events and feature quantities and other variables, and based on such expression and known evidence E, the diagnosis is based on the known state of the X variables, which are known X ij together constitutes evidence E.
  • E X 1, 2X2, 3 X 3, 1 X 4, 0 X 5,0
  • the comma separates the variable identifier from the state identifier, which is the state identifier.
  • the more X variables (whose states are known) in E the more accurate the diagnosis.
  • some states are known to have little or no help for accurate diagnosis, and some help a lot.
  • the intersection of different H kj is treated as an empty set.
  • a DUCG example is shown in Figure 1, where B variables or events are represented by rectangles; X variables or events are represented by circles; BX variables or events are represented by double circles; G variables or events are logic gate variables or events, with logic gates Representation; D event is the default cause event for an X variable or event, represented by a pentagon. These variables and events can also be optionally represented by other graphical symbols. B, X, BX, G, and D variables or events are also known as nodes.
  • the variable Gi must have at least two inputs and use Or other graphical symbols connect the input variables to G i and express a logical combination of the various states of interest of the input variables. These logical combinations are illustrated by the logic gate description table LGS i .
  • LGS i the logic gate description table
  • B, X, BX, D, and G variables and their states can be defined according to the described object.
  • B, X, BX, D, and G can all be cause variables or events, called parent variables or events, and are uniformly expressed by V, V ⁇ B, X, BX, D, G ⁇ , and the subscripts are unchanged.
  • Result variables or events can only be X or BX variables or events, called subvariables or events.
  • the action variable F n;i expresses a causal relationship between the parent variable V i and the sub-variant X n or BX n , where F nk;ij expresses a causal relationship between the parent event V ii and the sub-event X nk or BX nk , F Nk;i expresses the causal relationship between the parent variable V i and the sub-event X nk , F n; ij expresses the causal relationship between the parent event V ij and the sub-variant X n or BX n , and can use a directed arc Or other graphical symbol representation, pointing from the cause to the result.
  • V ij Pr ⁇ V ij ⁇ , v ⁇ ⁇ b, x, bx, d, g ⁇ , and V ij and v ij are elements of the event vector V i and the parameter vector v i , respectively.
  • the above-mentioned action variable matrix F n;i can be a conditional action variable matrix, with a dotted circular arc Or other graphical symbol representation.
  • the conditional action variable matrix expresses a conditional relationship between the cause event vector V i and the result event vector X n or BX n , that is, whether F n;i is established according to whether the condition event Z n;i is satisfied.
  • Z n;i X 1,2 , when X 1,2 is true, Z n;i satisfies, F n;i holds, become When X 1,2 is not true, Z n;i is not satisfied, F n;i does not hold, been deleted.
  • variable letters in each graphical symbol can be omitted, and only the variables and their status labels are written, as shown in Figure 2.
  • the first value is the label of the variable
  • the second value separated by a comma is the status label of the variable.
  • the D variable has only one state, so only the variable label.
  • Non-D nodes with known states can be color-coded, as shown by X 110 , 1 in FIG. If there is only the first value, it indicates that the state of the non-D variable is unknown, as shown by other non-D nodes in FIG.
  • Rule 8 When a directed arc has no parent or no children, remove the directed arc from the DUCG.
  • Rule 11 The above rules can be used, used in combination, and reused in any order.
  • some of the state-to-detect X variables are cause-specific variables, that is, their states can directly determine whether the hypothetical cause event H kj (H kj ⁇ S H (y)) holds.
  • H kj specific state variable is true (positive)
  • H kj is confirmed; when all specific state variables X H kj is false (negative), H kj is excluded; in other cases, H kj of Status is pending.
  • US invention patent "Method for constructing an intelligent system processing uncertain causal relationship information"; patent number: US 8255353 B2; authorization date: August 28, 2012; rights holder: Zhang Zhan; inventor: Zhang Qin, Zhang Zhan.
  • the invention discloses a technical solution for optimally sorting the detectable X variables whose state is unknown, so that the state of the X variable is detected according to the sorting, and the optimized E + (y) can be effectively found, so that the possible cause event set is obtained.
  • S H (y+1) is as small as possible, and the probability of a real cause event is as large as possible.
  • a sorting method for determining an X variable to be detected in a DUCG by a computing device including at least one CPU, and detecting the state of a part or all of the detectable X variable whose state is unknown by serial or parallel, according to the sorting, constitutes E + ( y), so that under the condition of E(y+1) E + (y)E(y), the hypothetical cause event H kj in S H (y+1) is as small as possible, and the order of the real cause event H kj
  • the steps include: (1) determining the detectable state to be detected X variable based on the DUCG simplified by E(y), the subscript set is S X (y); (2) If there is only one element in S X (y), the sorting is completed; (3) the order importance I i (y) of X i (i ⁇ S X (y)) is calculated; (4) according to the order importance I i ( y) Sort X i (i ⁇ S X (y)), refer to the sort to
  • a method for determining 1 (1) the state to be detected X variable by a computing device comprising at least one CPU, the steps of which are: (1) collecting all possible possibilities in the DUCG based on evidence E(y) simplification The cause event H kj , whose set is S H (y); (2) for each H k in S H (y), find a state in which the stateless known variable is blocked in one-way connection with H k The X variable is unknown, and the state of the X variable is detectable.
  • the subscript set of these X variables is S X (y).
  • the value can be given when building the DUCG, or only after the known S H (y), the H k is given according to the site conditions.
  • the average value of (k ⁇ S ik (y)) is called the probability importance ⁇ i (y) of X i .
  • the specific calculation methods include but are not limited to:
  • the criterion is that the higher the cost, the lower the value of ⁇ i , including but not limited to the cost.
  • the state of the DUCG in the conditional simplification is unknown.
  • the detectable X variable is given at the scene.
  • a method of determining probability importance in 5 by a computing device comprising at least one CPU, characterized by: based on the DUCG simplified by E(y), for each H kj (H kj ⁇ S k The state of (y)) is unknown to the specific variable X i (i ⁇ S s (y)), and ⁇ i (y) and ⁇ i (y) are not calculated according to 3 and 5, ie, S is deducted from S X (y) s (y) becomes S Xs (y), but takes ⁇ i (y) and ⁇ i (y) in i ⁇ S s (y) to a maximum, including but not limited to That is to say, X i in 5 is limited to i ⁇ S Xs (y).
  • the larger i (y) and ⁇ i are, the larger I i (y) is.
  • the specific algorithms include but are not limited to:
  • I i (y) ⁇ i (y) ⁇ i ⁇ i (y)
  • I i (y) w 1 ⁇ i (y)+w 2 ⁇ i (y)+w 3 ⁇ i
  • a method for determining a ranking in 1(4) by a computing device comprising at least one CPU characterized in that the X variables of the state to be detected are sorted according to the size of I i (y), and the X variables of the prior ranking are sorted.
  • a method for determining whether sorting in 1(5) is ended by a computing device comprising at least one CPU, characterized in that if there is only one hypothetical event in S H (y), the sorting ends; if all states are detectable X The state of the variable is determined, the sorting ends; if there is only one element in S X (y), the sorting ends.
  • Figure 2 Example of a simple expression of DUCG
  • FIG. 3 Block diagram of the steps of the present invention
  • Figure 4 Original DUCG map in the embodiment
  • FIG. 8 is a DUCG diagram after the state of the first five X variables in the first example is detected
  • Figure 12 DUCG diagram after detecting the states of X 8 and X 10 in Figure 9;
  • Figure 13 DUCG diagram based on E(2) simplification
  • Figure 14 Split and simplify the results of Figure 13 in accordance with BX 1 ;
  • Figure 15 Split and simplify the results of Figure 13 in accordance with BX 2 .
  • Fig. 4 be the original DUCG diagram with the parameters:
  • All r n;i 1; where X 14 is an unobservable variable for various reasons, and the remaining X variables are observable variables; X 15 and X 16 are specific variables of BX 3,1 , ie, X 15 When X or X 16,1 is detected as true, BX 3,1 must occur. When X 15,0 and X 16,0 are detected to be true, BX 3,0 must occur. In other cases, BX 3 The status is pending. Since ⁇ B 1,1 , B 2,1 , B 2,2 , B 3,1 ⁇ are equivalent to ⁇ BX 1,1 , BX 2,1 , BX 2,2 , BX 3,1 ⁇ , the B variable No longer as a diagnostic object. a n; D is a shorthand for the parameters of the directed arc between X n and D n .
  • Our task is to detect the state of as few X variables as possible with minimal cost, so that the set of possible cause events S H (y) is as small as possible, and the probability of real cause events is as large as possible.
  • S X (0) ⁇ 4,5,6,7,8,9, 10,11,12,13,15,16 ⁇ .
  • Figure 4 should be split according to H k and simplified according to the simplification rule. The result is shown in Figure 5-7.
  • the sorting probability can be calculated as:
  • Sorting probability before detection (1) the original BX 3,1 is sorted out, and the possible result space S H (1) is reduced; (2) The sorting probability of BX 1,1 is much larger than the other two. That is to say, in the case where only the state detection of five X variables is performed, it is basically confirmed that BX 1, 1 is the cause of the real object system abnormality.
  • the state detectable variables are ⁇ X 4 , X 5 , X 6 , X 8 , X 9 , X 10 ⁇ .
  • the state of these variables is unknown, and is unidirectionally connected with H k ⁇ ⁇ BX 1 , BX 2 ⁇ involved in S H (1), and the intermediate stateless known variable is blocked.
  • ⁇ i (1) can be calculated as follows:
  • the sort result is:
  • the weight coefficient of the subgraph Can be counted as:

