CN109523032A - Construction Method of Evidence Network for Equipment Parameter Monitoring Based on Independence Analysis and Test - Google Patents

Abstract

The method for constructing the equipment parameter monitoring evidence network based on the independence analysis and test is characterized in that different equipment parameters of target equipment are synchronously acquired by utilizing a plurality of sensors, and each equipment parameter is used as a node in the parameter monitoring evidence network of the target equipment. And synchronously sampling by each sensor at the same time interval in the running time of the target equipment, and taking the monitoring data sampled by each sensor in the running time of the target equipment as the node input data of each node. And constructing a undirected evidence network structure model based on the maximum information coefficient among the nodes. And finally, determining the direction of the undirected edge according to the noise model on the basis of the undirected evidence network model, and constructing a device parameter monitoring evidence network corresponding to the target device. According to the invention, historical data of a plurality of different device parameters of the target device are detected by using the sensor, and the network structure is constructed based on the historical data, so that the objectivity and the accuracy of the evidence network structure are improved.

Description

Construction Method of Evidence Network for Equipment Parameter Monitoring Based on Independence Analysis and Test

technical field

The invention relates to the technical field of multiple information fusion in equipment monitoring, in particular to a method for building an evidence network for equipment parameter monitoring based on independence analysis and testing.

Background technique

In the fusion of multiple information in equipment monitoring, data correlation and information uncertainty are ubiquitous. In order to sort out the relationship between multiple information, and then realize the effective fusion of multiple information, it is necessary to consider the relationship between many different types of information that are interconnected and affect each other, as well as various internal and external factors in the process of equipment parameter information collection. Uncertainty caused by combination. If these uncertainties cannot be described correctly, the relationship between information cannot be described reasonably, and the effective fusion of multiple information cannot be realized.

As an organic combination of graph model and evidence theory, evidence network has the ability of graph model to express complex relationship and the advantage of evidence theory to describe uncertain knowledge. The expert system represented by the network structure can describe the causal relationship between different information qualitatively and quantitatively, and make inferences based on the corresponding observation data. Therefore, the evidence network is currently one of the most effective theoretical models in the field of complex uncertain knowledge representation and reasoning, and is widely used in the management and decision-making practice of complex systems such as weaponry system capability assessment, reliability assessment, status monitoring, and medical diagnosis.

The information expression of the evidence network consists of two parts: one part is a network structure represented by a directed acyclic graph, each node in the network represents a parameter in the actual sample data, and the connection between nodes represents the causal relationship between variables, the arrow The node pointed by the tail represents the cause, and the node pointed by the arrow represents the result; the other part is the evidence network parameter, or the confidence rule base, which expresses the degree of influence from the cause to the result. When solving the multivariate information fusion problem in equipment monitoring, the evidence network structure constructed is the parameter network involved in the multivariate information fusion problem.

As a structural model of the evidence network, the graph model describes the relationship between variables and is the basis for analyzing and solving problems by applying evidence network reasoning methods. If the evidence network structure model misses or misjudges the relationship between variables, it will affect the results of network reasoning and further reduce the reliability of the evidence network model to analyze and solve problems.

At present, the method of directly constructing the network structure model based on expert experience and knowledge is mostly used in the application research of evidence network. Although such a method is simple and fast, it has certain deficiencies and limitations in constructing evidence network structure only relying on domain knowledge. The accuracy of the results based on such network model analysis and inference is naturally not high.

SUMMARY OF THE INVENTION

Aiming at the defects in the prior art, the present invention provides a method for constructing an evidence network for equipment parameter monitoring based on independent analysis and testing. With the improvement of sensor sampling technology and database storage capacity, there are often a large amount of relevant historical data that can be used in the analysis of practical problems. The present invention uses sensors to collect historical data of multiple different device parameters of the target device, and constructs a device parameter monitoring evidence network based on the historical data, which is conducive to improving the objectivity of the constructed evidence network structure, and further improving the reasoning and analysis results of the evidence network Accuracy in multivariate information fusion problems.

For realizing the purpose of the present invention, adopt following technical scheme to realize:

Referring to Figure 4, the construction method of evidence network for equipment parameter monitoring based on independent analysis and testing includes the following steps:

S1 determines the target device, uses k (k>1) sensors to collect k device parameters of the target device, and each device parameter is used as a node in the target device parameter monitoring evidence network, that is, there are k nodes.

