CN110443344B - Momentum wheel fault diagnosis method and device based on K2ABC algorithm - Google Patents

Momentum wheel fault diagnosis method and device based on K2ABC algorithm Download PDF

Info

Publication number
CN110443344B
CN110443344B CN201910535762.4A CN201910535762A CN110443344B CN 110443344 B CN110443344 B CN 110443344B CN 201910535762 A CN201910535762 A CN 201910535762A CN 110443344 B CN110443344 B CN 110443344B
Authority
CN
China
Prior art keywords
algorithm
food source
node
momentum wheel
new
Prior art date
Legal status (The legal status 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 status listed.)
Active
Application number
CN201910535762.4A
Other languages
Chinese (zh)
Other versions
CN110443344A (en
Inventor
戴光明
彭雷
王茂才
宋芳然
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China University of Geosciences
Original Assignee
China University of Geosciences
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 China University of Geosciences filed Critical China University of Geosciences
Priority to CN201910535762.4A priority Critical patent/CN110443344B/en
Publication of CN110443344A publication Critical patent/CN110443344A/en
Application granted granted Critical
Publication of CN110443344B publication Critical patent/CN110443344B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/01—Probabilistic graphical models, e.g. probabilistic networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Algebra (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a momentum wheel fault diagnosis method and a momentum wheel fault diagnosis device based on a K2ABC algorithm, which are based on the K2ABC algorithm combining an artificial bee colony algorithm and a K2 algorithm and are applied to a momentum wheel fault diagnosis modeling method, so that the defects of complex calculation, inflexible model and the like existing in the conventional momentum wheel fault diagnosis modeling method are overcome, and the fault diagnosis result of a momentum wheel is more accurate.