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Databases & Information Systems (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computational Mathematics (AREA)
  • Algebra (AREA)
  • Probability & Statistics with Applications (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Fuzzy Systems (AREA)
  • Operations Research (AREA)
  • Biomedical Technology (AREA)
  • Automation & Control Theory (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Debugging And Monitoring (AREA)

Abstract

一种确定DUCG中状态待检测的X变量的排序方法,包括:步骤1:基于以E(y)为条件化简后的DUCG,确定可检测的状态待检测X变量,其下标集合为S X(y),y为时间序号;步骤2:如果S X(y)中的元素只有一个,排序结束;步骤3:计算X i(i∈S X(y))的排序重要度I i(y);步骤4:按照排序重要度I i(y)对X i(i∈S X(y))进行排序,参考排序对i∈S X(y)的X变量进行状态检测;以及步骤5:将y增加1,重复步骤1-5,直到无可检测的待检测X变量。使用本方法的技术方案,能够以尽可能小的代价尽快诊断对象系统异常的原因,从而采取有效措施,使对象系统恢复正常。

Description

一种基于动态不确定因果图的启发式检测系统异常原因的方法 技术领域
本发明涉及智能信息处理技术,处理对象为在动态不确定因果图DUCG中表达的不确定因果关系类信息,使用本发明提出的技术方案,通过计算机处理,对DUCG中的待检测变量进行优化排序,依次串行或并行进行检测以确定这些变量的状态,目的是以尽可能小的代价尽快诊断对象系统异常的原因,从而采取有效措施,使对象系统恢复正常。
背景技术
工业系统、社会系统和生物系统中存在大量导致系统异常的原因事件,例如线圈短路、泵故障停转、零部件失效、子系统失灵、传导通路阻塞、异物进入、某组织或机体被污染、感染、伤害或自然失效等,可用事件变量Bn或BXn表示,n为事件变量标号、Bnk或BXnk为变量Bn或BXn的k状态。Bn和BXn的区别在于Bn为根原因变量,没有输入,而BXn有输入,可受其他因素影响。k=0表示Bn或BXn处于正常状态;k=1、2、3...表示Bn或BXn处于不同的异常状态。Bn和BXn的状态多数不能直接检测,或难以直接检测。此外,系统中还存在大量与Bn或BXn有因果关系的特征量,如温度、压力、流量、速度、频率、各种化验或物理试验结果、调查结果、影像学结果、声学结果等。有些特征量使BXnk(k≠0)发生的可能性增加或减少,例如所处地域、时间、环境、季节、宗教、肤色、经历、血缘关系、嗜好、性格、居住条件、工作条件等。这些特征量均可用事件变量Xi表示,i为变量标号、Xij为Xi的j状态。j=0表示Xi处于正常状态;j=1、2、3...表示Xi处于不同的异常状态。人们可通过检测上述Xi的状态来诊断系统异常的原因Bnk或BXnk(k≠0),从而及时采取有效措施,使系统恢复正常。动态不确定因果图(简称DUCG)就是完成这种诊断的一种智能技术方案。DUCG是一种表达上述原因事件和特征量以及其它变量之间不确定因果关系、并基于这种表达和已知证据E进行诊断的方法,诊断基于X变量的已知状态,这些状态已知的Xij共同构成证据E。例如E=X1,2X2,3X3,1X4,0X5,0,其中逗号将变量标识与状态标识分开,后者为状态标识。一般来说,E中的X变量(其状态已知)越多,诊断越准确。但有些状态已知X变量对于准确诊断帮助不大或没有帮助,有些则帮助很大。实践中,获取X变量的状态是要付出代价的。因此只能有选择和(或)分先后地检测X变量的状态。本项发明要解决的技术问题是:基于DUCG,在已知当前时间序号y(y=0、1、2、...)获得的证据E(y)、以及基于E(y)诊断出的可能原因事件集合SH(y)的前提下,寻找下一步应该检测哪些状态待检测X变量的状态,得到新证据E+(y),以获得更新后的证据E(y+1)=E+(y)E(y),使DUCG在基于E(y+1)的条件下诊断出的可能原因事件集合SH(y+1)尽可能小,或改变其中的排序以使真实原因排序靠前,从而以尽可能小的代价尽快诊断出系统异常的原因。其中y=0指没有收到任何证据的情况,也就是E(0)为全集的情况。
SH(y)或SH(y+1)中的可能原因事件用Hkj表示,其中Hk标识一个或一组由k 标识的变量,例如H1=BX1、H2=BX2B4、等等;j标识这组变量的状态组合,例如H1,2=BX1,2、H2,3=B1,3X4,2、等等。不同的Hkj的交集被作为空集对待。
一个DUCG示例如图1所示,其中B变量或事件用矩形表示;X变量或事件用圆形表示;BX变量或事件用双圆表示;G变量或事件是逻辑门变量或事件,用逻辑门表示;D事件是X变量或事件的缺省原因事件,用五边形表示。这些变量和事件也可以选择用其它图形符号表示。B、X、BX、G和D变量或事件又称为节点。
变量Gi必须有至少两个输入,并用
Figure PCTCN2016083906-appb-000001
或其它图形符号将输入变量与Gi相连,表达输入变量各种所关注状态的逻辑组合,这些逻辑组合用逻辑门说明表LGSi说明。例如图1中G1用LGS1说明:G11,1=B3,1X111,1,G11,2=B3,1X111,2,G1,0=其它情况,等等。
B、X、BX、D和G变量及其状态的物理意义可以根据所描述对象定义。其中B、X、BX、D和G均可为原因变量或事件,称为父变量或事件,用V统一表达,V∈{B,X,BX,D,G},下标不变。例如V2=X2,V3,2=B3,2,等等。结果变量或事件只能是X或BX变量或事件,称为子变量或事件。
作用变量Fn;i表达父变量Vi与子变量Xn或BXn之间的因果关系,其中Fnk;ij表达父事件Vii与子事件Xnk或BXnk之间的因果关系,Fnk;i表达父变量Vi与子事件Xnk之间的因果关系,Fn;ij表达父事件Vij与子变量Xn或BXn之间的因果关系,可用有向弧
Figure PCTCN2016083906-appb-000002
或其它图形符号表示,从原因指向结果。Fnk;ij≡(rn;i/rn)Ank;ij。其中rn;i>0为父变量Vi与子变量Xn或BXn之间的因果关系强度,rn≡∑irn;i,Ank,ij为Vij导致Xnk或BXnk发生这一随机事件,ank;ij≡Pr{Ank;ij},满足
Figure PCTCN2016083906-appb-000003
Figure PCTCN2016083906-appb-000004
fnk;ij是Vij对Xnk的概率的贡献值,满足
Figure PCTCN2016083906-appb-000005
分别是矩阵Fn;i、fn;i、An;i、an;i的元素。vij=Pr{Vij},v∈{b,x,bx,d,g},Vij和vij分别是事件向量Vi和参数向量vi的元素。
上述作用变量矩阵Fn;i可以是条件作用变量矩阵,用虚线有向弧
Figure PCTCN2016083906-appb-000006
或其它图形符号表示。条件作用变量矩阵表达其原因事件向量Vi与结果事件向量Xn或BXn之间是条件作用关系,即根据条件事件Zn;i是否满足来判定Fn;i是否成立。例如Zn;i=X1,2,当X1,2为真时,Zn;i满足、Fn;i成立、