Each sensor samples synchronously at the same time interval during the running time of the target device, and the monitoring data obtained by each sensor during the running time of the target device is used as the node input data of each node. It is assumed that each sensor samples n times during the running time of the target device , then the node input data of each node includes n sample data. The set of node input data of all nodes is denoted as data set D.

S2 builds an undirected evidence network structure model based on the maximum information coefficient.

S2.1 Calculate the maximum information coefficient between each node in the data set D and other nodes in the data set D.

For any node x _i in the data set D, calculate the maximum information coefficient between the node x _i and any other node x _j in the data set D, the method is as follows:

Take the set of all values within the range of the minimum value and the maximum value in the node input data of node x _i as the value range of node x _i , and take the minimum value and all values within the range of the maximum value in the node input data of node x _j The collection of values is used as the value range of node x _j . Divide the value interval of node x _i into a sub-interval, and divide the value interval of node x _j into b sub-intervals. The division of the value interval can be arbitrarily selected under the condition of satisfying the constraint ab<B(n), In order to achieve the purpose of maximizing the MIC value, the division of the value range is not required to be evenly divided, and such a division method is recorded as G. The maximum information coefficient list of node x _i is denoted as Indicates the maximum information coefficient between node x _i and node x _j , which has symmetry.

In the formula, the variable n represents the number of sample sets, and the function B(n) limits the size of the search grid. According to experiments, it is more appropriate to set B(n) as B(n)=n ^0.6 . In the present invention, B(n) is set to n ^0.6 . I(D| _G ) represents the mutual information value of the data set D divided by G. Suppose the joint distribution of node x _i and node j under the division of G is p( _xi , x _j ), and the marginal distributions are respectively p( _xi ) and p(x _j ), then the mutual information value can be calculated as follows ;

According to the same method as above, calculate the maximum information coefficient between node _xi and other nodes in the data set D, and obtain the list of maximum information coefficients of node _xi

S2.2 Correct the list of maximum information coefficients between nodes in the data set D.

For any node x _i in the data set D, its revised maximum information coefficient list is recorded as

In order to avoid the generation of triangular loops due to the introduction of redundant edges, the maximum information coefficient of each node needs to be corrected. In the present invention, a penalty factor δ is introduced. The larger δ is, the less likely it is to introduce redundant edges, but the possibility of missing edges in the actual evidence network structure increases; the smaller δ is, the less likely it is to miss the edges existing in the actual evidence network. However, the possibility of introducing redundant edges increases. The value of penalty factor δ is determined based on empirical knowledge, generally δ∈[0, 0.5]. Node h represents the list of maximum information coefficients that co-occur in node x _i A list of the maximum information coefficients of node x _j before node x _j in the middle A kind of node before the node x _i in the middle. m represents the total number of such nodes.

S2.3 Determine whether there is an undirected edge between nodes.

S2.3.1 List of maximum information coefficients after correction for all nodes Arranged in descending order, record its maximum value as

S2.3.2 Set the isolation threshold factor γ, the value of the isolation threshold factor γ is determined based on empirical knowledge, generally γ∈[0,0.3]. if Then the node x _i is an isolated node, and there is no undirected edge, otherwise, execute S2.3.3;

S2.3.3 If the corrected maximum information coefficient between node x _i and node x _j satisfies:

Then there is an undirected edge between node x _i and node x _j . Among them, α is the connection threshold factor. The larger α is, the harsher the connection conditions are, and it is not easy to introduce redundant edges; the smaller α is, the looser the connection conditions are, and it is not easy to cause omissions. The value of connection threshold factor α is determined based on empirical knowledge, and generally takes α∈[0.7,0.9].

S3 is based on the undirected evidence network model, determines the direction of the undirected edge according to the noise-added model, and constructs the device parameter monitoring evidence network corresponding to the target device.

S3.1 On the premise that there is an undirected edge between node x _i and node x _j , respectively construct nonlinear regression models x _i :=f(x _j )+n ₁ and x _j :=f(x _i )+ n ₂ , f(x _j ) in the model _xi := f(x _j )+n ₁ refers to the nonlinear regression function that describes the node x _j as the cause and node _xi as the result obtained through fitting, and n ₁ refers to the model x _i :＝f(x _j )+noise item in n ₁ ; f( _xi ) in the model x _j :＝f(x _i )+n ₂ refers to the fitted description node x _i as the cause, the node x _j is the nonlinear regression function of the result, and n ₂ refers to the noise item in the model x _j :=f( _xi )+n ₂ .