Description

Momentum wheel fault diagnosis method and device based on K2ABC algorithm
Technical Field
The invention relates to the field of satellite momentum wheel fault diagnosis, and simultaneously relates to the field of Bayesian network structure learning in computer technology, in particular to a K2ABC algorithm based on combination of an artificial bee colony algorithm and a K2 algorithm and a device based on the algorithm, wherein the algorithm can reduce the dependence of the K2 algorithm on node sequences so as to improve the accuracy of the K2 algorithm in momentum wheel fault diagnosis modeling.
Background
The momentum wheel is a key component of a satellite attitude control system, and the uncertainty existing in the satellite momentum wheel fault cannot be better solved by the traditional satellite momentum wheel fault diagnosis technology such as an observer method and an expert system method. The Bayesian network as a probability network has obvious advantages in reasoning uncertainty problems, and the model established by Bayesian network structure learning influences the accuracy of the final reasoning result. The Bayesian network structure learning is an NP difficult problem, the K2 algorithm is a commonly used Bayesian network structure learning algorithm, but the K2 algorithm has the problem of being excessively dependent on the order of model nodes. Since the K2 algorithm needs to give a node ranking before running, the ranking is in most cases unknown and is usually determined according to expert experience, but the expert experience cannot guarantee objectivity and accuracy, and the ranking is difficult to realize when the number of nodes is large. In addition, experiments show that the node ranking can affect the learning effect of the K2 algorithm, the better node ranking can enable the K2 algorithm to learn a network structure which is more suitable for sample data, and the poorer node ranking can cause the learning effect of the K2 algorithm to be reduced. Therefore, the research on how to obtain the better node order has a positive influence on the K2 algorithm and the Bayesian network structure learning, so that the satellite momentum wheel fault can be better diagnosed.
Disclosure of Invention
The invention aims to provide a K2ABC algorithm and a device based on combination of an artificial bee colony algorithm and a K2 algorithm, and the K2ABC algorithm and the device are applied to a momentum wheel fault diagnosis modeling method so as to overcome the defects of complex calculation, inflexible model and the like of the conventional momentum wheel fault diagnosis modeling method and enable the fault diagnosis result of the momentum wheel to be more accurate.
The technical scheme adopted by the invention for solving the technical problem is as follows: a momentum wheel fault diagnosis method based on a K2ABC algorithm is constructed, and comprises the following steps:
s1, establishing a Bayesian network model of the momentum wheel fault based on the K2ABC algorithm and by combining the momentum wheel fault data;
s2, based on the Bayesian network model established in the step S1, utilizing Bayesian network parameter learning to obtain parameters of each network node in the model;
s3, based on the model with the node parameters obtained in the step S2 and the obtained evidence information, carrying out probability calculation on various reasons which may cause the momentum wheel to break down by adopting a Bayesian network inference algorithm; the evidence information refers to adding state information to a certain node or a plurality of nodes during application to reason about the influence on the states of other nodes;
and S4, finding out the node with the maximum posterior probability by comparing the probabilities calculated in the reasoning stage, and taking the node as a final diagnosis result.
Further, in the momentum wheel fault diagnosis method based on the K2ABC algorithm of the present invention, the K2ABC algorithm in step S1 includes the following steps:
s101, acquiring momentum wheel fault data which are discrete data;
s102, extracting variables in fault data to serve as nodes in a network model to be built;
s103, setting algorithm execution parameters;
s104, randomly initializing the bee colony, and calculating the fitness value of the initial bee colony;
s105, the hiring bee searches for new food sources, calculates the fitness value of each food source by using a fitness function based on K2 search and correlation, and selects the food sources needing to be reserved according to a greedy mechanism;
s106, calculating the selection probability of each food source according to the following formula:
Figure BDA0002101122470000031
wherein p isiProbability of being selected for the ith food source, fitiThe fitness function value corresponding to the ith food source is max (fit), and the max (fit) is the maximum value of the fitness function values corresponding to all the food sources;
s107, selecting a food source to be followed by the following bee by roulette according to the probability calculated in the step S106, further searching for a new food source, then calculating each fitness value by utilizing a fitness function based on K2 searching and relevance, and reserving the food source with higher fitness value;
s108, judging the size relation between the continuous non-updating times and the maximum limiting times of the sequence, and if the continuous non-updating times are larger than the maximum limiting times, eliminating the corresponding food sources and searching new food sources for substitution by the scout bees;
s109, recording the current optimal food source;
s110, if the termination condition of the algorithm is met, outputting the current optimal food source; otherwise, go to step S105.