Figure PCTCN2016083906-appb-000007
成为
Figure PCTCN2016083906-appb-000008
当X1,2非真时,Zn;i不满足、Fn;i不成立、
Figure PCTCN2016083906-appb-000009
被删除。
为简单起见,每个图形符号中的变量字母可以被省略,只写该变量及其状态的标号,如图2所示。其中第一个数值为变量的标号,用逗号隔开的第二个数值为该变量的状态标号。D变量只有一个状态,故只有变量标号。状态已知的非D节点可以用颜色标出,如图2中X110,1所示。如果只有第一个数值,表示该非D变量的状态未知,如图2中的其它非D节点所示。
当收到证据E(y)后,可采用下述规则对DUCG进行化简:
规则1:如果E(y)显示Zn;i不满足,将Fn;i从DUCG中删除,当E(y)显示Zn;i已经满足, 条件有向弧Fn;i成为普通有向弧Fn;i,图形上体现为将
Figure PCTCN2016083906-appb-000010
改为
Figure PCTCN2016083906-appb-000011
规则2:如果E(y)显示Vij(V∈{B,X})为真,但Vij却不是Xn或BXn的父事件,将有向弧Fn;i从DUCG中删除。
规则3:如果E(y)显示Xnk为真,但Xnk不可能被Vi(V∈{B,X,BX,G,D})的任何状态引起,将有向弧Fn;i从DUCG中删除。
规则4:如果E(y)显示{B,X}类型节点状态未知且无输出有向弧,将该节点及其输入有向弧从DUCG中删除。
规则5:如果E(y)显示Xn0为真,且Xn0与异常证据E’(y)无任何连通关系,将Xn0从DUCG中删除。
规则6:如果E(y)显示一组状态未知节点除非通过Xn0,否则不与Xnk(k≠0)相连,将这组状态未知节点及与之相连的有向弧和D节点从DUCG中删除。
规则7:因任何原因导致Gi没有输出,将Gi及其输入有向弧
Figure PCTCN2016083906-appb-000012
从DUCG中删除;当Gi没有输入,将Gi及其输出有向弧从DUCG中删除。
规则8:当有向弧没有父节点或没有子节点,将这条有向弧从DUCG中删除。
规则9:当存在一组节点和有向弧与E(y)中涉及的节点无连通关系,将这组节点和有向弧从DUCG中删除。
规则10:如果E(y)显示Xnk为真,但Xnk因任何原因没有输入,对Xnk增加一个虚拟事件Dn作为其输入,在从Dn到Xnk的有向弧中,ank;nD=1且ank’;nD=0,k≠k’.rn;D可以为任何值。Dn可用符号
Figure PCTCN2016083906-appb-000013
表示。
规则11:上述规则可以按照任何顺序单独使用、联合使用、重复使用。
化简后的DUCG中的B或BX变量的异常状态事件Hkj=Bkj或Hkj=BXkj可构成SH(y)中的可能原因事件。
在化简后的DUCG中,状态待检测X变量中有一些是原因特异性变量,即它们的状态可以直接决定假设原因事件Hkj(Hkj∈SH(y))是否成立。当Hkj的特异性X变量状态为真(阳性)时,Hkj被确认;当Hkj的所有特异性X变量状态为假(阴性)时,Hkj被排除;其它情况下,Hkj的状态待定。
本发明既有技术参考文献:
[1]中国发明专利:“一种处理不确定因果关系类信息的智能系统的构造方法”,专利号:200680055266.X;授权日期:2010年4月14日;权利人:张湛;发明人:张勤、张湛。
[2]美国发明专利:“Method for constructing an intelligent system processing uncertain causal relationship information”;专利号:US 8255353 B2;授权日期:2012年8月28日;权利人:张湛;发明人:张勤、张湛。
[3]中国发明专利:“一种构造立体DUCG智能系统用于动态故障诊断的方法”;专利号:2013107185964;授权日期:2015年2月;权利人:张湛;发明人:张勤、董春玲。
[4]Q.Zhang.“Dynamic uncertain causality graph for knowledge representation and reasoning:discrete DAG cases”,Journal of Computer Science and Technology,vol.27,no.1,pp.1-23,2012.
[5]Q.Zhang,C.Dong,Y.Cui and Z.Yang.“Dynamic uncertain causality graph for knowledge representation and probabilistic reasoning:statistics base,matrix and fault diagnosis”,IEEE Trans.Neural Networks and Learning Systems,vol.25,no.4,pp.645-663,2014.
[6]Q.Zhang.“Dynamic uncertain causality graph for knowledge representation and probabilistic reasoning:directed cyclic graph and joint probability distribution”,IEEE Trans.Neural Networks and Learning Systems,vol.26,no.7,pp.1503-1517,2015.
[7]Q.Zhang.“Dynamic uncertain causality graph for knowledge representation and probabilistic reasoning:continuous variable,uncertain evidence and failure forecast”,IEEE Trans.Systems,Man and Cybernetics,vol.45,no.7,pp.990-1003,2015.
[8]Q.Zhang and S.Geng.“Dynamic uncertain causality graph applied to dynamic fault diagnosis of large and complex systems”,IEEE Trans.Reliability,vol.64,no.3,pp 910-927,2015
[9]Q.Zhang and Z.Zhang.“Dynamic uncertain causality graph applied to dynamic fault diagnoses and predictions with negative feedbacks”,IEEE Trans.Reliability,DOI:10.1109/TR.2015.2503759,2015.
发明内容
本发明公开了一种技术方案,对状态未知的可检测X变量进行优化排序,使得按照这个排序对X变量的状态进行检测,可以有效寻找到优化的E+(y),使得可能原因事件集合SH(y+1)尽可能小,真实原因事件的概率尽可能大。本专利申请是已授权中国发明专利ZL 2006 8 0055266.X、2013107185964和美国发明专利US 8255353 B2的后续专利申请,是对上述授权专利技术的进一步发展。
本发明的技术方案如下:
1、一种通过包含至少一个CPU的计算装置确定DUCG中状态待检测的X变量的排序方法,按照该排序串行或并行检测部分或全部状态未知的可检测X变量的状态,构成E+(y),使得在E(y+1)=E+(y)E(y)的条件下,SH(y+1)中的假设原因事件Hkj尽可能少,真实原因事件Hkj的排序尽可能靠前,其步骤包括:(1)基于以E(y)为条件化简后的DUCG,确定可检测的状态待检测X变量,其下标集合为SX(y);(2)如果SX(y)中的元素只有一个,排序结 束;(3)计算Xi(i∈SX(y))的排序重要度Ii(y);(4)按照排序重要度Ii(y)对Xi(i∈SX(y))进行排序,参考排序对i∈SX(y)的X变量进行状态检测;(5)如果需要继续排序,将y增加1,重复上述(1)-(5)的步骤,直到无可检测的待检测X变量。
2、一种通过包含至少一个CPU的计算装置确定1(1)所述状态待检测X变量的方法,其步骤为:(1)搜集基于证据E(y)化简后的DUCG中的所有可能原因事件Hkj,其集合为SH(y);(2)针对SH(y)中的每个Hk,寻找与Hk单向相连的、之间无状态已知变量阻断的状态未知X变量,且该X变量状态可检测,这些X变量的下标集合为SX(y)。
3、一种通过包含至少一个CPU的计算装置确定1(3)中影响排序重要度Ii(y)的结构重要度的方法,其特征在于:针对1(1)中的状态待检测变量Xi(i∈SX(y)),计算其连接的在2中确定的SH(y)中不同Hk(k∈SiK)的个数,记为mi(y),mi(y)越大,Xi的结构重要度λi(y)越小,λi(y)>0,其计算方法以能够表征mi(y)越大,λi(y)越小为准,包括但不限于λi(y)=1/(mi(y))n(n=1,2,...),Xi所连接的Hk的下标集合记为SiK(y),其中Hkj∈SH(y)中的Hk的状态j的集合记为SkJ(y)。
4、一种通过包含至少一个CPU的计算装置确定1(3)中影响排序重要度Ii(y)的关注重要度的方法,其特征在于:对SH(y)中的可能原因事件Hkj按照j≠0的所有异常状态受关注的程度综合打分,记为ωk,1≥ωk>0,称为关注重要度,受关注程度越大,ωk的值越大,ωk的值可在建造DUCG时给定,也可在已知SH(y)后仅对其中的Hk根据现场情况给定。
5、一种通过包含至少一个CPU的计算装置确定1(3)中影响排序重要度Ii(y)的概率重要度的方法,其特征在于:基于以E(y)为条件化简后的DUCG,计算如果将非特异性状态待检测Xi的各可能状态Xig(g∈SiG(y),SiG(y)为Xi的可能状态集合,包括或不包括状态0)加入E(y)(亦即将条件从E(y)改为XigE(y)),SH(y)中的Hk各状态的条件概率的变动幅度,这个变动幅度对所有单向连通的Hk(k∈Sik(y))的平均值称为Xi的概率重要度ρi(y),变动幅度越大,ρi(y)的值越大,具体计算方法包括但不限于:
Figure PCTCN2016083906-appb-000014
Figure PCTCN2016083906-appb-000015
Figure PCTCN2016083906-appb-000016
Figure PCTCN2016083906-appb-000017
Figure PCTCN2016083906-appb-000018
这些公式中的E(y)和ωk也可以不考虑,相当于被分别或一并去掉,也就是令E(y)=全集、ωk=1。
6、一种通过包含至少一个CPU的计算装置确定1(3)中影响排序重要度Ii(y)的代价重要度的方法,其特征在于:对可检测变量Xi(i∈SX(y))的检测困难程度(用j=1标识)、等待时间(用j=2标识)、成本(用j=3标识)以及检测对对象系统造成损害的程度(用j=4标识)综合打分,也可分别对这四项分别打分之后权重相加,权重系数σij(j=1、2、3、4)可在建造DUCG时给定,也可根据现场情况给定,综合打分或分别打分权重相加后的值称为代价,代价越大,代价重要度βi越小,1≥βi>0,其准则为代价越高,βi的值越低,包括但不限于代价最高为100时,βi=1/100=0.01;代价最低为1时,βi=1/1=1,其余为中间状况,βi可在建造DUCG时给定,也可针对以E(y)为条件化简后的DUCG中的状态未知可检测X变量在现场给定。
7、一种通过包含至少一个CPU的计算装置确定5中概率重要度的方法,其特征在于:基于以E(y)为条件化简后的DUCG,针对每个Hkj(Hkj∈Sk(y))的状态未知特异性变量Xi(i∈Ss(y)),不根据3和5计算λi(y)和ρi(y),即从SX(y)中扣除Ss(y)而成为SXs(y),而是令i∈Ss(y)中的λi(y)和ρi(y)取最大值,包括但不限于
Figure PCTCN2016083906-appb-000019
也就是说,5中的Xi仅限于i∈SXs(y)。
8、一种通过包含至少一个CPU的计算装置综合计算1(3)中的X变量的排序重要度Ii(y)的方法,其特征在于:2-7中的λi(y)、ρi(y)、βi越大,则Ii(y)越大,具体算法包括但不限于:
Figure PCTCN2016083906-appb-000020
Ii(y)=λi(y)βiρi(y)
Figure PCTCN2016083906-appb-000021
Ii(y)=λi(y)ρi(y)
Figure PCTCN2016083906-appb-000022
Ii(y)=βiρi(y)
Figure PCTCN2016083906-appb-000023
Ii(y)=ρi(y)
Ii(y)=w1λi(y)+w2ρi(y)+w3βi
Figure PCTCN2016083906-appb-000024
其中w1、w2和w3为三个权重系数,wi≥0(i=1、2、3),wi=0表示此项不予考虑,w1、w2和w3在建造DUCG时或现场给定。
9、一种通过包含至少一个CPU的计算装置确定1(4)中排序的方法,其特征在于:按照Ii(y)大小对状态待检测X变量排序,当排序靠前的X变量为排序靠后的X变量的唯一上游或下游X变量时,可将排序靠后的X变量从排序中删除,同时也可将Ii(y)=0的Xi从排序中删除。
10、一种通过包含至少一个CPU的计算装置确定1(5)中排序是否结束的方法,其特征在于:如果SH(y)中只有一个假设事件,排序结束;如果所有状态可检测的X变量的状态均已确定,排序结束;如果SX(y)中的元素只有一个,排序结束。
以上步骤的总体框图如图3所示。
附图说明
图1:DUCG示例;
图2:DUCG的简单表达方式示例;
图3:本发明的步骤框图;
图4:实施例中的原始DUCG图;
图5:y=0时对图4按照BX1进行拆分化简的结果;
图6:y=0时对图4按照BX2进行拆分化简的结果;
图7:y=0时对图4按照BX3进行拆分化简的结果;
图8:检测例1中排序前5个X变量的状态后的DUCG图;
图9:y=1时化简后的DUCG;
图10:y=1时对图9按照BX1进行拆分化简的结果;
图11:y=1时对图9按照BX2进行拆分化简的结果;
图12:检测图9中X8和X10的状态后的DUCG图;
图13:基于E(2)化简后的DUCG图;
图14:按照BX1拆分并化简图13的结果;
图15:按照BX2拆分并化简图13的结果。
具体实施方式
下面结合附图和具体实施例对本发明的技术方案进行详细描述。
设图4为原始DUCG图,其中的参数为:
Figure PCTCN2016083906-appb-000025
Figure PCTCN2016083906-appb-000026
Figure PCTCN2016083906-appb-000027
Figure PCTCN2016083906-appb-000028
Figure PCTCN2016083906-appb-000029
Figure PCTCN2016083906-appb-000030