S3.2 Calculate the residuals of the two models in S3.1 respectively n ₁ = _xi -f(x _j ) and n ₂ =x _j -f(x _i ), f(x _j ) refers to the node x _j Bring the node input data into the nonlinear regression model x _i : the value of the node x _i obtained after =f(x _j )+n ₁ , f( _xi ) refers to bringing the node input data of node x _i into the nonlinear regression model x _j : the value of node x _j obtained after =f( _xi )+n ₂ .

S3.3 Carry out T-test on the two groups of variables of x _i and n ₁ , x _j and n ₂ respectively. By computing the test statistic and Obtain the corresponding t values, which are recorded as t ₁ and t ₂ values, and then refer to the T test statistics table to obtain the p values, and record the obtained p values as p ₁ and p ₂ respectively. in, Indicates the mean value of the node input data corresponding to the node x _j , σ ₁ indicates the standard deviation of the node input data corresponding to the node x _j , Indicates the mean value of the node input data corresponding to node _xi , σ ₂ indicates the standard deviation of the node input data corresponding to node _xi , and n indicates the number of samples of node input data collected by each node in the data set D.

S3.4 Compare the size of p ₁ and p ₂ , if p ₁ > p ₂ , accept the model x _i := f(x _j )+n ₁ , consider the direction of the undirected edge between node x _i and node x _j is pointing from node x _j to node x _i ; if p ₂ >p ₁ , accept the model x _j := f( _xi )+n ₂ , and consider the direction of the undirected edge between node x _j and node x _i to be from Node x _i points to node x _j ;

S3.5 Determine the directions of all undirected edges according to S3.1 to S3.4, and check whether there is a cycle in the directed graph on this basis. If there is a cycle, remove the edge with the smallest p-value. So far, the construction of the device parameter monitoring evidence network corresponding to the target device has been completed.

In the present invention, the maximum information coefficient between each pair of nodes in the data set is used as a description of the strength of the causal relationship between nodes. The larger the maximum information coefficient, the stronger the causal relationship between nodes, and the more likely there is an edge between nodes, and vice versa. side. In order to avoid the generation of redundant edges, a penalty factor is introduced to modify the maximum information coefficient. Design connection rules to analyze the modified maximum information coefficient, and then determine the undirected evidence network structure. Further, on the basis of generating the undirected evidence network structure, determine the directed network structure, that is, determine the direction of the edges. Using the noise-added model to analyze the causal relationship of each undirected edge to determine the direction of the edge.

The present invention has the following beneficial effects:

(1) For the target object, that is, the target device, multiple sensors are used to collect different device parameters of the target device synchronously, and the network structure is constructed by using the historical data of multiple different device parameters of the target device, which is conducive to improving the quality of the evidence network structure. Objectivity, and then improve the accuracy of the reasoning and analysis results of the evidence network in the multi-information fusion problem. From the constructed network structure, the relationship and interaction between different device parameters of the target device can be effectively described.

(2) The present invention uses the maximum information coefficient to measure the degree of correlation between nodes. As a coefficient for identifying the correlation between variables, the maximum information coefficient can accurately identify whether the relationship between variables is a functional relationship or a non-functional relationship. Through such properties, a more reliable undirected network structure can be obtained in the end.

(3) The present invention uses a noise-added model to determine the direction of edges in the network, and then obtains a directed network structure. Adding noise model is a relatively mature method of direction determination, which can effectively use data to determine the direction of edges between nodes. Through such a mechanism, the direction of each edge is gradually determined, and finally a directed evidence network structure is formed.

Description of drawings

Fig. 1 is the evidence network structural diagram of three kinds of engine parameter monitoring to be established in the embodiment; Wherein Fig. 1 (1) represents the evidence network structural diagram of the turbine engine parameter monitoring to be established; Fig. 1 (2) represents the horizontal horizontal structure to be established Figure 1(3) shows the structure diagram of the evidence network for parameter monitoring of the rotary engine to be established.

Fig. 2 utilizes the undirected network model of the turbine engine parameter monitoring that the inventive method builds;

Fig. 3 is a structure diagram of the evidence network for turbine engine parameter monitoring constructed by the method of the present invention.