Further, in the momentum wheel fault diagnosis method based on the K2ABC algorithm of the present invention, in step S11, if the momentum wheel fault data is non-discrete data, the momentum wheel fault data is discretized first to form discrete pair momentum wheel fault data.
Further, in the momentum wheel fault diagnosis method based on the K2ABC algorithm of the present invention, the method for searching for a new food source is: obtaining a new food source by adopting a domain search strategy, and then sequentially carrying out reverse operator operation and mutual information-based immune operator operation on the new food source to obtain a final new food source individual; the domain search strategy refers to a domain search strategy adopting three-point exchange: generating a new solution through local change on the basis of an original food source, wherein the specific method comprises the steps of randomly generating three exchange points 1,2 and 3 on an original variable sequence, and sequentially exchanging numbers of the three points, namely, original data of a 2 nd exchange point is used as new data of a 1 st exchange point, original data of a 3 rd exchange point is used as new data of the 2 nd exchange point, and original data of the 1 st exchange point is used as new data of the 3 rd exchange point;
the reversal operator is to randomly generate two different sites in the variable sequencing and carry out reverse sequence operation on the variable sequence between the two sites;
the specific process of updating the food source based on the immune operator is as follows:
step 1: in the sequence of variables S ═ x1,x2,…,xn]Optionally a node xiObtaining a new sequence S to be adjusted1＝[xi,xi+1,…,xn]And a reserved sequence S2＝[x1,x2,…,xi-1]；
steo 2: from the new sequence S1The first node of the node is started, the node with the maximum mutual information value with the current node is searched in the rest nodes, and the node is inserted behind the current node;
step 3: repeating the above steps until a new sequence S1Finishing the adjustment of all the nodes in the node B;
step 4: and combining the adjusted new sequence with the reserved sequence to obtain an adjusted variable sequence.
Further, in the momentum wheel fault diagnosis method based on the K2ABC algorithm of the present invention, in step S107, the searching for a fitness function based on the K2 and the correlation to calculate each fitness value specifically includes:
step 1: after the food sources are initialized, scoring evaluation is carried out on the food sources by adopting K2 search, and the current optimal food sources are recorded;
step 2: in the stage of hiring bees and following bees, solving the correlation degree r of sample data corresponding to the searched new food source and the current optimal food source;
step 3: if the correlation r is larger than the threshold value delta, adopting K2 to search and calculate the fitness function value of the new food source, and performing subsequent operation; otherwise, the new food source is discarded, the original food source is retained, and step2 is transferred.
step 4: in the scout bee phase, a K2 search is used to score new food sources.
Further, in the momentum wheel fault diagnosis method based on the K2ABC algorithm of the present invention, in step S107, the food source with a higher remaining fitness value specifically refers to: the correlation degree of the sample data corresponding to the current variable sequence and the current optimal variable sequence is solved, and if the correlation degree is smaller than a given threshold value delta, the variable sequence is abandoned, so that the sequence is prevented from being scored by repeatedly calling K2 for searching.
Furthermore, in the momentum wheel fault diagnosis method based on the K2ABC algorithm, in order to reduce the complexity of food source coding, a serial number coding mode is adopted, and the node serial number x is usediAs a unique identification of the variable.
The invention also provides a momentum wheel fault diagnosis device based on the K2ABC algorithm, which is provided with a computer storage medium, wherein the computer storage medium stores computer executable instructions, and the computer executable instructions are used for realizing the momentum wheel fault diagnosis method based on the K2ABC algorithm.
The momentum wheel fault diagnosis method and device based on the K2ABC algorithm have the following beneficial effects: the K2ABC algorithm can better fit sample data and learn a better network structure, so that the satellite momentum wheel fault can be better diagnosed.
Drawings
The invention will be further described with reference to the following drawings and examples, in which:
FIG. 1 is a flow chart based on the K2ABC algorithm;
FIG. 2 is a model diagram based on the K2ABC algorithm;
FIG. 3 is a diagram of a domain search strategy implementation process;
FIG. 4 is an inversion operator implementation.
Detailed Description
For a more clear understanding of the technical features, objects and effects of the present invention, embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
Introduction to K2 Algorithm
The K2 algorithm starts with an undirected graph that contains all nodes but no edges, and during the search, the algorithm compares the nodes in the variable ordering in turn. In variable sequencing, the current node is only possibly the father node of the subsequent node, so that in the process of searching the father node, the algorithm only needs to traverse the network structure formed by the nodes arranged in front of the current node, and the nodes behind the current node do not need to be considered. Meanwhile, the number of father nodes of any variable in the network structure is set by the algorithm to be not more than the given maximum number of father nodes, and when the score of the network structure formed by each variable and the father node of the variable is maximum, the final score of the network structure is also maximum.