Figure PCTCN2016083906-appb-000031
所有的rn;i=1;其中X14因各种原因是不可观测变量,其余X变量均是可观测变量;X15和X16为BX3,1的特异性变量,即当X15,1或X16,1被检测到为真时,BX3,1必定发生,当X15,0和X16,0被检测到均为真时,BX3,0必定发生,其它情况下BX3的状态待定。由于{B1,1,B2,1,B2,2,B3,1}与{BX1,1,BX2,1,BX2,2,BX3,1}等价,所以B变量不再作为诊断对象。an;D为Xn与Dn之间的有向弧的参数的简写。
我们的任务是通过最小的代价检测尽可能少的X变量的状态,以使得可能原因事件集合SH(y)尽可能小、真实原因事件的概率尽可能大。
例1:y=0(无任何证据)的情况
根据2,当无证据时(y=0),根据图4,SH(0)={H1,1,H2,1,H2,2,H3,1}={BX1,1,BX2,1,BX2,2,BX3,1}、状态待检测的可检测变量为{X4,X5,X6,X7,X8,X9,X10,X11,X12,X13,X15,X16}。这些变量状态均未知、与SH(0)中的Hk单向连通、且中间无状态已知变量阻断,因而SX(0)={4,5,6,7,8,9,10,11,12,13,15,16}。但由于X15和X16是BX3,1的特异性变量,所以被从SX(0)中删除,即SXs(0)={4,5,6,7,8,9,10,11,12,13}。相应地,S4G(0)=S5G(0)=S7G(0)=S8G(0)=S9G(0)=S10G(0)=S11G(0)=S12G(0)=S13G(0)={1}、S6G(0)={1,2}。
基于图4,根据[8]所述DUCG算法,由于不同Hkj的交集为空,即不同Hkj不能同时发生,应根据Hk对图4进行拆分,并根据化简规则进行化简,结果如图5-7所示。
由于无证据,E(0)=全集,各拆分子图的概率ζi(y)=ζi(0)=Pr{E(0)}=1(i∈{1,2,3}),各拆分子图的权重系数
Figure PCTCN2016083906-appb-000032
基于图5-7,
根据参考文献所述的算法,可得BX1,1,BX2,1,BX2,2和BX3,1的状态概率
Figure PCTCN2016083906-appb-000033
分别为:
Figure PCTCN2016083906-appb-000034
Figure PCTCN2016083906-appb-000035
Figure PCTCN2016083906-appb-000036
Figure PCTCN2016083906-appb-000037
排序概率
Figure PCTCN2016083906-appb-000038
的计算结果如下:
Figure PCTCN2016083906-appb-000039
按照3,基于图4,S12k(0)=S13K(0)={1}、S4K(0)=S7K(0)=S9K(0)={1,2}、S6K(0)=S11K(0)={2,3}、S5K(0)=S8K(0)=S10K(0)={1,2,3};相应地,m12(0)=m13(0)=1,m4(0)=m6(0)=m7(0)=m9(0)=m11(0)=2,m5(0)=m8(0)=m10(0)=3。由于BX1和BX3只有一个标号为“1”的异常状态,所以,S1J(0)=S3J(0)={1}。BX2有标号为“1”和“2”的两个异常状态,所以S2J(0)={1,2}。
按照4,设ωk=1(k∈{1,2,3})。
按照5,采用下式计算:
Figure PCTCN2016083906-appb-000040
其中i∈SXs(0)={4,5,6,7,8,9,10,11,12,13}。基于图4,由于ωk=1,k∈{1,2,3},于是
Figure PCTCN2016083906-appb-000041
其中:
Figure PCTCN2016083906-appb-000042
所以:
Figure PCTCN2016083906-appb-000043
同理,
Figure PCTCN2016083906-appb-000044
Figure PCTCN2016083906-appb-000045
Figure PCTCN2016083906-appb-000046
Figure PCTCN2016083906-appb-000047
Figure PCTCN2016083906-appb-000048
Figure PCTCN2016083906-appb-000049
Figure PCTCN2016083906-appb-000050
Figure PCTCN2016083906-appb-000051
Figure PCTCN2016083906-appb-000052
按照7,采用
Figure PCTCN2016083906-appb-000053
进行计算,可得:
ρ15(0)=ρ16(0)=max{0.0299,0.030331,0.040694,0.020938,0.01785,
                   0.020838,0.021232,0.027049,0.063,0.00792}=0.063。
于是有:
i ρi(0)
4 0.0299
5 0.030331
6 0.040694
7 0.020938
8 0.01785
9 0.020938
10 0.021232
11 0.027049
12 0.063
13 0.00792
15 0.063
16 0.063
按照6,设βi值给定如下:
i βi
4 0.01
5 0.5
6 0.02
7 1
8 0.5
9 0.5
10 0.5
11 1
12 1
13 1
15 1
16 1
按照3,设n=1,根据mi(0),可算得λi(0)如下:
i mi(0) λi(0)
4 2 1/2
5 3 1/3
6 2 1/2
7 1 1
8 3 1/3
9 1 1
10 1 1
11 1 1
12 1 1
13 1 1
15 1 1
16 1 1
其中,按照7,令λ15(0)=λ16(0)=1。
按照8,采用
Figure PCTCN2016083906-appb-000054
进行计算,可得:
Figure PCTCN2016083906-appb-000055
Figure PCTCN2016083906-appb-000056
Figure PCTCN2016083906-appb-000057
Figure PCTCN2016083906-appb-000058
Figure PCTCN2016083906-appb-000059
Figure PCTCN2016083906-appb-000060
Figure PCTCN2016083906-appb-000061
Figure PCTCN2016083906-appb-000062
Figure PCTCN2016083906-appb-000063
Figure PCTCN2016083906-appb-000064
Figure PCTCN2016083906-appb-000065
Figure PCTCN2016083906-appb-000066
按照9,排序结果为:
序号 i Ii(0)
1 12 0.229442
2 15 0.229442
3 16 0.229442
4 11 0.098511
5 7 0.076255
6 10 0.038663
7 9 0.038128
8 13 0.028844
9 5 0.018411
10 8 0.010835
11 6 0.001482
12 4 0.000544