Fig. 4 is a flowchart of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below in combination with the accompanying drawings in the embodiments of the present invention, and further detailed descriptions will be given, but the embodiments of the present invention are not limited thereto.

In the following, three engines are used as target devices, and the equipment parameter monitoring evidence network construction method based on independent analysis and testing provided by the present invention is used to construct the equipment parameter monitoring evidence networks of the three engines respectively. The three types of engines are turbine engines, horizontal engines, and rotary engines.

Figure 1 is a structure diagram of the evidence network for three kinds of engine parameter monitoring to be established, in which:

Figure 1(1) shows the evidence network structure diagram of the turbine engine parameter monitoring to be established. Node A represents the ratio of fuel flow and static pressure at the outlet of the high-pressure compressor of the turbine engine, and node B represents the flow rate of the high-pressure turbine coolant of the turbine engine. , node C represents the exhaust enthalpy of the turbine engine.

Figure 1(2) shows the evidence network structure diagram of horizontal engine parameter monitoring to be established. Node A indicates the outlet temperature of the low-pressure compressor of the horizontal engine, node B indicates the outlet temperature of the high-pressure compressor of the horizontal engine, and node C indicates the outlet temperature of the horizontal engine. The low -pressure turbine exit temperature of the engine.

Figure 1(3) shows the evidence network structure diagram of the parameter monitoring of the rotary engine to be established. Node A represents the combustion air ratio of the combustor of the rotary engine, node B represents the core speed of the rotary engine, and node C represents the bypass pressure of the rotary engine

In order to verify the feasibility and validity of the equipment parameter monitoring evidence network construction method based on independent analysis and testing provided by the present invention, 200 groups of monitoring data are collected for each of the three engine parameters involved in Fig. 1, and the method proposed by the present invention is adopted The equipment parameter monitoring evidence network of the above three engines is constructed. Take the turbine engine among Fig. 1 (1) as example to illustrate the steps of the present invention:

S1 determines that the target equipment is a turbine engine, and uses sensors to simultaneously sample the fuel flow rate and static pressure ratio at the outlet of the high-pressure compressor, the flow rate of the high-pressure turbine coolant, and the exhaust enthalpy during the operating life cycle of the turbine engine at the same time interval.

Note that "the ratio of fuel flow and static pressure at the outlet of the high-pressure compressor" is node A, "high-pressure turbine coolant flow rate" is node B, and "exhaust enthalpy" is node C. The monitoring data sampled by each sensor is used as the node input data of each node. The set of node input data of all nodes is denoted as data set D.

S2 builds an undirected evidence network structure model based on the maximum information coefficient:

S2.1 Calculate the maximum information coefficient between two nodes:

Table 1 Maximum information coefficient between nodes

maximum information coefficient A B C A 1 1 B 1 1 0.8778 C 1

Then the maximum information coefficient list of each node is shown in Table 2 to Table 4:

Table 2 The maximum information coefficient list of node A

Table 3 List of maximum information coefficients of node B

list of maximum information coefficients MIC_B A 1 B 1 C

Table 4 List of maximum information coefficients of node C

list of maximum information coefficients MIC_C C 1 B A

S2.2 Correct the maximum information coefficient between each node;

Taking the maximum information coefficient between node A and node C as an example, if δ=0.1, then:

Table 5 The maximum information coefficient between the node after the correction

S2.3 Determine whether there is an anonymous side between nodes:

S2.3.1

S2.3.2 Set the isolation threshold factor γ=0.3, according to the result of S2.3.1, it is obvious that there is no isolated node;

S2.3.3 Set the connection threshold factor α=0.8, and check whether the corrected maximum information coefficient between nodes satisfies:

If you are satisfied, there will be an influence. After inspection, there are undirected edges between node A and node B, node B and node C. as shown in picture 2.

S3 is based on the undirected evidence network model, and determines the direction of the undirected edge according to the noise model: (take node A and node B as an example for analysis)

S3.1 There is an undirected edge between node A and node B, and a nonlinear regression model is constructed respectively:

A:=-0.0062B ² +0.927B+19.553 and B:=-0.0127A ² +3.3923A-76.891;

S3.2 Calculate respectively the residuals n ₁ =Af(B) and n ₂ =Bf(A) of the two models in step S3.1;

S3.3 Carry out T test on two groups of variables of x _i and n ₁ , x _j and n ₂ respectively, and the obtained p values are: p ₁ =2.5×10 ^-198 and p ₂ =5.06×10 ^-67 ;

S3.4 Compare the sizes of p ₁ and p ₂ , because p ₂ > p ₁ , so accept model B:=f(A)+n ₂ , and consider the direction of the undirected edge between node A and node B to be from node A point to node B;

S3.5 Determine the direction of the edge between node B and node C from node B to node C according to S3.1-S3.4, and determine that there is no loop in the network. The network structure obtained at this time is the evidence network monitoring of the turbine engine parameter monitoring in Figure 1 (1), as shown in Figure 3.