The specific implementation process of the algorithm is as follows: the K2 algorithm starts from the first node in the variable ordering, searches the node which makes the network structure score the maximum for each node in turn as the candidate father node, and then adds the score value P of the father nodenewAnd the value of credit P of the parent node is not addedoldMaking a comparison if Pnew＞PoldIf the candidate node is taken as the father node of the candidate node, an edge is added between the two nodes, otherwise, all the remaining nodes are continuously traversed, and the termination condition of the algorithm is that the father node number of each node reaches a given value or the score value P of the network structurenewNo longer increased.
2. Artificial bee colony algorithm related introduction
An artificial bee colony algorithm, called ABC algorithm for short, is an intelligent optimization algorithm which is provided by simulating the bee honey collection behavior in nature. In 2005, the ABC algorithm was proposed by kariboga and successfully applied to the function optimization problem, and in 2006, the ABC algorithm was successfully applied to solve the TSP problem. The algorithm has the characteristics of strong optimizing capability, simplicity, easy realization and the like, and is widely concerned in the field of group intelligent optimization algorithms at present.
In the bee colony intelligent search model, three basic components of a food source, a hired bee and a non-hired bee are mainly contained. Food sources are determined by various factors, usually evaluated by the "revenue ratio", and non-hired bees mainly comprise two types, namely scout bees and follower bees, and are mainly used for searching and exploiting food sources.
Hiring bees: also called leading bees, are used for storing relevant information of a certain food source, and then hire bees to share the information with other bees with a certain probability so as to improve the overall honey collection benefit. Hiring bees will decide to lead the role based on the profitability of the bees. Bees with high profit ratios can recruit more following bees, thereby enabling the overall algorithm to converge quickly to the optimum.
Detecting bees: and searching a new food source nearby the food source, and acquiring nectar profit information of the related food source. When the algorithm is in local optimum, the scout bees can effectively jump out and perform scout of new food sources, and the jumping-out condition is that the bee colony does not find a better food source in a period of time.
Bee following: whether the hiring bees are followed is determined near the honeycomb according to the income ratio of the food sources, the following bees can strengthen the traversal of excellent food sources in the hiring bees, the elite effect of the hiring bees is highlighted, and the convergence of the algorithm is accelerated.
3. Momentum wheel fault diagnosis modeling method based on K2ABC algorithm
The momentum wheel fault diagnosis method based on the K2ABC algorithm specifically comprises the following steps:
s1, establishing a Bayesian network model of the momentum wheel fault based on the K2ABC algorithm and by combining the momentum wheel fault data;
table 1 shows the identifier corresponding to each node in the model and the value of each node, which are used to form momentum wheel fault data. The X node represents whether the momentum wheel fails or not, A, B, C three nodes represent three common failure modes of the momentum wheel respectively, six nodes such as D-I represent failure modes of four components of the momentum wheel, twelve nodes such as J-U represent causes which may cause the momentum wheel to fail, namely failure modes of all parts in the momentum wheel.
Table 1 network model node description
Figure BDA0002101122470000071
S2, based on the Bayesian network model established in the step S1, utilizing Bayesian network parameter learning to obtain parameters of each network node in the model;
s3, based on the model with the node parameters obtained in the step S2 and the obtained evidence information, carrying out probability calculation on various reasons which may cause the momentum wheel to break down by adopting a Bayesian network inference algorithm; the evidence information refers to adding state information to a certain node or a plurality of nodes during application to reason about the influence on the states of other nodes;
and S4, finding out the node with the maximum posterior probability by comparing the probabilities calculated in the reasoning stage, and taking the node as a final diagnosis result.
Aiming at the defects of the K2 algorithm, the invention provides a K2ABC algorithm which combines an artificial bee colony algorithm with a K2 algorithm. The algorithm mainly improves the food source updating strategy and the fitness function of the artificial bee colony algorithm. An immune operator based on mutual information is provided in the aspect of a food source updating strategy, and a fitness function based on K2 search and correlation is provided in the aspect of the fitness function. The algorithm can enable the K2 algorithm to learn a network structure which fits sample data better under the condition that node sequencing is not given in advance, and the problem that the K2 algorithm depends on the node sequencing is solved.