取前5个X变量进行检测。设检测结果为:X7,1、X11,0、X12,1、X15,0、X16,0。图4成为图8。其中,由于X15和X16是BX3,1的特异性变量,检测结果均为阴性(状态为0),可知BX3的状态为BX3,0
根据例1所给参数,按照前述化简规则2、3和5,图8被化简为图9。其中,E(1)=E+(0)E(0)=E+(0)=X7,1X11,0X12,1。基于E(1),类似于例1,根据[8]所述DUCG算法,将图9拆分和化简为图10和图11。
根据文献[4-9]的算法,分别求拆分子图10和图11的概率ζi(y)=ζi(1):基于图10,
Figure PCTCN2016083906-appb-000067
基于图11,
Figure PCTCN2016083906-appb-000068
其中,原始参数a11,0;6=(- - -)被修改为a11,0;6=(1 1-0.3 1-0.8)。这是因为X11,0为负证据,表示异常状态均不发生。根据文献[4],X11,0=1-X11,1
根据文献[8]的算法,分图的权重系数
Figure PCTCN2016083906-appb-000069
Figure PCTCN2016083906-appb-000070
Figure PCTCN2016083906-appb-000071
按照1,根据[4-9]的DUCG算法,y=0+1=1时,Hkj的状态概率
Figure PCTCN2016083906-appb-000072
为:
Figure PCTCN2016083906-appb-000073
其中:
Figure PCTCN2016083906-appb-000074
Figure PCTCN2016083906-appb-000075
同理,
Figure PCTCN2016083906-appb-000076
Figure PCTCN2016083906-appb-000077
根据
Figure PCTCN2016083906-appb-000078
排序概率可算得为:
Figure PCTCN2016083906-appb-000079
与检测前的排序概率
Figure PCTCN2016083906-appb-000080
比较,(1)原来排序第一的BX3,1被排除了,可能结果状态空间SH(1)缩小了;(2)BX1,1的排序概率远大于其余两个。也就是说,在只做了5个X变量的状态检测的情况下,已基本上确诊BX1,1就是真实的对象系统异常的原因。
例2:y=1的情况
仍如例1,根据图9,E(1)=X7,1X11,0X12,1、SH(1)={H1,1,H2,1,H2,2}={BX1,1,BX2,1,BX2,2}、状态可检测变量为{X4,X5,X6,X8,X9,X10}。这些变量状态均未知、与SH(1)中涉及的Hk∈{BX1,BX2}单向连通、且中间无状态已知变量阻断。但按照9,X4为X7,1的唯一上游变量,可以从排序中删除,所以SX(1)={5,6,8,9,10}。由于无特异性变量,SXs(1)=SX(1)。与y=0时相同,S4G(1)=S5G(1)=S8G(1)=S9G(1)=S10G(1)={1}、S6G(1)={1,2}。
按照3,基于图9,S13K(1)={1}、S4K(1)=S5K(1)=S8K(1)=S9K(1)=S10K(1)={1,2}、S6K(1)={2}、;相应地,m6(1)=1,m4(1)=m5(1)=m8(1)=m10(1)m9(1)=2。S1J(1)={1}、S2J(1)={2}。仍如y=0的情况,ωk=1(k∈{1,2})。
按照5,仍如例1采用下式计算:
Figure PCTCN2016083906-appb-000081
其中i∈SXs(0)={4,5,6,8,9,10}。
基于图9,由于ωk=1,k∈{1,2},于是
Figure PCTCN2016083906-appb-000082
其中:
Figure PCTCN2016083906-appb-000083
Figure PCTCN2016083906-appb-000084
Figure PCTCN2016083906-appb-000085
Figure PCTCN2016083906-appb-000086
所以:
Figure PCTCN2016083906-appb-000087
同理,
Figure PCTCN2016083906-appb-000088
Figure PCTCN2016083906-appb-000089
Figure PCTCN2016083906-appb-000090
Figure PCTCN2016083906-appb-000091
Figure PCTCN2016083906-appb-000092
计算结果列表为:
i ρi(1)
4 0
5 0.092519
6 0.233261
8 0.065047
9 0.000199
10 0.06299
按照6,βi值仍如例1给定如下:
i βi
4 0.01
5 0.5
6 0.02
8 0.5
9 0.5
10 0.5
根据mi(1),可算得λi(1)如下:
i mi(1) λi(1)
4 2 1/2
5 2 1/2
6 1 1/1
8 3 1/3
9 1 1
10 1 1
按照8,采用
Figure PCTCN2016083906-appb-000093
进行计算,可得:
Figure PCTCN2016083906-appb-000094
Figure PCTCN2016083906-appb-000095
Figure PCTCN2016083906-appb-000096
Figure PCTCN2016083906-appb-000097
Figure PCTCN2016083906-appb-000098
Figure PCTCN2016083906-appb-000099
排序结果为:
序号 i Ii(1)
1 10 0.448449
2 5 0.329338
3 8 0.154363
4 6 0.066427
5 9 0.001417
6 4 0
按照9,由于X5是X10的唯一上游变量,可以将X5从排序中删除。又由于X4的排序概率等于0,对其检测没有意义,可从排序中删除。于是上述排序成为:
序号 i Ii(1)
1 10 0.448449
2 8 0.154363
3 6 0.066427
4 9 0.001417
对前3个变量进行检测,设检测结果为X10,1,X8,1和X6,0。则图9成为图12。由于a8,1;6,0=a11,0;6,0=“-”,表示相应的因果关系不存在。根据化简规则,图12被化简为图13。其中E(2)=E+(1)E(1)=X10,1X8,1X6,0X7,1X8,1X10,1X12,1。为计算
Figure PCTCN2016083906-appb-000100
Figure PCTCN2016083906-appb-000101
Figure PCTCN2016083906-appb-000102
图13被拆分并化简为为图14和图15。根据文献[8]所述算法,可得:
基于图14,
Figure PCTCN2016083906-appb-000103
基于图15,
Figure PCTCN2016083906-appb-000104
根据文献[8]的算法,分图的权重系数
Figure PCTCN2016083906-appb-000105
可算得为:
Figure PCTCN2016083906-appb-000106
Figure PCTCN2016083906-appb-000107
由于ζ2(0)=0或ξ2(0)=0,图15不能成立,应当删除。只有图14是唯一成立的子 图。也就是说,BX1,1是SH(2)中的唯一事件。按照10,排序结束。对象系统异常的原因被唯一确定为BX1,1,也就是B1,1