Construct the evidence network of horizontal engine parameter monitoring in Fig. 1(2) and the evidence network of rotor engine parameter monitoring in Fig. 1(3) according to the same method steps as above for constructing the evidence network of turbine engine parameter monitoring, and the results are shown in the table

Table 6 evidence network structure learning results

Table 6 shows the results of the two engine parameter monitoring evidence network structure learning. The fourth column "redundant edge" in Table 6 describes the number of edges that do not exist in the evidence network structure of equipment parameter monitoring to be established but are finally constructed using the method of the present invention and exist in the evidence network structure of equipment parameter monitoring The fifth column "missing edges" describes the number of edges that exist in the evidence network structure for equipment parameter monitoring to be established but do not exist in the evidence network structure for equipment parameter monitoring constructed by the method of the present invention. It can be seen from Table 6 that the invention can effectively and accurately build the evidence network structure that obtains equipment parameter monitoring based on small sample data. There are neither redundant edges nor missing edges in the network structure of the three engine equipment parameter monitoring in the instance. For the three engine parameter monitoring network structures in the instance, the present invention can accurately build its parameter monitoring network structure for each type of engine with only 200 sample data. The present invention solves the problem of the network structure relying on the experience of expert experience in the network modeling process of evidence, and puts forward the data -driven evidence network structure learning method, which greatly enhances the objectivity of the network structure of the evidence, and then improves the results of the network reasoning of the evidence network reasoning. Reliability, so it has high degree of utilization value, and has good application prospects in solving the problem of multiple information integration.

In summary, although the present invention has been disclosed above with preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art may make various modifications without departing from the spirit and scope of the present invention. Dynamic and moisturizing, so the protection scope of the present invention shall prevail the scope of the definition of claims.

Claims (5)

1. The equipment parameter monitoring evidence network construction method based on independence analysis and test is characterized by comprising the following steps of:

s1, determining target equipment, and collecting k equipment parameters of the target equipment by using k (k is more than 1) sensors respectively, wherein each equipment parameter is used as a node in a parameter monitoring evidence network of the target equipment, namely k nodes are provided;

the method comprises the steps that each sensor synchronously samples within the running time of target equipment according to the same time interval, monitoring data sampled within the running time of the target equipment by each sensor is used as node input data of each node, and if each sensor samples for n times within the running time of the target equipment, the node input data of each node comprises n sample data; the set of node input data of all nodes is recorded as a data set D;

s2, constructing a undirected evidence network structure model based on the maximum information coefficient;

s2.1, calculating the maximum information coefficient between each node in the data set D and each other node in the data set D;

s2.2, correcting the maximum information coefficient list among the nodes in the data set D;

s2.3, judging whether undirected edges exist between the nodes;

s3, based on the undirected evidence network model, determining the undirected direction according to the noise model, and constructing a device parameter monitoring evidence network corresponding to the target device.

2. The independence analysis and testing based equipment parameter monitoring evidence network construction method according to claim 1, characterized in that in S2.1 for any node x in the data set D_iCalculating node x_iWith any other node x in the data set D_jThe maximum information coefficient in between, the method is as follows:

node x_iThe node of (1) sets all values within the range of the minimum value and the maximum value in the input data as a node x_iThe value interval of (2), the node x_jThe node of (1) sets all values within the range of the minimum value and the maximum value in the input data as a node x_jThe value range of (1); node x_iIs divided into a sub-intervals, and the node x is divided into a sub-intervals_jThe value interval is divided into b sub-intervals, the division of the value interval is not required to be uniform, the value interval can be arbitrarily divided under the condition that the constraint ab < B (n) is met, and the division mode is marked as G; node x_iIs tabulated as Representing a node x_iAnd node x_jThe maximum information coefficient in between, has symmetry;

where the variable n represents the number of sample sets, the function b (n) limits the size of the search grid, and b (n) n^0.6；I(D|_G) Representing the mutual information value of the data set D with the division mode G, and setting a node x_iAnd node_jWith a joint distribution under the division of G of p (x)_i,x_j) Marginal distribution is p (x) respectively_i) And p (x)_j) Then the mutual information value can be calculated as follows;

in the same way as above, node x is computed_iObtaining the maximum information coefficient between the node x and other nodes in the data set D_iList of maximum information coefficients