Fig. 2 shows a flow chart of the K2ABC algorithm. The algorithm firstly needs to initialize various main parameters and food sources, the searching process mainly comprises three stages of employing bees, following bees and detecting bees, fitness functions based on K2 searching and relevance are adopted to score the food sources in the searching process, the food sources are updated by combining an updating strategy, and when a termination condition is met, the algorithm directly returns the found optimal network model and the corresponding variable sequencing thereof
Referring to FIG. 1, the K2ABC algorithm includes the following steps:
s101, acquiring momentum wheel fault data which are discrete data; if the momentum wheel fault data are non-discrete data, discretization processing is firstly carried out on the momentum wheel fault data to form discrete momentum wheel fault data.
And S102, extracting variables in the fault data to serve as nodes in the network model to be built.
S103, setting algorithm execution parameters.
S104, randomly initializing the bee colony, and calculating the fitness value of the initial bee colony. The initialization of bee colony is food source initialization, specifically, a plurality of food sources (solutions) are randomly generated according to parameter setting, namely, a plurality of full arrays from 1 to n are randomly generated. In order to reduce the complexity of food source coding, a serial number coding mode is adopted in the text, and the node serial number x is considerediAs a unique identification of the variable, the solution to the variable ordering problem for n nodes can be expressed as S ═ xi]I is 1,2, …, n. For example, there is a variable ordering problem with 10 nodes, where a solution can be expressed as S ═ 35186102749]. The advantage of sequence number coding is that: 1) legal constraints of variable ordering problem are implied in food source initialization and other basic operations, namely node numbers do not appear repeatedly; 2) the generation of illegal network structures cannot be caused in other subsequent operations; 3) the coding mode is simple and easy to realize.
S105, the hiring bee sequentially searches for new food sources by adopting a domain search strategy, a reverse operator and mutual information, calculates the fitness value of each food source by utilizing a fitness function based on K2 search and correlation, and selects the food sources needing to be reserved according to a greedy mechanism.
The domain search strategy adopts a three-point exchange domain search strategy, and a new solution can be generated through local change on the basis of the original food source. The specific method is to randomly generate three exchange points 1,2 and 3 on the original variable sequence, and exchange numbers at the three points in turn, namely 1 ← 2,2 ← 3 and 3 ← 1. The specific exchange process is shown in fig. 3, taking the ordering problem with 10 nodes as an example.
The reversal operator operation means that two different sites are randomly generated in variable sequencing, and the variable sequence between the two sites is subjected to reversal operation, and the operator is favorable for small-range migration of an algorithm solution. The specific implementation process is shown in fig. 4.
Aiming at the variable ordering problem, an immune operator based on mutual information is used for updating the variable sequence. Mutual information expresses the strength of the dependency relationship between two variables, and the larger the mutual information is, the stronger the correlation relationship between the two variables is. Therefore, in the variable ordering problem, "antigen" can be regarded as a variable sequence with a small mutual information value, and "vaccination" is a process of adjusting the mutual information value of the variable sequence according to a certain rule. The specific process of updating the food source based on the immune operator will be described in detail below.
step 1: in the variable sequence S ═ x1,x2,…,xn]Optionally a node xiObtaining a new sequence S to be adjusted1＝[xi,xi+1,…,xn]And a reserved sequence S2＝[x1,x2,…,xi-1]；
step 2: from the new sequence S1The first node of the node is started, the node with the maximum mutual information value with the current node is searched in the rest nodes, and the node is inserted behind the current node;
step 3: repeating the above steps until a new sequence S1Finishing the adjustment of all the nodes in the node B;
step 4: combining the adjusted new sequence with the reserved sequence to obtain an adjusted variable sequence Snew。
Compared with the original sequence, the adjusted variable sequence can not only keep partial node sequencing of the original sequence, but also increase mutual information values among the sequence nodes, thereby improving the dependency relationship among the nodes and updating the variable sequence towards a more optimal direction.
S106, calculating the selection probability of each food source according to the following formula:
Figure BDA0002101122470000101
wherein p isiProbability of being selected for the ith food source, fitiThe fitness function value corresponding to the ith food source is max (fit), and the max (fit) is the maximum value of the fitness function values corresponding to all the food sources;
s107, the follower bees select food sources to be followed by roulette according to the probability calculated in the step S106, further search new food sources by sequentially adopting a domain search strategy, a reverse operator and an immune operator based on mutual information, then calculate each fitness value by utilizing a fitness function based on K2 search and correlation degree, and reserve the food sources with higher fitness values.