Claims (10)

  1. 一种通过包含至少一个CPU的计算装置确定DUCG中状态待检测的X变量的排序方法,按照该排序串行或并行检测部分或全部状态未知的可检测X变量的状态,构成E+(y),使得在E(y+1)=E+(y)E(y)的条件下,SH(y+1)中的假设原因事件Hkj尽可能少,真实原因事件Hkj的排序尽可能靠前,其步骤包括:(1)基于以E(y)为条件化简后的DUCG,确定可检测的状态待检测X变量,其下标集合为SX(y);(2)如果SX(y)中的元素只有一个,排序结束;(3)计算Xi(i∈SX(y))的排序重要度Ii(y);(4)按照排序重要度Ii(y)对Xi(i∈SX(y))进行排序,参考排序对i∈SX(y)的X变量进行状态检测;(5)如果需要继续排序,将y增加1,重复上述(1)-(5)的步骤,直到无可检测的待检测X变量。
  2. 根据权利要求1所述的通过包含至少一个CPU的计算装置确定DUCG中状态待检测的X变量的排序方法,其特征在于,所述步骤(1)包括如下分步骤:(1)搜集基于证据E(y)化简后的DUCG中的所有可能原因事件Hkj,其集合为SH(y);(2)针对SH(y)中的每个Hk,寻找与Hk单向相连的、之间无状态已知变量阻断的状态未知X变量,且该X变量状态可检测,这些X变量的下标集合为SX(y)。
  3. 根据权利要求1所述的通过包含至少一个CPU的计算装置确定DUCG中状态待检测的X变量的排序方法,其特征在于,所述步骤(3)包括如下分步骤:针对权利要求1(1)中的状态待检测变量Xi(i∈SX(y)),计算其连接的在权利要求2中确定的SH(y)中不同Hk(k∈SiK)的个数,记为mi(y),mi(y)越大,Xi的结构重要度λi(y)越小,λi(y)>0,其计算方法以能够表征mi(y)越大,λi(y)越小为准,包括但不限于λi(y)=1/(mi(y))n(n=1,2,...),Xi所连接的Hk的下标集合记为SiK(y),其中Hkj∈SH(y)中的Hk的状态j的集合记为SkJ(y)。
  4. 根据权利要求1所述的通过包含至少一个CPU的计算装置确定DUCG中状态待检测的X变量的排序方法,其特征在于,所述步骤(3)包括如下分步骤:对SH(y)中的可能原因事件Hkj按照j≠0的所有异常状态受关注的程度综合打分,记为ωk,1≥ωk>0,称为关注重要度,受关注程度越大,ωk的值越大,ωk的值可在建造DUCG时给定,也可在已知SH(y)后仅对其中的Hk根据现场情况给定。
  5. 根据权利要求1所述的通过包含至少一个CPU的计算装置确定DUCG中状态待检测的X变量的排序方法,其特征在于,所述步骤(3)包括如下分步骤:基于以E(y)为条件化简后的DUCG,计算如果将非特异性状态待检测Xi的各可能状态Xig(g∈SiG(y),SiG(y)为Xi的可能状态集合,包括或不包括状态0)加入E(y)(亦即将条件从E(y)改为XigE(y)),SH(y)中的Hk各状态的条件概率的变动幅度,这个变动幅度对所有单向连通的Hk(k∈Sik(y))的平均值称为Xi的概率重要度ρi(y),变动幅度越大,ρi(y)的值越大,具体计算方法包括但不限于:
    Figure PCTCN2016083906-appb-100001
    Figure PCTCN2016083906-appb-100002
    Figure PCTCN2016083906-appb-100003
    Figure PCTCN2016083906-appb-100004
    Figure PCTCN2016083906-appb-100005
    这些公式中的E(y)和ωk也可以不考虑,相当于被分别或一并去掉,也就是令E(y)=全集、ωk=1。
  6. 根据权利要求1所述的通过包含至少一个CPU的计算装置确定DUCG中状态待检测的X变量的排序方法,其特征在于,所述步骤(3)包括如下分步骤:对可检测变量Xi(i∈SX(y))的检测困难程度(用j=1标识)、等待时间(用j=2标识)、成本(用j=3标识)以及检测对对象系统造成损害的程度(用j=4标识)综合打分,也可分别对这四项分别打分之后权重相加,权重系数σii(j=1、2、3、4)可在建造DUCG时给定,也可根据现场情况给定,综合打分或分别打分权重相加后的值称为代价,代价越大,代价重要度βi越小,1≥βi>0,其准则为代价越高,βi的值越低,包括但不限于代价最高为100时,βi=1/100=0.01;代价最低为1时,βi=1/1=1,其余为中间状况,βi可在建造DUCG时给定,也可针对以E(y)为条件化简后的DUCG中的状态未知可检测X变量在现场给定。
  7. 根据权利要求5所述的通过包含至少一个CPU的计算装置确定概率重要度的方法,其特征在于:基于以E(y)为条件化简后的DUCG,针对每个Hkj(Hkj∈Sk(y))的状态未知特异性变量Xi(i∈Ss(y)),不根据权利要求3和权利要求5计算λi(y)和ρi(y),即从SX(y)中扣除Ss(y)而成为SXs(y),而是令i∈Ss(y)中的λi(y)和ρi(y)取最大值,包括但不限于λi(y)≥1、
    Figure PCTCN2016083906-appb-100006
    也就是说,权利要求5中的Xi仅限于i∈SXs(y)。
  8. 根据权利要求1所述的通过包含至少一个CPU的计算装置确定DUCG中状态待检测的X变量的排序方法,其特征在于,所述步骤(3)包括如下分步骤:权利要求2-7中的λi(y)、ρi(y)、βi越大,则Ii(y)越大,具体算法包括但不限于:
    Figure PCTCN2016083906-appb-100007
    Ii(y)=λi(y)βiρi(y)
    Figure PCTCN2016083906-appb-100008
    Ii(y)=λi(y)ρi(y)
    Figure PCTCN2016083906-appb-100009
    Ii(y)=βiρi(y)
    Figure PCTCN2016083906-appb-100010
    Ii(y)=ρi(y)
    Ii(y)=w1λi(y)+w2ρi(y)+w3βi
    Figure PCTCN2016083906-appb-100011
    其中w1、w2和w3为三个权重系数,wi≥0(i=1、2、3),wi=0表示此项不予考虑,w1、w2和w3在建造DUCG时或现场给定。
  9. 根据权利要求1所述的通过包含至少一个CPU的计算装置确定DUCG中状态待检测的X变量的排序方法,其特征在于,所述步骤(4)包括如下分步骤:按照Ii(y)大小对状态待检测X变量排序,当排序靠前的X变量为排序靠后的X变量的唯一上游或下游X变量时,可将排序靠后的X变量从排序中删除,同时也可将Ii(y)=0的Xi从排序中删除。
  10. 根据权利要求1所述的通过包含至少一个CPU的计算装置确定DUCG中状态待检测的X变量的排序方法,其特征在于,所述步骤(5)包括如下分步骤:如果SH(y)中只有一个假设事件,排序结束;如果所有状态可检测的X变量的状态均已确定,排序结束;如果SX(y)中的元素只有一个,排序结束。
PCT/CN2016/083906 2016-04-29 2016-05-30 一种基于动态不确定因果图的启发式检测系统异常原因的方法 WO2017185449A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/096,675 US11397741B2 (en) 2016-04-29 2016-05-30 Heuristic inspection method to find causes of system abnormality based on dynamic uncertain causality graph

