CN112085202A - Automobile fault diagnosis method based on hybrid Bayesian network - Google Patents

Abstract

The application discloses an automobile fault diagnosis method based on a hybrid Bayesian network, which comprises the following steps: and obtaining a hybrid Bayesian network by adopting a hybrid Bayesian network structure learning method combining a sequence generation method based on conditional entropy and a K2 algorithm fusing expert information, and then carrying out automobile fault diagnosis by using the hybrid Bayesian network. The performance effect same as that of the mutual information sequencing can be obtained under the condition of no expert knowledge, but the calculation amount is simpler and more convenient. A small amount of expert knowledge is blended into the K2 algorithm to obtain a better performance effect, and the deviation possibly caused by the missing of sample data or large difference of input sequence can be corrected to learn a better network structure. The problem of high computational complexity in the automobile fault diagnosis method is solved, so that the automobile fault diagnosis is quicker and the diagnosis result is more accurate.

Description

Vehicle Fault Diagnosis Method Based on Hybrid Bayesian Network

technical field

The invention relates to the fields of information technology and machine learning, and more particularly, to a vehicle fault diagnosis method based on a hybrid Bayesian network.

Background technique

With the development of automobile technology, the structure and electronic control part of the automobile are complex, the failure phenomenon is diversified, and the cause of the failure is complicated, random and fuzzy, resulting in a certain uncertainty of the failure.

Bayesian network is a causal graph model that can express the relationship between random variables based on probability theory and graph theory. Bayesian network has a solid theoretical foundation and is an effective method for dealing with uncertain knowledge expression and reasoning. It can deal with the expression and reasoning of uncertain problems in automobile fault diagnosis, and effectively integrate domain prior knowledge and real-time sensor data. The distribution characteristics of the auto fault diagnosis system can be realized.

When realizing automobile fault diagnosis based on Bayesian network, it is necessary to establish the structure of Bayesian network first. The scoring-based search method mainly includes two parts, namely the scoring function and the search algorithm. The search algorithm is used to continuously search the network structure, and each structure is scored through the scoring function. After the algorithm is completed, the network structure with the highest score is output, among which the most classic It is the K2 algorithm. The K2 algorithm takes the sample data and the artificially given order as input, starts with an empty node set and gradually increases the connection edges with the parent node and calculates the score, and outputs the network structure after the algorithm ends. The method based on dependency analysis mainly judges the mutual relationship between nodes and determines its direction through independence detection. With the increase of the number of nodes in this method, the number of independent detections of the algorithm will increase exponentially, resulting in large structural differences.

The classic K2 algorithm has the limitation of requiring human input to determine the order of nodes. At present, the method based on mutual information is used to generate the initial network graph and search the graph to obtain the corresponding node sequence as the input of the K2 algorithm. However, the method based on mutual information needs to generate an initial structure. The graph and the direction of judging the edge have high algorithm complexity.

SUMMARY OF THE INVENTION

The purpose of the present invention is to propose a vehicle fault diagnosis method based on a hybrid Bayesian network, so as to overcome the problem of high computational complexity in the existing vehicle fault diagnosis methods, and to make the vehicle fault diagnosis faster and the diagnosis result more accurate.

In order to achieve the above purpose, the proposed scheme is as follows:

1. A vehicle fault diagnosis method based on a hybrid Bayesian network, characterized in that it comprises the following steps:

Collect the information on the relationship between all nodes in the vehicle fault diagnosis problem provided by the group experts, use the D-S evidence theory to fuse the obtained node relationship, make decisions on the fusion results and convert the effective information after the decision into an adjacency matrix;

According to the obtained adjacency matrix as the prior structure, the prior structure is combined with the K2 algorithm, and the search space of the K2 algorithm is limited by the prior structure;

Calculate the entropy function of each node in the car fault data and sort to get the node sequence;

Inputting the node sequence into the K2 algorithm fused with the prior structure to learn the Bayesian network structure to obtain a hybrid Bayesian network structure;

Car fault diagnosis is performed based on the hybrid Bayesian network structure, and a car fault diagnosis result is obtained.

Preferably, the information on the relationship between all nodes in the vehicle fault diagnosis problem provided by the group experts is collected, and the obtained node relationship is fused using the D-S evidence theory, including:

In the process of collecting expert knowledge, there are three causal relationships between any two nodes (i, j), including: i→j means that the expert determines that there is a directed edge from i to j, and j→i means that the expert determines There is a directed edge that j points to i, and i≠j means that there is no pointing relationship between nodes;

Collect all expert knowledge and get the reliability distribution table of all experts;

Applying the D-S fusion rule to fuse the reliability allocation table to obtain a final fusion result;

Among them, if the two basic probability assignments of m ₁ and m ₂ are known, the DS fusion rules are as follows:

Among them, m(A) is the fusion result, A, B, C represent the propositions or assumptions in automobile fault diagnosis; φ represents the empty set, K represents the conflict coefficient, when K=0, it indicates the difference between m ₁ and m ₂ No conflict.

Preferably, the decision on the fusion result includes:

After obtaining the final fusion result m(A _h ), 1≤h≤4, according to the preset threshold θ, decide whether there is a corresponding directed edge. The decision-making method is as follows:

If m(A ₁ )>θ, then determine i→j;

If m(A ₂ )>θ, then determine j→i;

If m(A ₃ )>θ, then determine i≠j;

If m(A ₄ )>θ or m(A _h ) are both less than θ, it is considered that the relationship between the two nodes cannot be determined, and subsequent processing is treated as no expert information.

Preferably, calculating the entropy function of each node in the vehicle fault data and sorting to obtain a node sequence, including:

Calculate and sort the information entropy of each node in the vehicle fault data respectively, and select the first node as the starting node;

Calculate and sort the conditional entropy of the remaining nodes under the condition that the selected node is known, and select a node as the Nth order node;

Repeat the calculation of the above conditional entropy until the final sequence point is obtained. The selected nodes in each calculation of the conditional entropy include the nodes selected before this calculation, wherein N is a positive integer greater than 1 and less than the total number of nodes;

According to the starting node and each sequence node, a node sequence is obtained.

Preferably, when learning the Bayesian network structure, the Bayesian information metric BIC score is used as the scoring function.

Preferably, inputting the node sequence into the K2 algorithm fused with prior structure for Bayesian network structure learning includes:

Correction processing is performed on illegal structures obtained in the search process, where the illegal structures are structures that do not conform to the properties of the acyclic graph.

Preferably, the correction includes:

If there is a bidirectional arc between two nodes in the matrix, select one of the two at random and set it to 0;

If there is a closed loop structure in the matrix, delete edges or reverse edges randomly in the loop to meet the requirement that the Bayesian network is a directed acyclic graph.

Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

1) The present invention adopts the principle of entropy function to obtain the node sequence order from the data, which avoids the uncertainty of the artificial input sequence and the complexity of obtaining the node sequence order by using the directed spanning tree algorithm in the prior art.

2) The present invention adopts the idea of integrating the initial structure into the classical k2 algorithm, which greatly reduces the network search and algorithm running time.

Description of drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only These are some embodiments of the present application. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without any creative effort.

1 is a flowchart of a Bayesian network structure learning method based on knowledge and data mixing according to an embodiment of the present invention;

Fig. 2 is the corresponding relation diagram of the network structure and the adjacency matrix in the embodiment of the present invention;

3 is a schematic structural diagram of a Bayesian network in an embodiment of the present invention;

FIG. 4 is an example diagram of an illegal structure in an embodiment of the present invention.

Detailed ways

Based on the K2 algorithm, the invention proposes a new hybrid Bayesian network structure learning method based on the fusion of knowledge and data by integrating expert knowledge and node pre-sorting methods, and the obtained hybrid Bayesian network can be used for Car fault diagnosis.

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

Information entropy (entropy function) is an effective method to deal with uncertain information. When uncertain information is expressed by probability, information entropy can be converted into Shannon entropy. Shannon entropy is widely used in the field of information processing.

