CN106803101B - Odometer method for diagnosing faults based on Hidden Markov Model - Google Patents
Odometer method for diagnosing faults based on Hidden Markov Model Download PDFInfo
- Publication number
- CN106803101B CN106803101B CN201611256419.9A CN201611256419A CN106803101B CN 106803101 B CN106803101 B CN 106803101B CN 201611256419 A CN201611256419 A CN 201611256419A CN 106803101 B CN106803101 B CN 106803101B
- Authority
- CN
- China
- Prior art keywords
- hidden markov
- quantity data
- markov model
- observed quantity
- odometer
- 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
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C25/00—Manufacturing, calibrating, cleaning, or repairing instruments or devices referred to in the other groups of this subclass
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/29—Graphical models, e.g. Bayesian networks
- G06F18/295—Markov models or related models, e.g. semi-Markov models; Markov random fields; Networks embedding Markov models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Physiology (AREA)
- Genetics & Genomics (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Probability & Statistics with Applications (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Manufacturing & Machinery (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Complex Calculations (AREA)
Abstract
The present invention provides a kind of odometer method for diagnosing faults based on Hidden Markov Model.This method comprises: the characteristic in extraction odometer operational process is as observed quantity data, observed quantity data are pre-processed, pretreated observed quantity data are input in Hidden Markov Model and are trained, obtain wheel it is normal, skid, the Hidden Markov Model of three kinds of states of locking, establish the malfunction classifier of odometer；Observed quantity data to be identified are input to the malfunction classifier of odometer, are matched respectively with the Hidden Markov Model of the various states of wheel, the state of the corresponding wheel of observed quantity data to be identified is determined according to matching result.The present invention improves a lot for the diagnostic accuracy of malfunction, is improved by genetic algorithm to the parameter training part in Hidden Markov Model, after simulation comparison, the results showed that genetic algorithm can comparatively fast reach stable state on training speed.
Description
Technical field
The present invention relates to fault diagnosis technology field more particularly to a kind of odometer failures based on Hidden Markov Model
Diagnostic method.
Background technique
HMM (Hidden Markov Model, hidden Markov model) is a doubly stochastic process, random comprising two
Process, one is the process of another state is transferred under hidden state from a state, the second is under each hidden state
Generate the process of observation.It include three kinds of rudimentary algorithms in HMM, every kind of algorithm is corresponding to solve a basic problem: (1) HMM mould
Shape parameter seeks the probability (assessment) of the observation sequence and Model Matching it has been determined that provide observation sequence；(2) successful in training
In model, the most possible hidden state path (decoding) that element generates given observation sequence is searched；(3) HMM model unknown parameters,
Training sequence is provided, generates HMM (study) by finding local optimum.
In the fault diagnosis of odometer (Odometry), there are mainly two types of algorithms, one is Forward-backward algorithm, is used
To solve the problems, such as HMM probability calculation；The second is Baum-Welch algorithm, is the solution of HMM parameter Estimation and training problem.
In Hidden Markov Model, parameter training has a significant impact for the pattern classification effect of last model.It is former hidden
The likelihood function value in B-W algorithm in Markov model restrains final determining optimal solution by monotone increasing, but B-W is calculated
It is that convergence rate is slower that method is put really, and numerical value operation is complicated, and the very easy result that optimal solution is taken as to locally optimal solution.
Summary of the invention
The embodiment provides a kind of odometer method for diagnosing faults based on Hidden Markov Model, to realize
The fault diagnosis of odometer is effectively performed.
To achieve the goals above, this invention takes following technical solutions.
A kind of odometer method for diagnosing faults based on Hidden Markov Model, comprising:
The characteristic in odometer operational process is extracted as observed quantity data, the observed quantity data are located in advance
Reason, the pretreatment include amplitude normalization, scalar quantization processing；
Pretreated observed quantity data are input in Hidden Markov Model and are trained, obtain wheel it is normal,
It skids, the Hidden Markov Model of three kinds of states of locking, establishes the malfunction classifier of odometer；
Observed quantity data to be identified are input to the malfunction classifier of the odometer, it is various with wheel respectively
The Hidden Markov Model of state is matched, and determines the corresponding wheel of the observed quantity data to be identified according to matching result
State.
Further, the characteristic extracted in odometer operational process is as observed quantity data, comprising:
Extract the velocity measurement v of the stored count pulse output in odometerodo；
Wherein, nkFor the pulse number at k moment, nk-1For the pulse number of k last moment, N is that odometer completes a circle
The umber of pulse issued afterwards, Δ t are time interval of the k moment to the k-1 moment, and d is the wheel footpath of wheel；
Observed quantity data vobservation(t) calculation formula is as follows:
vobservation(t)=vodo(t)-vreal(t)
Wherein vreal(t) be any moment t train actual motion speed；
Further, described to pre-process to the observed quantity data, which includes amplitude normalization, scalar
Quantification treatment, comprising:
Amplitude normalized is carried out to the observed quantity data using normalized function mapminmax, then passes through information source
Lloyds data compression algorithm in coding techniques carries out scalar quantization processing to the observed quantity data after amplitude normalized.