The artificial bee colony algorithm usually adopts a roulette method to select food sources for improvement in the bee following stage, and for the variable sorting problem, the larger the fitness function value is, the more the variable sequence can learn the correct network structure, so the maximum value of the fitness function is required to be searched in the invention. Since the K2-based search and relevance function selected by the invention evaluates to a negative number, the probability calculation formula in step S106 is defined, causing the superior food source to be selected by the following bees with a greater probability.
In fact, the best method for evaluating the variable sequences is to substitute the variable sequences into the K2 algorithm for learning, and evaluate the sequences according to the advantages and disadvantages of the learned network structures, so that the K2 search can be introduced into the fitness function of the variable sequences, and the score of the network structure corresponding to each variable sequence is directly used as the score of the sequence. However, since the search space of the K2 algorithm increases with the number of nodes, which results in that the algorithm takes too long, it is not desirable to rely on the K2 search as the fitness function in the whole swarm search process.
In order to keep the good evaluation effect of K2 search on variable sequences and solve the problems of time complexity increase, efficiency reduction and the like caused by the fact that K2 search is completely used in an algorithm, the invention introduces a concept of correlation. First, a definition of the degree of correlation used in the present invention is given.
Assuming X, Y is any two column vectors of equal length, the correlation between X and Y can be expressed as:
Figure BDA0002101122470000111
wherein C (X, Y) represents the covariance of X and Y. Generally speaking, the larger the correlation r, the stronger the correlation between two vectors. In order to reduce the number of times of the algorithm calling K2 search, variable sequences can be screened first, and a part of 'poor' variable sequences are eliminated. The screening method is that the correlation degree of the current variable sequence and the sample data corresponding to the current optimal variable sequence is solved, and if the correlation degree is smaller than a given threshold value delta, the variable sequence is abandoned, so that the sequence is prevented from being scored by repeatedly calling K2 for searching.
The realization process of the fitness function in the K2ABC algorithm is as follows:
step 1: after the food sources are initialized, scoring evaluation is carried out on the food sources by adopting K2 search, and the current optimal food sources are recorded;
step 2: in the stage of hiring bees and following bees, solving the correlation degree r of sample data corresponding to the searched new food source and the current optimal food source;
step 3: if the correlation r is larger than the threshold value delta, adopting K2 to search and calculate the fitness function value of the new food source, and performing subsequent operation; otherwise, the new food source is discarded, the original food source is retained, and step2 is transferred.
step 4: in the scout bee phase, a K2 search is used to score new food sources.
S108, judging the relation between the continuous non-updating times of the sequence and the maximum limit time limit, and if the continuous non-updating times are larger than the maximum limit time limit, eliminating the corresponding food source and searching for a new food source by the scout bees for replacement;
and S109, recording the current optimal food source.
S110, if the current iteration times of the algorithm exceed the set maximum iteration times, outputting the current optimal food source; otherwise, go to step S105.
4. Experimental conditions
(1) Standard test network
The invention selects 3 common standard test networks for evaluating the feasibility and effectiveness of the algorithm, namely an Asia network, a Car network and an Alarm network. The Asia network has 8 nodes and 8 edges; the Car network has 18 nodes and 17 edges; the Alarm network has 37 variables and 45 sides, and the three networks are respectively representatives of a small network, a medium network and a large network.
(2) Test data
Four groups of data of 1000, 2000, 5000 and 10000 are respectively generated based on the three networks, and all experimental results are average values of the algorithms after 100 times of operation in order to avoid the contingency.
(3) Comparison of results with standards
The experiment compares the number of Correct Edges (CE), Missing Edges (ME), extra edges (AE) and Reversed Edges (RE) of the network learned by each algorithm compared with the standard network, and calculates the hamming distance (SHD) of each network from the following formula.
SHD＝ME+AE+RE
Table 2 shows specific experimental data of the K2ABC algorithm and the Random-K2 algorithm for randomly generating node sequences. According to the results in the table, the correct sides of the network structure learned by the K2ABC algorithm are all more than the correct sides of the network structure learned by the Random-K2 algorithm, and the network structure learned by the K2ABC algorithm has less SHD, which indicates that the network structure learned by the K2ABC algorithm is closer to a standard network, so that the K2ABC algorithm can better fit sample data and learn a better network structure.
TABLE 2K 2ABC vs. Random-K2 Algorithm results
Figure BDA0002101122470000131
While the present invention has been described with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, which are illustrative and not restrictive, and it will be apparent to those skilled in the art that various changes and modifications can be made therein without departing from the spirit and scope of the invention as defined in the appended claims.