3. The independence analysis and testing based equipment parameter monitoring evidence network construction method according to claim 2, characterized in that in S2.2, for any node x in the data set D_iThe corrected maximum information coefficient is listed as

Wherein, delta is a penalty factor, and is delta epsilon [ c ], [ alpha ]0，0.5](ii) a Node h represents a common occurrence at node x_iList of maximum information coefficientsMiddle node x_jFront and node x_jList of maximum information coefficientsMiddle node x_iThe previous class of nodes, m, represents the total number of such nodes.

4. The independence analysis and test based equipment parameter monitoring evidence network construction method according to claim 3, wherein the S2.3 implementation method is as follows:

s2.3.1 modified maximum information coefficient list for all nodesIn descending order, the maximum value is recorded as

S2.3.2 setting an isolation threshold factor gamma, taking gamma as 0, 0.3](ii) a If it is notThen node x_iFor an orphaned node, there is no undirected edge, otherwise, S2.3.3 is performed;

s2.3.3 if node x_iAnd node x_jThe corrected maximum information coefficient satisfies:

then node x_iAnd node x_jThere is no directional edge between them, where α is the connection threshold factor, and α E is taken as [0.7, 0.9 ]]。

5. The independence analysis and test based equipment parameter monitoring evidence network construction method according to claim 4, wherein the implementation method of S3 is as follows:

s3.1 at node x_iAnd node x_jRespectively constructing nonlinear regression models x on the premise of existence of undirected edges_i:＝f(x_j)+n₁And x_j:＝f(x_i)+n₂Model x_i:＝f(x_j)+n₁F (x) of (1)_j) Description node x obtained by finger fitting_jFor reasons, node x_iAs a resulting nonlinear regression function, n₁Finger model x_i:＝f(x_j)+n₁The noise term of (1); model x_j:＝f(x_i)+n₂F (x) of (1)_i) Description node x obtained by finger fitting_iFor reasons, node x_jAs a resulting nonlinear regression function, n₂Finger model x_j:＝f(x_i)+n₂The noise term of (1);

s3.2 calculating residual n of two models in S3.1 respectively₁＝x_i-f(x_j) And n₂＝x_j-f(x_i)，f(x_j) Finger node x_jBy substituting the node input data into a non-linear regression model x_i:＝f(x_j)+n₁Node x obtained later_iValue of (a), f (x)_i) Finger node x_iBy substituting the node input data into a non-linear regression model x_j:＝f(x_i)+n₂Node x obtained later_jThe value of (d);

s3.3 Pair x_iAnd n₁、x_jAnd n₂Carrying out T test on the two groups of variables respectively; by calculating test statisticsAndobtaining the corresponding t value, which is recorded as t₁And t₂Value, recheckReading the T test statistical table to obtain p values, and respectively recording the obtained p values as p₁And p₂(ii) a Wherein,representing a node x_jMean, σ, of the corresponding node input data₁Representing a node x_jThe standard deviation of the input data of the corresponding node,representing a node x_iMean, σ, of the corresponding node input data₂Representing a node x_iThe standard deviation of the corresponding node input data, n represents the sample number of the node input data collected by each node in the data set D;

s3.4 comparison of p₁And p₂If p is₁＞p₂Then accept model x_i:＝f(x_j)+n₁Consider node x_iAnd node x_jThe direction without the edge therebetween is the slave node x_jPointing to node x_i(ii) a If p is₂＞p₁Then accept model x_j:＝f(x_i)+n₂Consider node x_jAnd node x_iThe direction without the edge therebetween is the slave node x_iPointing to node x_j；

S3.5, determining the directions of all the non-directional edges according to S3.1 to S3.4, and checking whether a loop exists in the directed graph or not on the basis; if the loop exists, deleting the edge with the minimum p value; and thus, constructing the equipment parameter monitoring evidence network corresponding to the target equipment.