Further, described pretreated observed quantity data are input in Hidden Markov Model is trained,
Obtain wheel it is normal, skid, the Hidden Markov Model of three kinds of states of locking, establish the malfunction classifier of odometer,
Include:
Choose respectively wheel it is normal, skid, the observed quantity data of odometer under three kinds of states of locking, use the sight
Measurement data utilize genetic algorithm optimization after B-W algorithm training Hidden Markov Model parameter, obtain wheel it is normal, beat
Sliding, three kinds of states of locking Hidden Markov Model, establishes the malfunction classifier of odometer.
Further, the genetic algorithm includes the following steps:
(1) parameter set of practical problem is encoded；
(2) initial value and each genetic operator of parameter are determined；
(3) it enables k=0 indicate evolutionary generation, and generates initialization population X (0)；
(4) fitness function is chosen, the value of fitness function is calculated in each generation, using fitness function value as standard
To judge whether individual enters the next generation；
(5) compared with termination condition, if do not met, k=k+1 algebra is enabled to increase a generation；
(6) selection operation is carried out, the basic principle of genetic algorithm " survival of the fittest " is based on, is chosen from current group X (k-1)
The low individual of fitness is given up in choosing, selects remaining individual X (k) into iterative process next time；
(7) crossover operation is carried out, the basic principle of genetic algorithm " information exchange " is based on, utilizes crossover probability Pc, according to
Crossover operator carries out crossover operation to the individual in X (k), then individual information is individual from former generation in the next generation；
(8) mutation operation variation is carried out, some structural reform in X (k) transitional population is randomly choosed according to mutation probability Pm
Become its value；
Observed quantity data described in the use utilize the B-W algorithm training Hidden Markov Model after genetic algorithm optimization
Parameter includes the following steps:
(1) parameter to be estimated: state Initial component π=(π1,π2,...πN), probability transfer matrix A=(aij), observe to
Metric density function bj(y)；
(2) fitness function: adapting to value function and be taken as objective function, and taking lnp (Y=y | X)-Viol (X) is objective function；
(3) generate initial population: parameter π, A etc. are randomly generated in [0,1] and carry out binary processing, according to receipts
It holds back precision and determines number of bits；
(4) value for setting each genetic operator, if select probability P=0.1, crossover probability Pc=0.25, mutation probability Pm=
0.01；
(5) genetic algorithm termination condition is set, maximum number of iterations is taken as max_iter=30, convergence precision ε=
0.0001。
Further, the malfunction classifier that observed quantity data to be identified are inputted to the odometer, will
Observed quantity data to be identified are matched with the Hidden Markov Model of the various states of wheel respectively, true according to matching result
Determine the state of the corresponding wheel of the observed quantity data to be identified, comprising:
The malfunction classifier that observed quantity data to be identified are inputted to the odometer, utilizes Forward-backward algorithm
By observed quantity data to be identified respectively with wheel it is normal, skid, the Hidden Markov Model of three kinds of states of locking carries out
Match, obtains the corresponding matching value of Hidden Markov Model of three kinds of states, three matching values are compared, it will be maximum
The corresponding normal condition of matching value, slipping state or locking state are as the corresponding wheel of the observed quantity data to be identified
State.
As can be seen from the technical scheme provided by the above-mentioned embodiment of the present invention, the present invention introduces Hidden Markov method
In the ODO fault diagnosis of column control positioning subsystem, the spy that can characterize ODO failure is extracted by a large amount of data statistic analysis
Sign amount, and process data into the input data of HMM using the method for amplitude normalization and scalar quantization passes through 50 groups of data
Statistical analysis, the results showed that validity of the HMM for ODO fault diagnosis.And by genetic algorithm to the parameter training in HMM
Part improve and with B-W algorithm comparison, by simulation comparison, genetic algorithm can reach surely quickly on training speed
State, and training precision improves 86%, there is corresponding improvement, it was demonstrated that the validity that parameter training is transformed in genetic algorithm.
The additional aspect of the present invention and advantage will be set forth in part in the description, these will become from the following description
Obviously, or practice through the invention is recognized.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, required use in being described below to embodiment
Attached drawing be briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for this
For the those of ordinary skill of field, without any creative labor, it can also be obtained according to these attached drawings other
Attached drawing.
Fig. 1 is a kind of basic composition schematic diagram of HMM provided in an embodiment of the present invention；
Fig. 2 is HMM Troubleshooting Flowchart provided in an embodiment of the present invention；
Fig. 3 is the calculation flow chart of genetic algorithm provided in an embodiment of the present invention；
Fig. 4 is HMM training pattern curve provided in an embodiment of the present invention；
Fig. 5 is the improved training curve of genetic algorithm under normal condition provided in an embodiment of the present invention；
Fig. 6 is HMM and HMM training curve contrast schematic diagram after improvement provided in an embodiment of the present invention.
Specific embodiment
Embodiments of the present invention are described below in detail, the example of the embodiment is shown in the accompanying drawings, wherein from beginning
Same or similar element or element with the same or similar functions are indicated to same or similar label eventually.Below by ginseng
The embodiment for examining attached drawing description is exemplary, and for explaining only the invention, and is not construed as limiting the claims.