D-S evidence theory (Dempster-Shafer), also known as belief function theory, has been greatly developed in theory and application since it was put forward, and it is an effective tool for dealing with uncertain information at present.

Referring to FIG. 1 , it shows a flowchart of an automobile fault diagnosis method based on a hybrid Bayesian network according to an embodiment of the present invention. Assuming that the engine is difficult to start cold at a temperature of 20 to 25 degrees Celsius is the parent node, and its associated child nodes are fuel quality, spark plug, water temperature sensor, fuel pressure regulator, fuel pump circuit, fuel pipe, and cylinder compression pressure. The above components are composed of Bayesian network diagram.

Include the following steps:

S1. Collect the information on the relationship between all nodes in the vehicle fault diagnosis provided by the group experts, use the D-S evidence theory to fuse the obtained node relationship, make a decision on the fusion result and convert the effective information after the decision into an adjacency matrix.

Through the collection of group expert knowledge on the relationship between all nodes in the problem of automobile fault diagnosis, the D-S evidence theory is used for fusion, and accurate prior information is obtained as the known causal relationship.

In the process of collecting expert knowledge, there are three causal relationships between any two nodes (i, j), that is, i→j means that the expert determines that there is a directed edge from i to j, and j→i means that the expert determines that there is a directed edge. j points to the directed edge of i, i≠j indicates that there is no pointing relationship between nodes. After all the knowledge collection is completed, the reliability distribution of all experts can be obtained as shown in Table 1.

Table 1

i→j j→i i≠j uncertain E₁ m₁(A₁) m₁(A₂) m₁(A₃) m₁(A₄) E₂ m₂(A₁) m₂(A₁₂) m₂(A₃) m₂(A₄) … … … … … E_n m_n(A₁) m_n(A₂) m_n(A₃) m_n(A₄) DS m(A₁) m(A₂) m(A₃) m(A₄)

where E _k (1≤k≤n) represents the kth expert, A ₁ , A ₂ ......A _h represents any hypothesis or proposition in the automobile fault diagnosis problem, m ₁ , m ₂ ...... ... m _n represents the probability distribution of the hypothesis or proposition. After obtaining the opinion distribution of all n experts, apply the DS rule for fusion, and after applying the DS rule fusion to the n expert opinions, the final fusion result is obtained, DS fusion The rules are as follows:

Let the recognition frame (sample space) Ω be a set of mutually independent but overall exhaustive events as follows:

Ω={θ ₁ , θ ₂ , ..., θ _i , ..., θ _n };

The power set of Ω 2 ^Ω represents all hypotheses or propositions, expressed as:

For a specific recognition framework Ω, its basic information distribution function is the mapping of 2θ on [ ⁰ , 1];

其中φ表示空集。m: 2 ^θ → [0, 1]; satisfy:

where φ represents the empty set.

Given the _two basic probability assignments _of m1 and m2, the combination rules of DS evidence theory are as follows.

Among them, m(A) is the fusion result, A, B, C respectively represent propositions or assumptions in automobile fault diagnosis; φ represents the empty set, K represents the conflict coefficient, when K=0, it means that the difference between m ₁ and m ₂ There is no conflict between.

After obtaining the final fusion result m(A _h ), 1≤h≤4, according to the preset threshold θ, decide whether there is a corresponding directed edge. The decision-making method is as follows:

If m(A ₁ )>θ, then determine i→j;

If m(A ₂ )>θ, then determine j→i;

If m(A ₃ )>θ, then determine i≠j;

If m(A ₄ )>θ or m(A _h ) are both less than θ, it is considered that the relationship between the two nodes cannot be determined, and subsequent processing is treated as no expert information.

The clear inter-node relationship after fusion is given in the form of an adjacency matrix, and the corresponding relationship between the network structure and the adjacency matrix is shown in Figure 2.

If the multi-expert information is fused, it can be accurately obtained that there is a directed edge from node 1 to node 2, and there is no connected edge between node 3 and node 5, node 4 and node 8, that is, 1→2, node 3≠5, node 4≠ 8. Convert the result of multi-expert knowledge fusion into the form of adjacency matrix A:

S2. According to the obtained adjacency matrix as a priori structure, the priori structure is combined with the K2 algorithm, and the search space of the K2 algorithm is limited by the priori structure.

In the implementation process of the original K2 algorithm, for each node, starting from an empty node set, according to the pre-input node sequence relationship, the connection edge with the parent node is continuously increased, and the network score changes are judged. If the score increases, the node is input. An empty set of parent nodes. In the embodiment of the present invention, 1→2 in the adjacency matrix after fusion means that node 1 must be included in the parent node set of node 2. Therefore, the process of searching for parent nodes for some nodes during the improvement process will no longer start from an empty node set. For node 3 and node 5, node 4 and node 8, there is clearly no directed edge. If in the subsequent search process, even adding the directed edge can make the overall score of the entire network higher, according to the prior network The pre-set of , forcibly cancels the directed edge.

S3. Calculate the entropy function of each node in the vehicle fault data and sort to obtain a node sequence.

The specific implementation can be:

Calculate and sort the information entropy of each node in the vehicle fault data respectively, and select the first node as the starting node;

Calculate and sort the conditional entropy of the remaining nodes under the condition that the selected node is known, and select a node as the Nth order node;

Repeat the calculation of the above conditional entropy until the final sequence point is obtained, the selected nodes in each calculation of the conditional entropy include the nodes selected before this calculation, wherein N is a positive integer greater than 1 and less than the total number of nodes;

According to the starting node and each sequence node, a node sequence is obtained.

Among them, the information entropy can be calculated as follows:

Suppose X is a discrete random variable, and the possible values of X are {x ₁ , x ₂ ... x _n }, then the Shannon entropy is expressed as:

where p(x) is the marginal probability distribution function of X.

Conditional entropy can be calculated as follows:

Let p(x, y) be the joint distribution function of discrete variables X and Y, and its possible values are {(x ₁ , y ₁ ), (x ₂ , y ₂ ),...(x _n , y _n ) }, then the joint distribution function is expressed as:

Under the condition of given X, the information entropy of Y is called the conditional entropy, then the conditional entropy can be expressed as:

where p(x, y) is the joint probability distribution function of X and Y, and p(x) is the marginal probability distribution function of X.

Conditional entropy can also be expressed as:

H(Y|X)=H(X,Y)-H(X);

Referring to FIG. 3 , it shows a Bayesian network structure for automobile fault diagnosis in an embodiment of the present invention, where the Bayesian network structure includes 8 nodes. Taking the Bayesian network structure as an example to illustrate the specific process of obtaining the node sequence, as shown below:

(1) Calculate and sort the information entropy of 8 nodes respectively, the result is: H(1)<H(3)<H(5)<H(4)<H(7)<H(6)<H( 8)<H(2), if H(1) is the smallest after calculation, select node 1 as the starting node.

(2) Calculate the conditional entropy of the remaining nodes under the known conditions of node 1 and sort them. The result is: H(2|1)<H(4|1)<H(3|1)<H(5|1)<H(6|1)<H(7|1)<H(8| 1), select node 2 as the second sequential node.

(3) Calculate and sort the conditional entropy under the known conditions of nodes 1 and 2. The result is: H(4|1,2)<H(3|1,2)<H(5|1,2)< H(6|1,2)<H(7|1,2)<H(8|1,2), select node 4 as the third sequence point.

(4) Calculate and sort the conditional entropy when nodes 1, 2, and 4 are known. The result is: H(3|1,2,4)<H(5|1,2,4)<H(6| 1,2,4)<H(7|1,2,4)<H(8|1,2,4), then node 3 is the fourth sequence point.

(5) Calculate and sort the conditional entropy when nodes 1, 2, 4, and 3 are known, and the result is: H(6|1,2,4,3)<H(5|1,2,4,3 )<H(7|1,2,4,3)<H(8|1,2,4,3), then node 6 is the fifth sequence point.

(6) Calculate and sort the conditional entropy under known nodes 1, 2, 4, 3, and 6. The result is: H(5|1,2,4,3,6)<H(8|1,2 ,4,3,6)<H(7|1,2,4,3,6), then node 5 is the sixth sequence point.

(7) Calculate and sort the conditional entropy when nodes 1, 2, 4, 3, 6, and 5 are known, and the result is: H(7|1,2,4,3,6,5)<H(8 |1,2,4,3,6,5), then node 7 is the seventh sequence point, and node 8 is the last sequence point.

The obtained node sequence is: node 1 - node 2 - node 4 - node 3 - node 6 - node 5 - node 7 - node 8.

S4, inputting the node sequence into the K2 algorithm fused with the prior structure to learn the Bayesian network structure to obtain a hybrid Bayesian network structure;

The resulting node sequence is input into the K2 algorithm fused with expert knowledge. (changing the node order of the same level does not affect the learning result of the K2 algorithm), in the embodiment of the present invention, the Bayesian information metric BIC score is used as the scoring function:

score _BIC (G:D)=score _MLE (G:D)-0.5logM*DIM[G];

代表网络结构的维度。where score _MLE (G:D) is the maximum likelihood scoring function,

Represents the dimension of the network structure.

When using the improved K2 algorithm for structure learning, the structure state obtained in the search process may become infeasible because it no longer conforms to the properties of the acyclic graph. At this time, the illegal structure (as shown in Figure 4) needs to be corrected. Steps as follows:

If two nodes in the matrix are mutually dependent, it is unreasonable to have a bidirectional arc, and one of the two can be randomly selected and set to 0;

If there is a closed loop structure in the matrix, the deletion edge or the reverse edge should be randomly selected in the loop to meet the requirement that the Bayesian network is a directed acyclic graph.

S5. Perform automobile fault diagnosis based on the hybrid Bayesian network structure to obtain an automobile fault diagnosis result.

The embodiment of the present invention adopts a hybrid Bayesian network structure learning method combining the conditional entropy-based sequence generation method and the K2 algorithm fused with expert information. In the case of no expert knowledge, the same performance effect as using mutual information sorting can be obtained, but the amount of calculation is simpler. On top of this, the method proposed in the present invention is obtained by adding the expert knowledge in the matrix A. Compared with the MWST and MMHC algorithms that only use data for sorting and searching, a small amount of expert knowledge can be integrated into the K2 algorithm to obtain better performance. , and can correct the deviations that may be caused by missing sample data or large differences in input order to learn a better network structure.

The following is to verify the improved K2 algorithm in the implementation of the present invention:

In order to avoid the randomness of the process, the algorithm is run ten times and the edge addition (IE), the edge subtraction (ME), the number of reverse edges (RE) and the Hamming distance (SHD) are selected as the evaluation indicators. The maximum and minimum hill climbing method (MMHC), maximum spanning tree algorithm (MWST), mutual information sorting (MI) and conditional entropy sorting (Con-En) are compared to verify the hybrid Bayesian network of fusion knowledge and data provided by the present invention. The advantages of structure learning methods are shown in Table 2.

Table 2

On this basis, by continuously increasing the amount of expert knowledge, more inter-node relationships can be determined after fusion, the corresponding results can be recorded, and the principle of the invention can be further verified, as shown in Table 3.

table 3

Determine the number of relationships IE ME RE 3 0.2 0.1 1.2 5 0.3 0 0.8 7 0 0 0.2

It can be seen from the learning results in Table 3 that as the number of experts increases, more relationships between nodes can be determined, the performance of Bayesian network structure learning continues to improve, and the learned network structure is closer to the original structure, that is, increasing the number of experts. The amount of knowledge can improve the accuracy of learning results. Therefore, it can be seen from the above analysis that expert knowledge plays an important role in Bayesian network structure learning.

Finally, it should also be noted that in this document, relational terms such as first and second are used only to distinguish one entity or operation from another, and do not necessarily require or imply these entities or that there is any such actual relationship or sequence between operations. Moreover, the terms "comprising", "comprising" or any other variation thereof are intended to encompass non-exclusive inclusion such that a process, method, article or device comprising a list of elements includes not only those elements, but also includes not explicitly listed or other elements inherent to such a process, method, article or apparatus. Without further limitation, an element qualified by the phrase "comprising a..." does not preclude the presence of additional identical elements in a process, method, article or apparatus that includes the element.

The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same and similar parts between the various embodiments can be referred to each other.

The above description of the disclosed embodiments enables any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, this application is not intended to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (7)

1. a vehicle fault diagnosis method based on hybrid Bayesian network, is characterized in that, comprises the following steps: Collect the information on the relationship between all nodes in the vehicle fault diagnosis problem provided by the group experts, use the D-S evidence theory to fuse the obtained node relationship, make decisions on the fusion results and convert the effective information after the decision into an adjacency matrix; According to the obtained adjacency matrix as the prior structure, the prior structure is combined with the K2 algorithm, and the search space of the K2 algorithm is limited by the prior structure; Calculate the entropy function of each node in the car fault data and sort to get the node sequence; Inputting the node sequence into the K2 algorithm fused with the prior structure to learn the Bayesian network structure to obtain a hybrid Bayesian network structure; Car fault diagnosis is performed based on the hybrid Bayesian network structure, and a car fault diagnosis result is obtained. 2. The method according to claim 1, wherein the collection of information on the relationship between all nodes in the vehicle fault diagnosis problem provided by group experts, adopts D-S evidence theory to fuse the acquired node relationship, comprising: In the process of collecting expert knowledge, there are three causal relationships between any two nodes (i, j), including: i→j means that there is a directed edge from i to j determined by experts, and j→i means determined by experts There is a directed edge that j points to i, and i≠j means that there is no pointing relationship between nodes; Collect all expert knowledge and get the reliability distribution table of all experts; Applying the D-S fusion rule to fuse the reliability allocation table to obtain a final fusion result; Among them, if the two basic probability assignments of m ₁ and m ₂ are known, the DS fusion rule is as follows:

Among them, m(A) is the fusion result, A, B, C respectively represent the proposition or hypothesis in the automobile fault diagnosis problem; Φ represents the empty set, K represents the conflict coefficient, when K=0, it means that the difference between m ₁ and m ₂ There is no conflict between. 3. The method according to claim 2, wherein the decision on the fusion result comprises: After obtaining the final fusion result m(A _h ), 1≤h≤4, according to the preset threshold θ, decide whether there is a corresponding directed edge. The decision-making method is as follows: If m(A ₁ )>θ, then determine i→j; If m(A ₂ )>θ, then determine j→i; If m(A ₃ )>θ, then determine i≠j; If m(A ₄ )>θ or m(A _h ) are both less than θ, it is considered that the relationship between the two nodes cannot be determined, and subsequent processing is treated as no expert information. 4. The method according to claims 1 to 3, wherein calculating the entropy function of each node in the vehicle fault data and sorting to obtain a node sequence, comprising: Calculate and sort the information entropy of each node in the vehicle fault data respectively, and select the first node as the starting node; Calculate and sort the conditional entropy of the remaining nodes under the condition that the selected node is known, and select a node as the Nth order node; Repeat the calculation of the above conditional entropy until the final sequence point is obtained, the selected nodes in each calculation of the conditional entropy include the nodes selected before this calculation, wherein N is a positive integer greater than 1 and less than the total number of nodes; According to the starting node and each sequence node, a node sequence is obtained. 5 . The method according to claim 1 , wherein when learning the Bayesian network structure, the Bayesian information metric BIC score is used as the scoring function. 6 . 6. The method according to claim 1, wherein, inputting the node sequence into the K2 algorithm fused with prior structure for Bayesian network structure learning, comprising: Correction processing is performed on illegal structures obtained in the search process, where the illegal structures are structures that do not conform to the properties of the acyclic graph. 7. The method of claim 6, wherein the correction comprises: If there is a bidirectional arc between two nodes in the matrix, select one of the two at random and set it to 0; If there is a closed loop structure in the matrix, delete edges or reverse edges randomly in the loop to meet the requirement that the Bayesian network is a directed acyclic graph.

Images