Claims (6)

1. A momentum wheel fault diagnosis method based on a K2ABC algorithm is characterized by comprising the following steps:
s1, establishing a Bayesian network model of the momentum wheel fault based on the K2ABC algorithm and by combining the momentum wheel fault data;
s2, based on the Bayesian network model established in the step S1, utilizing Bayesian network parameter learning to obtain parameters of each network node in the model;
s3, based on the model with the node parameters obtained in the step S2 and the obtained evidence information, carrying out probability calculation on various reasons which may cause the momentum wheel to break down by adopting a Bayesian network inference algorithm; the evidence information refers to adding state information to a certain node or a plurality of nodes during application to reason about the influence on the states of other nodes;
s4, finding out the node with the maximum posterior probability as the final diagnosis result by comparing the probabilities calculated in the inference stage;
the K2ABC algorithm in step S1 includes the following steps:
s101, acquiring momentum wheel fault data which are discrete data;
s102, extracting variables in fault data to serve as nodes in a network model to be constructed;
s103, setting algorithm execution parameters;
s104, randomly initializing the bee colony, and calculating the fitness value of the initial bee colony;
s105, the hiring bee searches for new food sources, calculates the fitness value of each food source by using a fitness function based on K2 search and correlation, and selects the food sources needing to be reserved according to a greedy mechanism;
s106, calculating the selection probability of each food source according to the following formula:
Figure FDA0003456239940000011
wherein p isiProbability of being selected for the ith food source, fitiThe fitness function value corresponding to the ith food source is max (fit), and the max (fit) is the maximum value of the fitness function values corresponding to all the food sources;
s107, selecting a food source to be followed by the following bee by roulette according to the probability calculated in the step S106, further searching for a new food source, then calculating each fitness value by utilizing a fitness function based on K2 searching and relevance, and reserving the food source with higher fitness value;
s108, judging the size relation between the continuous non-updating times and the maximum limiting times of the sequence, and if the continuous non-updating times are larger than the maximum limiting times, eliminating the corresponding food sources and searching new food sources for substitution by the scout bees;
s109, recording the current optimal food source;
s110, if the termination condition of the algorithm is met, outputting the current optimal food source; otherwise, turning to the step S105;
in the foregoing step S105 and step S107, the searching for the fitness function based on the K2 and the correlation degree to calculate each fitness value specifically includes:
step 1: after the food sources are initialized, scoring evaluation is carried out on the food sources by adopting K2 search, and the current optimal food sources are recorded;
step 2: in the stage of hiring bees and following bees, solving the correlation degree r of sample data corresponding to the searched new food source and the current optimal food source;
step 3: if the correlation r is larger than the threshold value delta, adopting K2 to search and calculate the fitness function value of the new food source, and performing subsequent operation; otherwise, abandoning the new food source, keeping the original food source, and switching to step 2;
step 4: in the scout bee phase, a K2 search is used to score new food sources.
2. The momentum wheel fault diagnosis method based on the K2ABC algorithm as claimed in claim 1, wherein in step S11, if the momentum wheel fault data is non-discrete data, the momentum wheel fault data is discretized first to form discrete pair momentum wheel fault data.
3. The momentum wheel fault diagnosis method based on K2ABC algorithm as claimed in claim 1, wherein the new food source is searched by the following method: obtaining a new food source by adopting a domain search strategy, and then sequentially carrying out reverse operator operation and mutual information-based immune operator operation on the new food source to obtain a final new food source individual;
the domain search strategy is a domain search strategy adopting three-point exchange: generating a new solution through local change on the basis of an original food source, wherein the specific method comprises the steps of randomly generating three exchange points 1,2 and 3 on an original variable sequence, and sequentially exchanging numbers of the three points, namely, original data of a 2 nd exchange point is used as new data of a 1 st exchange point, original data of a 3 rd exchange point is used as new data of the 2 nd exchange point, and original data of the 1 st exchange point is used as new data of the 3 rd exchange point;
the inversion operator is that two different sites are randomly generated in the variable sequencing, and the variable sequence between the two sites is subjected to the reverse sequence operation;
the specific process of updating the food source based on the immune operator is as follows:
step 1: in the variable sequence S ═ x1,x2,L,xn]Optionally a node xiObtaining a new sequence S to be adjusted1＝[xi,xi+1,L,xn]And a reserved sequence S2＝[x1,x2,L,xi-1]；
step 2: from the new sequence S1Starting with the first node, searching the rest nodes for the node with the most mutual information valueA large node, and inserting the node after the current node;
step 3: repeating the above steps until a new sequence S1Finishing the adjustment of all the nodes in the node B;
step 4: and combining the adjusted new sequence with the reserved sequence to obtain an adjusted variable sequence.
4. The momentum wheel fault diagnosis method based on K2ABC algorithm as claimed in claim 1, wherein in step S107, the food source with higher preservation fitness value is specifically: the correlation degree of the sample data corresponding to the current variable sequence and the current optimal variable sequence is solved, and if the correlation degree is smaller than a given threshold value delta, the variable sequence is abandoned, so that the sequence is prevented from being scored by repeatedly calling K2 for searching.
5. The momentum wheel fault diagnosis method based on K2ABC algorithm of claim 1, wherein to reduce the complexity of food source coding, a serial number coding method is adopted to encode the node serial number xiAs a unique identification of the variable.
6. A momentum wheel fault diagnosis device based on K2ABC algorithm, comprising a computer storage medium storing computer executable instructions for implementing the momentum wheel fault diagnosis method based on K2ABC algorithm as claimed in any one of claims 1-5.
CN201910535762.4A 2019-06-20 2019-06-20 Momentum wheel fault diagnosis method and device based on K2ABC algorithm Active CN110443344B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910535762.4A CN110443344B (en) 2019-06-20 2019-06-20 Momentum wheel fault diagnosis method and device based on K2ABC algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910535762.4A CN110443344B (en) 2019-06-20 2019-06-20 Momentum wheel fault diagnosis method and device based on K2ABC algorithm