Those skilled in the art of the present technique are appreciated that unless expressly stated, singular " one " used herein, " one
It is a ", " described " and "the" may also comprise plural form.It is to be further understood that being arranged used in specification of the invention
Diction " comprising " refer to that there are the feature, integer, step, operation, element and/or component, but it is not excluded that in the presence of or addition
Other one or more features, integer, step, operation, element, component and/or their group.It should be understood that when we claim member
Part is " connected " or when " coupled " to another element, it can be directly connected or coupled to other elements, or there may also be
Intermediary element.In addition, " connection " used herein or " coupling " may include being wirelessly connected or coupling.Wording used herein
"and/or" includes one or more associated any cells for listing item and all combinations.
Those skilled in the art of the present technique are appreciated that unless otherwise defined, all terms used herein (including technology art
Language and scientific term) there is meaning identical with the general understanding of those of ordinary skill in fields of the present invention.Should also
Understand, those terms such as defined in the general dictionary, which should be understood that, to be had and the meaning in the context of the prior art
The consistent meaning of justice, and unless defined as here, it will not be explained in an idealized or overly formal meaning.
In order to facilitate understanding of embodiments of the present invention, it is done by taking several specific embodiments as an example below in conjunction with attached drawing further
Explanation, and each embodiment does not constitute the restriction to the embodiment of the present invention.
The embodiment of the present invention introduce intelligent trouble diagnosis field based on Hidden Markov Model Analysis on Fault Diagnosis side
Method, to wheel footpath may occur in the process of running slide, the fault types such as locking carry out diagnosis detection, finally by practical existing
Field data verifies the accuracy of the diagnostic method.It is easily fallen into finally, attempting improvement B-W (Baum-Welch) algorithm using genetic algorithm
Enter the defect of local optimum, and achieves good effect.
Genetic algorithm (Genetic Algorithms, GA) has advantage for processing complicated optimum problem, can be to avoid
The problem of complicated calculations, can be obtained by optimal solution using three kinds of selection in genetic algorithm, intersection, variation operators.Pass through mould
The genetic mechanism and biological evolution process of quasi- nature form Global Optimization Algorithm For Analysis, by encoding to problem, construct
Corresponding individual, obtains potential disaggregation initial population in problem to be solved.According to the survival of the fittest, the principles such as the survival of the fittest are pressed
The individual individual good to the fitness size selection adaptation of institute's Solve problems enters the next generation, then is lost by intersection, variation etc.
It passes to operate and obtains performance more preferably population, on last evolution convergence to the best individual of adaptability generation upon generation of.Genetic algorithm tool
Having the advantage that (1), it is unrelated with its field for the solution of optimization problem, and can be from global search optimal solution；(2) hidden
Containing concurrency, not only initial solution is but also can be compared from group's with Parallel Multivalue；(3) genetic algorithm can be with it
He combines algorithm, and use is more flexible.
Genetic algorithm has the advantages that multivalue simultaneously scans for that the possibility for falling into locally optimal solution therefore can be reduced due to it
Property, and search speed is very fast, therefore can use genetic algorithm to optimize the B-W algorithm of Hidden Markov Model.
Embodiment one
It breaks down in train positioning unit and causes positioning-speed-measuring unpunctual, need quickly and accurately to find out failure hair
Failure is simultaneously repaired in raw position, reduces the time of maintenance.There are many train locating methods, and what is more generally used mainly has: being based on
The train locating method of ODO (odometer), the train positioning based on track circuit, the train locating method based on satellite navigation,
The train locating method of inquiry response devices, in addition there are application inertial sensors, Doppler radar, accelerometer etc. to be based on pass more
The train combined positioning method of sensor information fusion.Wherein, GPS positioning is absolute fix, will not generation system accumulated error,
GPS receiver device technology maturation, reliable performance；ODO positioning (i.e. wheel shaft speed sensor positioning) is relative positioning, can be generated
System accumulated error, but its advantage be it is stable and reliable for performance, at low cost, it is easy to maintain.Therefore the localization method based on GPS/ODO
It is the probing direction towards actual location demand.
ODO is connected with wheel, and in train operation, wheel rotates, and measurement accuracy is influenced by wheel rail relation.When train driving,
Wheel can generate phenomena such as skidding, locking.In addition with the increase of running time, wheel will appear abrasion, in also making in this way
Journey meter generates error.Therefore odometer realizes that the key technology of train positioning is the source analysis of odometer position error,
Currently, field service personnel is there is no analysis on Source of Error is carried out for odometer, when ODO breaks down, maintenance personal is general
Possible failure cause is judged by rule of thumb, corrigendum modification is carried out to the wheel footpath value in software, maintenance efficiency is lower, seriously affects
The efficiency of operation, so carrying out failure there is an urgent need to can rapidly and accurately determine failure cause when ODO breaks down and examining
It is disconnected.
The embodiment of the invention provides a kind of ODO method for diagnosing faults based on Hidden Markov Model, by hidden Ma Erke
Husband's method is introduced into the ODO fault diagnosis of column control positioning subsystem, and such as the basic composition schematic diagram that Fig. 1 is HMM, this method includes
Several processing such as selection, data prediction, the fault diagnosis based on HMM, the HMM parameter training based on genetic algorithm of observed quantity
Process introduces each treatment process separately below:
The selection of the observed quantity, the output of odometer (ODO) are pulse signal, calculate the principle of wheel velocity
Are as follows: before pulse signal is input to counter interface, shaping filter is carried out to it first, it is general to pass through plastic filter circuit reality
Existing, the result by counting interface calculates the number of pulse signal, finally calculates wheel velocity, theoretically, the accumulative meter of odometer
Rapid pulse rushes calculated measured value are as follows:
Wherein, nkFor the pulse number at k moment, nk-1For the pulse number of k last moment.N is that odometer completes a circle
The umber of pulse issued afterwards, the value are a determining value, and Δ t is the k moment to the time interval at k-1 moment and a fixation
Value, d are the wheel footpath of wheel.
Observed quantity data vobservation(t) calculation formula is as follows:
vobservation(t)=vodo(t)-vreal(t)
Wherein vrealIt (t) is any moment t train actual motion speed, the v when GPS positioning is in good conditionreal(t) it is equal to
The value v that tests the speed of GPS positioningGPS(t)
The data prediction refers to through the Lloyds data compression algorithm in source coding technique to training sequence
Scalar quantization processing is carried out, before scalar quantization processing, is needed at data in a certain range, so needing to carry out normalizing
Change processing, the present invention are normalized training data using normalized function mapminmax.In this way, by normalization
And scalar quantization treated training sequence can be used to train HMM model, establish the malfunction classifier of ODO.
Mapminmax function is the normalized function in Matlab, and Lloyds algorithmic function is that one of Matlab data compression is calculated
Method function.
The fault diagnosis based on HMM, Fig. 2 are a kind of HMM Troubleshooting Flowchart provided in an embodiment of the present invention,
By data prediction, the v under three kinds of states is chosen respectivelyobservation(t) observation sequence utilizes the B- after genetic algorithm optimization
W algorithm trains HMM parameter, obtains the HMM model of three kinds of normal, skidding, locking states, establishes the fault grader of ODO.
The malfunction classifier that observed quantity data to be identified are inputted to the odometer, utilizes Forward-backward algorithm
By observed quantity data to be identified respectively with wheel it is normal, skid, the Hidden Markov Model of three kinds of states of locking carries out
Match, obtains the corresponding matching value of Hidden Markov Model of three kinds of states, three matching values are compared, it will be maximum
The corresponding normal condition of matching value, slipping state or locking state are corresponding current as the observed quantity data to be identified
Wheel condition, to realize fault diagnosis.
Described includes: by genetic algorithm to HMM's using the B-W algorithm training HMM parameter after genetic algorithm optimization
Parameter training part is transformed, and is compared with B-W algorithm, it was demonstrated that the validity of transformation.Fig. 3 mentions for the embodiment of the present invention
A kind of calculation flow chart of the genetic algorithm supplied, including following treatment process:
HMM parameter training based on genetic algorithm finds λ according to maximum likelihood criterion*, so that
Wherein y is observation sequence, and λ is model parameter, and process is with Prescribed Properties:
Thus maximize p (Y=y | λ) is exactly one with constrained optimization problem in fact, and p (Y=y | λ) it can be by preceding
Backward algorithm calculates.
The amount that the i.e. individual X of penalty violates degree of restraint is defined first
M1,M2Take very big positive number.
Parameter training process based on genetic algorithm is as follows:
(1) parameter set of practical problem is encoded；
(2) initial value and each genetic operator of parameter are determined；
(3) it enables k=0 indicate evolutionary generation, and generates initialization population X (0)；
(4) suitable fitness function is chosen, the value of fitness function is calculated in each generation, is made with fitness function value
For standard come judge individual whether enter the next generation；
(5) compared with termination condition, if do not met, k=k+1 algebra is enabled to increase a generation；
(6) selection operation is carried out, the basic principle of genetic algorithm " survival of the fittest " is based on, is chosen from current group X (k-1)
Choosing, gives up the low individual of fitness, selects excellent individual X (k) into iterative process next time；
(7) crossover operation is carried out, the basic principle of genetic algorithm " information exchange " is based on, utilizes crossover probability Pc, according to
Crossover operator carries out crossover operation to the individual in X (k), then individual information is individual from former generation in the next generation；
(8) mutation operation variation is carried out, according to mutation probability PmRandomly choose some structural reform in X (k) transitional population
Become its value.
Observed quantity data described in the use utilize the B-W algorithm training Hidden Markov Model after genetic algorithm optimization
Parameter includes the following steps:
(1) parameter to be estimated: state Initial component π=(π1,π2,...πN), probability transfer matrix A=(aij), observe to
Metric density function bj(y)；
(2) fitness function: adapting to value function and be taken as objective function, and taking lnp (Y=y | X)-Viol (X) is objective function；
(3) generate initial population: parameter π, A etc. are randomly generated in [0,1] and carry out binary processing, according to receipts
It holds back precision and determines number of bits；
(4) value for setting each genetic operator, if select probability P=0.1, crossover probability Pc=0.25, mutation probability Pm=
0.01；
Genetic algorithm termination condition is set, maximum number of iterations is taken as max_iter=30, convergence precision ε=0.0001.
Embodiment two
The present invention using on 2 22nd, 2014 Jin Mei group iron transport line road acquisition data dally, slide detection examination
It tests, testing machine license number is DF85512 fifth wheel to two-way velocity sensor is above equipped with, and is taken turns defeated to the velocity sensor that rotates a circle
200 pulses out.50 groups of data in the case of normal, skidding, idle running are chosen respectively, and data measure under identical operating condition, have
Identical sample frequency chooses 30 groups therein as training data respectively, and 20 groups as observation data.First, by 30 groups of instructions
Practice data and parameter training is carried out to the corresponding HMM model of three kinds of states of ODO, get parms, Fig. 4 is that three kinds of states are corresponding
HMM training curve.
Three groups of data converge in fixed range when training is to step 10, have reached local optimum, then model parameter
It has been obtained that, parameter training is partially completed.
Then by 20 groups of test datas after normalization and scalar quantization, trained successful three instructions are sent to
Practice in model, the matching degree of test data and model is calculated using Forward-backward algorithm, choosing maximum value is that cycle tests is worked as
Preceding state tests HMM classifying quality.
Thus it proves, HMM model can model different faults data, and can be effective for observation data input
Fault diagnosis is realized in matching.From in table it will be seen that 3 observation data diagnosis occur is in locking fault diagnosis
Sipping fault, analyzing this reason may be that data variation rule between skidding data and locking data is close, therefore will appear
Such case, but the judgement for ODO failure and normal condition, the precision of HMM diagnostic result can achieve 100%.
Embodiment three
It, can using genetic algorithm since B-W algorithm has the disadvantages of convergence is slow, and calculating is complicated, easily falls into locally optimal solution
Effectively to solve the above problems, therefore the present invention is transformed by parameter training part of the genetic algorithm to HMM, and and B-W
Algorithm is compared, it was demonstrated that the validity of transformation.
The present invention chooses the data under normal condition and carries out genetic algorithm and the comparison of B-W algorithm parameter Simulation Training, it is assumed that
Population scale N=30.Specific step is as follows:
(1) parameter to be estimated: state Initial component π=(π1,π2,...πN), probability transfer matrix A=(aij), observe to
Metric density function bj(y)；The meaning of each parameter is exactly the modifier before it in fact, π vector: contains (hidden) model in the time
State-transition matrix A: the probability (probability) of one special hidden state when t=1 contains a hidden state to separately
The probability confusion matrix b of one hidden state: containing some special hidden state of given Hidden Markov Model, sees
The probability of some observation state observed.
(2) it fitness function: in most cases, adapts to value function and is taken as objective function, take lnp (Y=y | X)-
Viol (X) is objective function；Viol (X) expression penalty, and lnp (Y=y | X) indicate that the matching of list entries and population is closed
System, is an optimization problem；
(3) generate initial population: parameter π, A etc. are randomly generated in [0,1] and carry out binary processing, according to receipts
It holds back precision and determines number of bits；
(4) value for setting each genetic operator, if select probability P=0.1, crossover probability Pc=0.25, mutation probability Pm=
0.01；
(5) genetic algorithm termination condition is set, maximum number of iterations is taken as max_iter=30, convergence precision ε=
0.0001。
The data chosen under normal condition carry out genetic algorithm parameter Simulation Training, and it is as shown in Figure 5 to obtain result.Fig. 6 is
Improve the preceding comparison with training curve both after improvement, it can be seen that the training curve that parameter training obtains is carried out using genetic algorithm
The B-W algorithm compared in convergence rate in Hidden Markov can comparatively fast reach stable state and final convergence precision side
Face, in HMM after B-W algorithm parameter training 30 step of iteration, matching value loglik=-37.046634, after genetic algorithm is improved
Loglik=-5.0431 improves 86%.
In conclusion Hidden Markov method is introduced into the ODO fault diagnosis of column control positioning subsystem by the present invention, pass through
A large amount of data statistic analysis extracts the characteristic quantity that can characterize ODO failure, and utilizes amplitude normalization and scalar quantization
Method processes data into the input data of HMM, pass through the statistical analysis of 50 groups of data, the results showed that HMM is for ODO failure
The validity of diagnosis.And the parameter training part in HMM is improved by genetic algorithm and with B-W algorithm comparison, pass through
Simulation comparison, genetic algorithm can reach stable state quickly on training speed, and training precision improves 86%, have corresponding
Improve, it was demonstrated that the validity that genetic algorithm is transformed parameter training.
Example analysis results, using the odometer method for diagnosing faults based on Hidden Markov Model of the embodiment of the present invention
It improves a lot for normal and malfunction diagnostic accuracy, the parameter in Hidden Markov Model is instructed by genetic algorithm
Practice part to improve, shows that genetic algorithm can comparatively fast reach stable state on training speed after simulation comparison.
Those of ordinary skill in the art will appreciate that: attached drawing is the schematic diagram of one embodiment, module in attached drawing or
Process is not necessarily implemented necessary to the present invention.
As seen through the above description of the embodiments, those skilled in the art can be understood that the present invention can
It realizes by means of software and necessary general hardware platform.Based on this understanding, technical solution of the present invention essence
On in other words the part that contributes to existing technology can be embodied in the form of software products, the computer software product
It can store in storage medium, such as ROM/RAM, magnetic disk, CD, including some instructions are used so that a computer equipment
(can be personal computer, server or the network equipment etc.) executes the certain of each embodiment or embodiment of the invention
Method described in part.
All the embodiments in this specification are described in a progressive manner, same and similar portion between each embodiment
Dividing may refer to each other, and each embodiment focuses on the differences from other embodiments.Especially for device or
For system embodiment, since it is substantially similar to the method embodiment, so describing fairly simple, related place is referring to method
The part of embodiment illustrates.Apparatus and system embodiment described above is only schematical, wherein the conduct
The unit of separate part description may or may not be physically separated, component shown as a unit can be or
Person may not be physical unit, it can and it is in one place, or may be distributed over multiple network units.It can root
According to actual need that some or all of the modules therein is selected to achieve the purpose of the solution of this embodiment.Ordinary skill
Personnel can understand and implement without creative efforts.
The foregoing is only a preferred embodiment of the present invention, but scope of protection of the present invention is not limited thereto,
In the technical scope disclosed by the present invention, any changes or substitutions that can be easily thought of by anyone skilled in the art,
It should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with scope of protection of the claims
Subject to.
Claims (5)
1. a kind of odometer method for diagnosing faults based on Hidden Markov Model characterized by comprising
The characteristic in odometer operational process is extracted as observed quantity data, the observed quantity data are pre-processed,
The pretreatment includes amplitude normalization, scalar quantization processing；
Pretreated observed quantity data are input in Hidden Markov Model and are trained, obtain wheel it is normal, skid,
The Hidden Markov Model of three kinds of states of locking establishes the malfunction classifier of odometer；
Observed quantity data to be identified are input to the malfunction classifier of the odometer, respectively with the various states of wheel
Hidden Markov Model matched, the shape of the corresponding wheel of the observed quantity data to be identified is determined according to matching result
State；
The characteristic extracted in odometer operational process is as observed quantity data, comprising:
Extract the velocity measurement v of the stored count pulse output in odometerodo；
Wherein, nkFor the pulse number at k moment, nk-1For the pulse number of k last moment, N is that odometer is completed to send out after a circle
Umber of pulse out, Δ t are time interval of the k moment to the k-1 moment, and d is the wheel footpath of wheel；
Observed quantity data vobservation(t) calculation formula is as follows:
vobservation(t)=vodo(t)-vreal(t)
Wherein vreal(t) be any moment t train actual motion speed.
2. the odometer method for diagnosing faults according to claim 1 based on Hidden Markov Model, which is characterized in that institute
That states pre-processes the observed quantity data, which includes amplitude normalization, scalar quantization processing, comprising:
Amplitude normalized is carried out to the observed quantity data using normalized function mapminmax, then passes through message sink coding
Lloyds data compression algorithm in technology carries out scalar quantization processing to the observed quantity data after amplitude normalized.
3. the odometer method for diagnosing faults according to claim 2 based on Hidden Markov Model, which is characterized in that institute
State pretreated observed quantity data are input in Hidden Markov Model is trained, obtain wheel it is normal, skid,
The Hidden Markov Model of three kinds of states of locking establishes the malfunction classifier of odometer, comprising:
Choose respectively wheel it is normal, skid, the observed quantity data of odometer under three kinds of states of locking, use the observed quantity
Data utilize the B-W algorithm training Hidden Markov Model parameter after genetic algorithm optimization, obtain the normal of wheel, skidding, embrace
The Hidden Markov Model of dead three kinds of states, establishes the malfunction classifier of odometer.
4. the odometer method for diagnosing faults according to claim 3 based on Hidden Markov Model, which is characterized in that institute
The genetic algorithm stated includes the following steps:
(1) parameter set of practical problem is encoded；
(2) initial value and each genetic operator of parameter are determined；
(3) it enables w=0 indicate evolutionary generation, and generates initialization population X (0)；
(4) fitness function is chosen, the value of fitness function is calculated in each generation, is commented using fitness function value as standard
Sentence whether individual enters the next generation；
(5) compared with termination condition, if do not met, w=w+1 algebra is enabled to increase a generation；
(6) selection operation is carried out, the basic principle of genetic algorithm " survival of the fittest " is based on, is selected from current group X (w-1),
Give up the low individual of fitness, selects remaining individual X (w) into iterative process next time；
(7) crossover operation is carried out, the basic principle of genetic algorithm " information exchange " is based on, utilizes crossover probability Pc, calculated according to intersecting
Son carries out crossover operation to the individual in X (w), then individual information is individual from former generation in the next generation；
(8) mutation operation variation is carried out, some individual in X (k) transitional population is randomly choosed according to mutation probability Pm and changes it
Value；
Observed quantity data described in the use utilize the B-W algorithm training Hidden Markov Model after genetic algorithm optimization to join
Number, includes the following steps:
(1) parameter to be estimated: state Initial component π=(π1,π2,...πN), probability transfer matrix A=(aij), observation vector is close
Spend function bj(y)；
(2) fitness function: adapting to value function and be taken as objective function, and taking lnp (Y=y | X)-Viol (X) is objective function；
(3) generate initial population: parameter π, A are randomly generated in [0,1] and carry out binary processing, according to convergence precision
Determine number of bits；
(4) value for setting each genetic operator, if select probability P=0.1, crossover probability Pc=0.25, mutation probability Pm=0.01；
(5) genetic algorithm termination condition is set, maximum number of iterations is taken as max_iter=30,
Convergence precision ε=0.0001.
5. the odometer method for diagnosing faults according to any one of claims 1 to 4 based on Hidden Markov Model, special
Sign is, the malfunction classifier that observed quantity data to be identified are input to the odometer, respectively with wheel
The Hidden Markov Model of various states matched, determine that the observed quantity data to be identified are corresponding according to matching result
Wheel state, comprising:
The malfunction classifier that observed quantity data to be identified are inputted to the odometer, will be to using Forward-backward algorithm
The observed quantity data of identification are matched with the Hidden Markov Model of three kinds of the normal of wheel, skidding, locking states respectively, are obtained
To the corresponding matching value of Hidden Markov Model of three kinds of states, three matching values are compared, by maximum matching
It is worth the shape of corresponding normal condition, slipping state or locking state as the corresponding wheel of the observed quantity data to be identified
State.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611256419.9A CN106803101B (en) | 2016-12-30 | 2016-12-30 | Odometer method for diagnosing faults based on Hidden Markov Model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611256419.9A CN106803101B (en) | 2016-12-30 | 2016-12-30 | Odometer method for diagnosing faults based on Hidden Markov Model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106803101A CN106803101A (en) | 2017-06-06 |
| CN106803101B true CN106803101B (en) | 2019-11-22 |
Family
ID=58984635
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611256419.9A Active CN106803101B (en) | 2016-12-30 | 2016-12-30 | Odometer method for diagnosing faults based on Hidden Markov Model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106803101B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109117353B (en) * | 2018-08-20 | 2019-12-20 | 中国石油大学(北京) | Fault diagnosis result fusion method and device |
| CN109282832B (en) * | 2018-09-30 | 2020-10-02 | 重庆自行者科技有限公司 | Self-adaptive calibration method and system of inertia-assisted odometer suitable for special road surface |
| CN109101010B (en) * | 2018-09-30 | 2021-06-11 | 深圳市元征科技股份有限公司 | Automobile fault diagnosis method and related equipment |
| CN111767933A (en) * | 2019-05-17 | 2020-10-13 | 北京京东尚科信息技术有限公司 | Method and device for identifying vehicle driving state |
| CN111687840B (en) * | 2020-06-11 | 2021-10-29 | 清华大学 | Method, device and storage medium for capturing space target |
| CN111649940A (en) * | 2020-07-10 | 2020-09-11 | 上海三一重机股份有限公司 | Walking speed reducer fault model generation method and device and computer equipment |
| CN114021621A (en) * | 2021-10-13 | 2022-02-08 | 北京和利时系统集成有限公司 | Fault diagnosis method, system, storage medium and edge computing device |
| CN114252261B (en) * | 2021-12-21 | 2024-03-08 | 沈阳顺义科技股份有限公司 | Fault diagnosis method and system for steering system of comprehensive transmission device |
| CN114300038B (en) * | 2021-12-27 | 2023-09-29 | 山东师范大学 | Multi-sequence comparison method and system based on improved biological geography optimization algorithm |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101520651A (en) * | 2009-03-03 | 2009-09-02 | 华中科技大学 | Analysis method for reliability of numerical control equipment based on hidden Markov chain |
| CN101739025A (en) * | 2009-12-03 | 2010-06-16 | 天津理工大学 | Immunity genetic algorithm and DSP failure diagnostic system based thereon |
| CN102054179A (en) * | 2010-12-14 | 2011-05-11 | 广州大学 | Online state monitoring and fault diagnosis device and method for rotary machine |
| CN103036974A (en) * | 2012-12-13 | 2013-04-10 | 广东省电信规划设计院有限公司 | Cloud computing resource scheduling method and system based on hidden markov model |
| CN103217157A (en) * | 2012-01-18 | 2013-07-24 | 北京自动化控制设备研究所 | Inertial navigation/mileometer autonomous integrated navigation method |
| CN103744977A (en) * | 2014-01-13 | 2014-04-23 | 浪潮(北京)电子信息产业有限公司 | Monitoring method and monitoring system for cloud computing system platform |
| CN104793606A (en) * | 2015-04-15 | 2015-07-22 | 浙江大学 | Industrial fault diagnosis method based on improved KPCA (kernel principal component analysis) and hidden Markov model |
| CN105137328A (en) * | 2015-07-24 | 2015-12-09 | 四川航天系统工程研究所 | Analog integrated circuit early-stage soft fault diagnosis method and system based on HMM |
| CN106226097A (en) * | 2016-09-14 | 2016-12-14 | 西安理工大学 | Bullet train airduct safe condition diagnostic method based on hidden Markov model |
-
2016
- 2016-12-30 CN CN201611256419.9A patent/CN106803101B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101520651A (en) * | 2009-03-03 | 2009-09-02 | 华中科技大学 | Analysis method for reliability of numerical control equipment based on hidden Markov chain |
| CN101739025A (en) * | 2009-12-03 | 2010-06-16 | 天津理工大学 | Immunity genetic algorithm and DSP failure diagnostic system based thereon |
| CN102054179A (en) * | 2010-12-14 | 2011-05-11 | 广州大学 | Online state monitoring and fault diagnosis device and method for rotary machine |
| CN103217157A (en) * | 2012-01-18 | 2013-07-24 | 北京自动化控制设备研究所 | Inertial navigation/mileometer autonomous integrated navigation method |
| CN103036974A (en) * | 2012-12-13 | 2013-04-10 | 广东省电信规划设计院有限公司 | Cloud computing resource scheduling method and system based on hidden markov model |
| CN103744977A (en) * | 2014-01-13 | 2014-04-23 | 浪潮(北京)电子信息产业有限公司 | Monitoring method and monitoring system for cloud computing system platform |
| CN104793606A (en) * | 2015-04-15 | 2015-07-22 | 浙江大学 | Industrial fault diagnosis method based on improved KPCA (kernel principal component analysis) and hidden Markov model |
| CN105137328A (en) * | 2015-07-24 | 2015-12-09 | 四川航天系统工程研究所 | Analog integrated circuit early-stage soft fault diagnosis method and system based on HMM |
| CN106226097A (en) * | 2016-09-14 | 2016-12-14 | 西安理工大学 | Bullet train airduct safe condition diagnostic method based on hidden Markov model |
Non-Patent Citations (2)
| Title |
|---|
| 刘江等.《基于灰色理论的列车组合定位轮径校准方法研究》.《铁道学报》.2011,第33卷(第5期),第54-59页. * |
| 张增银等.《基于GEP和Baum-Welch算法训练HMM模型的研究》.《计算机工程与设计》.2010,第31卷(第9期),第2027-2029，2069页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106803101A (en) | 2017-06-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106803101B (en) | Odometer method for diagnosing faults based on Hidden Markov Model | |
| CN106443598B (en) | Radar fence based on convolutional neural networks cooperates with track spoofing distinguishing disturbance method | |
| CN108960303B (en) | Unmanned aerial vehicle flight data anomaly detection method based on LSTM | |
| CN102789545B (en) | Based on the Forecasting Methodology of the turbine engine residual life of degradation model coupling | |
| CN104408924B (en) | A kind of urban road abnormal traffic stream detection method based on coupled hidden markov model | |
| CN111832225A (en) | Method for constructing driving condition of automobile | |
| CN112669594B (en) | Method, device, equipment and storage medium for predicting traffic road conditions | |
| CN110132603A (en) | Boat diesel engine Fault Locating Method based on union confidence rule base and ant group algorithm | |
| CN111079836A (en) | Process data fault classification method based on pseudo label method and weak supervised learning | |
| CN112800855B (en) | Non-invasive real-time fault monitoring method for train bogie | |
| CN109964182A (en) | Method and system for vehicle analysis | |
| CN106647514A (en) | Cement enterprise carbon emission real-time on-line monitoring management system | |
| CN109799804B (en) | Diagnostic algorithm evaluation method and system based on random fault injection | |
| CN113822201B (en) | Deep learning method for underwater object shape recognition based on flow field velocity component time course | |
| CN114440900A (en) | Improved hidden Markov model map matching method and device | |
| CN111144462B (en) | Unknown individual identification method and device for radar signals | |
| JP5348941B2 (en) | Method and system for estimating movement state of portable terminal device | |
| CN115600051A (en) | Intelligent track maneuvering detection method and device based on short arc space-based optical observation | |
| CN115099328A (en) | Traffic flow prediction method, system, device and storage medium based on countermeasure network | |
| CN106886467A (en) | Method for optimizing is tested in multitask based on the comprehensive multi-target evolution of packet | |
| CN113743592A (en) | Telemetry data anomaly detection method based on GAN | |
| CN109685331A (en) | A kind of high-speed rail bogie sensor fault diagnosis method based on machine learning | |
| Wu et al. | Traffic sensor health monitoring using spatiotemporal graphical modeling | |
| Zeng et al. | Rail break prediction and cause analysis using imbalanced in-service train data | |
| CN116304969A (en) | Vehicle track multi-mode prediction method considering road information based on LSTM-GNN |
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 |