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610282052.1A CN105956665B (zh) 2016-04-29 2016-04-29 一种基于动态不确定因果图的启发式检测系统异常原因的方法
CN201610282052.1 2016-04-29

Publications (1)

Publication Number Publication Date
WO2017185449A1 true WO2017185449A1 (zh) 2017-11-02

Family

ID=56912957

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/083906 WO2017185449A1 (zh) 2016-04-29 2016-05-30 一种基于动态不确定因果图的启发式检测系统异常原因的方法

Country Status (3)

Country Link
US (1) US11397741B2 (zh)
CN (1) CN105956665B (zh)
WO (1) WO2017185449A1 (zh)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107944562B (zh) * 2017-10-17 2019-07-05 北京清睿智能科技有限公司 一种扩展的处理不确定因果关系类信息的智能系统的构造方法
CN108629417B (zh) * 2018-05-14 2019-12-03 北京予同智能科技有限公司 一种对ducg进行分层递归推理的高效方法
CN112448836B (zh) * 2019-09-04 2023-09-15 中兴通讯股份有限公司 故障根因确定方法、装置、服务器和计算机可读介质
CN112887119B (zh) * 2019-11-30 2022-09-16 华为技术有限公司 故障根因确定方法及装置、计算机存储介质
CN111060813B (zh) * 2019-12-09 2022-05-10 国网北京市电力公司 高压断路器操作机构的故障诊断方法及装置、电子设备
US11816080B2 (en) 2021-06-29 2023-11-14 International Business Machines Corporation Severity computation of anomalies in information technology operations
CN116545650B (zh) * 2023-04-03 2024-01-30 中国华能集团有限公司北京招标分公司 一种网络动态防御方法
CN117197546A (zh) * 2023-08-24 2023-12-08 佛山科学技术学院 基于因果机制的故障诊断方法及计算机设备

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2007134495A1 (fr) * 2006-05-16 2007-11-29 Zhan Zhang Procédé de construction de système intelligent traitant une information de relation causale incertaine
CN101127100A (zh) * 2006-08-18 2008-02-20 张湛 一种处理不确定因果关系类信息的智能系统的构造方法
CN103745261A (zh) * 2013-12-24 2014-04-23 张湛 一种构造立体ducg智能系统用于动态故障诊断的方法

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6820044B2 (en) * 2001-10-09 2004-11-16 University Of Maryland Method and apparatus for a common-cause failure module for probabilistic risk assessment tools
US9697503B1 (en) * 2011-04-22 2017-07-04 Angel A. Penilla Methods and systems for providing recommendations to vehicle users to handle alerts associated with the vehicle and a bidding market place for handling alerts/service of the vehicle

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2007134495A1 (fr) * 2006-05-16 2007-11-29 Zhan Zhang Procédé de construction de système intelligent traitant une information de relation causale incertaine
CN101127100A (zh) * 2006-08-18 2008-02-20 张湛 一种处理不确定因果关系类信息的智能系统的构造方法
CN103745261A (zh) * 2013-12-24 2014-04-23 张湛 一种构造立体ducg智能系统用于动态故障诊断的方法

Also Published As

Publication number Publication date
US11397741B2 (en) 2022-07-26
US20210256030A1 (en) 2021-08-19
CN105956665B (zh) 2017-06-06
CN105956665A (zh) 2016-09-21

Similar Documents

Publication Publication Date Title
WO2017185449A1 (zh) 一种基于动态不确定因果图的启发式检测系统异常原因的方法
Garg et al. Quantifying gesture information in brain hemorrhage patients using probabilistic dual hesitant fuzzy sets with unknown probability information
Zeng et al. Belief reliability: a new metrics for products’ reliability
Thao et al. A new fuzzy entropy on Pythagorean fuzzy sets
Liu et al. A FTA-based method for risk decision-making in emergency response
WO2019076026A1 (zh) 一种扩展的处理不确定因果关系类信息的智能系统的构造方法
EP2963552B1 (en) System analysis device and system analysis method
CN112382407A (zh) 一种风险管控方法、装置、电子设备和存储介质
CN107168842A (zh) 基于pmc模型的自适应顺序故障诊断方法
Gibson et al. Real-time mechanistic bayesian forecasts of covid-19 mortality
Zhou et al. A large group decision-making method based on fuzzy preference relation
CN111884859A (zh) 一种网络故障诊断方法、装置及可读存储介质
Leifeld et al. Fault detection for probabilistic Boolean networks
US9959153B2 (en) Assisting failure diagnosis in a system
Zaitseva et al. Healthcare system reliability analysis addressing uncertain and ambiguous data
Yang et al. A multi-components approach to monitoring process structure and customer behaviour concept drift
Hu et al. Who are the ‘silent spreaders’?: Contact tracing in spatio-temporal memory models
Daniel et al. [Retracted] Recurrent Neural Networks for Feature Extraction from Dengue Fever
Lopatka et al. Classification and Prediction of Diabetes Disease Using Modified k-neighbors Method
Schwarzerova et al. An innovative perspective on metabolomics data analysis in biomedical research using concept drift detection
Zeng Risk-adjusted performance monitoring in healthcare quality control
Kim Chi et al. Clinical data-driven approach to identifying COVID-19 and influenza from a gradient-boosting model
Phu et al. Enhanced Diagnosis of Influenza and COVID-19 Using Machine Learning
CN117891566B (zh) 智能软件的可靠性评估方法、装置、设备、介质和产品
US11783233B1 (en) Detection and visualization of novel data instances for self-healing AI/ML model-based solution deployment

Legal Events

Date Code Title Description
NENP Non-entry into the national phase

Ref country code: DE

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 16899959

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 16899959

Country of ref document: EP

Kind code of ref document: A1