Publications (2)

Publication Number Publication Date
CN110443344A CN110443344A (en) 2019-11-12
CN110443344B true CN110443344B (en) 2022-05-06

Family

ID=68429275

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910535762.4A Active CN110443344B (en) 2019-06-20 2019-06-20 Momentum wheel fault diagnosis method and device based on K2ABC algorithm

Country Status (1)

Country Link
CN (1) CN110443344B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6076083A (en) * 1995-08-20 2000-06-13 Baker; Michelle Diagnostic system utilizing a Bayesian network model having link weights updated experimentally
CN102393644A (en) * 2011-11-01 2012-03-28 北京航空航天大学 Ducted unmanned aerial vehicle anti-sway method based on optimized quadratic form control of artificial bee colony
CN104217251A (en) * 2014-08-12 2014-12-17 西北工业大学 Equipment failure Bayesian network prediction method based on K2 algorithm
CN106154182A (en) * 2016-08-26 2016-11-23 上海电力学院 A kind of based on the lithium battery method for diagnosing faults improving D S evidence theory
CN109508745A (en) * 2018-11-14 2019-03-22 上海交通大学 The detection method of gas turbine gascircuit fault based on Bayesian network model

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6076083A (en) * 1995-08-20 2000-06-13 Baker; Michelle Diagnostic system utilizing a Bayesian network model having link weights updated experimentally
CN102393644A (en) * 2011-11-01 2012-03-28 北京航空航天大学 Ducted unmanned aerial vehicle anti-sway method based on optimized quadratic form control of artificial bee colony
CN104217251A (en) * 2014-08-12 2014-12-17 西北工业大学 Equipment failure Bayesian network prediction method based on K2 algorithm
CN106154182A (en) * 2016-08-26 2016-11-23 上海电力学院 A kind of based on the lithium battery method for diagnosing faults improving D S evidence theory
CN109508745A (en) * 2018-11-14 2019-03-22 上海交通大学 The detection method of gas turbine gascircuit fault based on Bayesian network model

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
An artificial bee colony algorithm for learning Bayesian networks;Junzhong Ji et al.;《Soft Computing》;20130601;第986页第4节 *
基于贝叶斯网络的动量轮可靠性建模与评估;厉海涛 等;《***工程与电子技术》;20090228;第31卷(第2期);第485-487页第2-3节 *

Also Published As

Publication number Publication date
CN110443344A (en) 2019-11-12

Similar Documents

Publication Publication Date Title
Belouadah et al. Scail: Classifier weights scaling for class incremental learning
CN109783817B (en) Text semantic similarity calculation model based on deep reinforcement learning
CN108875916B (en) Advertisement click rate prediction method based on GRU neural network
CN109961098B (en) Training data selection method for machine learning
CN113094593B (en) Social network event recommendation method, system, device and storage medium
CN115099606B (en) Training method and terminal of power grid dispatching model
CN109558898B (en) Multi-choice learning method with high confidence based on deep neural network
CN111950393A (en) Time sequence action fragment segmentation method based on boundary search agent
CN115035345A (en) Cross-domain hyperspectral image classification technology based on improved ant colony optimization algorithm
CN110443344B (en) Momentum wheel fault diagnosis method and device based on K2ABC algorithm
Fortier et al. Learning Bayesian classifiers using overlapping swarm intelligence
Cano et al. A score based ranking of the edges for the PC algorithm
CN116561302A (en) Fault diagnosis method, device and storage medium based on mixed knowledge graph reasoning
CN113377884A (en) Event corpus purification method based on multi-agent reinforcement learning
Hu et al. Pwsnas: powering weight sharing nas with general search space shrinking framework
CN112906435B (en) Video frame optimization method and device
CN107480768A (en) Bayesian network structure adaptive learning method and device, storage device and terminal device
CN114154582A (en) Deep reinforcement learning method based on environment dynamic decomposition model
CN113591731A (en) Knowledge distillation-based weak surveillance video time sequence behavior positioning method
CN113095328A (en) Self-training-based semantic segmentation method guided by Gini index
CN110334395A (en) The satellite momentum wheel fault diagnosis method and system of initialization EM algorithm based on JADE
CN112201340A (en) Electrocardiogram disease determination method based on Bayesian network filtering
CN111783980B (en) Ranking learning method based on dual cooperation generation type countermeasure network
CN114386764B (en) GRU and R-GCN based OJ platform topic sequence recommendation method
CN116091884A (en) Construction method and system of dual-discriminant generation type countermeasure network model

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant