CN110288046A - A kind of failure prediction method based on wavelet neural network and Hidden Markov Model - Google Patents
A kind of failure prediction method based on wavelet neural network and Hidden Markov Model Download PDFInfo
- Publication number
- CN110288046A CN110288046A CN201910587643.3A CN201910587643A CN110288046A CN 110288046 A CN110288046 A CN 110288046A CN 201910587643 A CN201910587643 A CN 201910587643A CN 110288046 A CN110288046 A CN 110288046A
- Authority
- CN
- China
- Prior art keywords
- state
- hidden
- probability
- moment
- indicate
- 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.)
- Granted
Links
Classifications
-
- 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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- 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
- G06F18/29—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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses a kind of failure prediction method based on wavelet neural network and Hidden Markov Model, steps are as follows: 1, input sample;2, Data Dimensionality Reduction is carried out to sample data using wavelet neural network, updates input layer to hidden layer, the weight and bias of hidden layer to output layer return to 1, otherwise go to 3 if output layer numerical value and input layer difference exceed threshold value;3 output wavelet-neural network models;4 initialization hidden Markov models;5 use different samples, and sample data is replaced using wavelet neural network hidden layer neuron numerical value;6 establish Hidden Markov Model;7 update Hidden Markov Model parameter, design conditions probability using Forward-backward algorithm;If the 8 conditional probability convergences calculated, go to 9, otherwise return to 5;The 9 final Hidden Markov Model of output;10 input device history operation datas to be detected calculate the maximum decline probability of equipment using Hidden Markov Model.
Description
Technical field
The invention discloses a kind of failure prediction methods, more particularly to one kind to be based on wavelet neural network and Hidden Markov
The failure prediction method of model.
Background technique
The maintenance and guarantee of the vehicle-mounted operating facilities of urban track traffic experienced correction maintenance, preventive maintenance and condition maintenarnce
Three phases, wherein condition maintenarnce is intended to monitor and push away using the operation data of equipment in the case that equipment remains to normal work
It a possibility that disconnected equipment fault, to take maintenance measure, for the urban track traffic of high-frequency operation, can mention significantly
High efficiency of operation, security risk caused by reducing because of equipment fault.Currently, being based on nerve net compared to traditional regression forecasting
The failure prediction method of network is that can extract the inherent law of data merely with sample data without establishing visual mathematical model
And substantive characteristics, there are the advantages such as study, associative memory, simple inference, adaptive, but due to present in neural network itself
Limitation, such as neural network are more sensitive to initial weight, and the weight of network be by along the direction of minor betterment gradually
It is adjusted, algorithm can be made to fall into local extremum, weight convergence to local minimum point in this way;Network structure selects generally by passing through
Selection, excessive over-fitting easy to form are tested, too small easy formation does not restrain.
It is specifically included that currently with the means that neural network carries out failure predication and falls into locally optimal solution for anti-locking system,
It is chosen using initial value of the optimization algorithm to neural network, to improve the constringency performance of network;To reduce prediction model
Existing parameter uncertainty problem chooses suitable network structure using PCA scheduling algorithm, to be different from traditional experience
Selection mode.Reference
[1] Liu Haoran, Zhao Cuixiang, Li Xuan wait a kind of optimum algorithm of multi-layer neural network research based on improved adaptive GA-IAGA of
[J] Chinese journal of scientific instrument, 2016,37 (7): 1573-1580.
[2] probabilistic neural network structure optimization [J] Tsinghua University journal (natural science of Xing Jie, the Xiao Deyun based on PCA
Version), 2008,48 (1): 141-144.
[3]Bar-Itzhack I Y,Oshman Y.Attitude determination from vector
observations:quaternion estimation[J].Aerospace and Electronic Systems,IEEE
Transactions on,1985(1):128-136.
[4] hexapod robot independent navigation closed-loop control of Kong Lingwen, Li Pengyong, the Du Qiaoling based on fuzzy neural network
System designs [J] robot, 2018,40 (1): 16-23.
[5] implementation method [J] the equipment management of failure predication technology and maintenance under Li Xinjie condition maintenarnce mode, 2017
(17):68-70.
[6] Chen Xingtian, Xiong little Fu, Qi Xiaoguang, the big data compressing method for waiting a kind of for relay protection state evaluation
[J] Proceedings of the CSEE, 2015,35 (3): 538-548.
[7] Liu Nan designs [J] modern industry based on the aviation machine system diagnosability algorithm of hidden Markov model
It is economical and information-based, 2016,6 (5): 44-45.
[8] Wang Xiaofeng, Xia Jing, Han Jie wait to study [J] based on the Steam Turbine Fault Diagnosis Methods of hidden Markov model
China Construction Machinery journal, 2016,14 (6)
Summary of the invention
The invention proposes a kind of failure prediction method based on wavelet neural network and Hidden Markov Model, utilizes mind
The extraction that through network the data of acquisition are carried out with failure cause, achievees the purpose that Data Dimensionality Reduction, recycles Hidden Markov Model
To failure carry out probabilistic forecasting, for guarantee network structure and Weight selected reliability, using Algorithms of Wavelet Analysis, with small echo letter
The hidden neuron excitation function to replace network is counted, the weight and hidden layer threshold value of input layer to hidden layer are respectively by wavelet function
Scale and translation parameters replace.Key step of the invention are as follows: step 1: the acquisition of sample data.Step 2: small echo is utilized
Analysis building neural network hidden layer neuron network, is constructed automatic coding machine, is dropped using neural network to sample data
Dimension.Step 3: equipment fault prediction is carried out using Hidden Markov Model.
In order to solve problem above, present invention employs following technical solutions: one kind is based on wavelet neural network and hidden horse
The failure prediction method of Er Kefu model, which comprises the following steps:
Step 1: the acquisition of sample data, including history operation data, maintenance data, environmental data, history operation data
Refer to the time of equipment failure-free operation, maintenance data refer to the time being safely operated after the number of maintenance of equipment, maintenance, environmental data
Refer to electric current, the voltage, running temperature, humidity, the vibration degree of mobile unit of pcb board;
Step 2: neuroid is established and Data Dimensionality Reduction;
It is mutually indepedent due to that can not accomplish between data collected, for example, equipment service time inherently by environment
The influence of factor, and maintenance frequency also will affect the time operated normally after maintenance of equipment, and these factors can not utilize standard
Data model distinguish, there are correlations for the data of sensor acquisition itself, to prevent the over-fitting to data, need pair
Data carry out dimension-reduction treatment, and data are indicated in the form of mutually independent.
The excitation function of traditional neural network hidden layer is replaced to construct three layers of neuroid using wavelet function, by small echo
Weight of the scale and translation function of function as input layer to hidden layer, the mode for replacing traditional empirical value to choose avoid
There is local convergence, data are approached using wavelet function, improve the iteration speed of traditional neural network.
Step 2.1 primary condition:
Neural network input sample is determined to the initial connection weight of hidden layer neuron and biasing, hidden layer neuron is extremely
The initial connection weight of output layer and biasing, input layer, hidden layer, output layer neuron number;
In primary condition, after non-failure operation time of selected equipment, temperature, humidity, voltage, maintenance frequency, maintenance
Neuron of the fault-free service time as neural network input layer and output layer, hidden layer neuron number are 3, initial weight
It is 1/7, is biased to the random value of [- 0.25,0.25].
Step 2.2 hidden layer excitation function:
In formula:Indicate wavelet function;aj, bjIndicate the scaling function and translation function of j-th of neuron of hidden layer, x table
Show the signal that is input to hidden layer of the input signal after weight and biasing calculate;Then k-th of neural network output may be expressed as:
In formula, fk(x) k-th of output valve of neural network, x are indicatediIndicate xthiA sample i-th dimension input, n indicate to hide
Layer neuron number, m indicate that sample inputs number, and n < m, wkjIndicate the company that j-th of neuron of hidden layer is exported to k-th
Meet weight, wjiIndicate i-th of input sample xiTo the connection weight of j-th of neuron of hidden layer, λiIndicate input layer biasing, λj
Hidden layer biasing is indicated, by fk(x) it is divided into three parts:
(1) input of j-th of neuron of hidden layer:
(2) output of j-th of neuron of hidden layer:
(3) output of k-th of node of output layer:
Step 2.3 autocoding:
Input function is approached using the neuron output of step 2.2, defines system error function:
Ask error function E to wkj、wji、λj、λi, scaling function ajWith translation function bjPartial derivative:
Step 2.4: above-mentioned partial derivative is directed to, using gradient descent algorithm to wkj、wji、λj、λi、ajAnd bjIt is updated, it is fixed
The Learning Step of adopted gradient descent algorithm is β, then by p+1 parameter factors of p-th of Sample Refreshment are as follows:
The next sample of step 2.5 training, return step 2.2 calculate neural network according to updated parameter factors
Output, compared with initial data, computing system error, if error amount is less than the error threshold of setting, judgement exports result at this time
Initial data is approached, deconditioning, hidden layer is the single order character representation of system at this time;
Probabilistic forecasting of the step 3 based on Hidden Markov Model:
According to step 2, from initial data XN×mExtract the failure sequence H of equipmentN×r=[h1,h2,...,hr], r is number
According to characteristic dimension, and r < m, N are number of samples, and m indicates that the data dimension for including in each sample, h are indicated from each
The required data extracted in sample
Step 3.1 primary condition:
Hidden Markov Model is denoted as λ=(N, M, π, A, B),
(1) N indicates the hidden state number of Hidden Markov Model, using the change procedure of device parameter as hidden state
Random process, N=(N1,N2,N3,...,Nn), the hidden state of t moment system is qt, qt∈N;
(2) M indicates the observation state of system, indicates the failure sequence that neural network is extracted, M=[M1,M2,...,Mr], t
When etching system observation state be Ot, Ot∈M;
(3) π indicates the probability matrix of initial hidden, π=(π1,π2,...,πn), πi=P (q1=Ni),1≤i≤n;
q1The original state of expression system, NiIndicate that i-th of hidden state of hidden Markov model, p () indicate system
Original state is the probability of i-th of hidden state;
(4) A is state-transition matrix, indicates that equipment is transferred to the probability square of another hidden state by current hidden state
Battle array, A=(aij)n×n, n × n representing matrix dimension, wherein aijIndicate the probability that state j is transferred to by state i, aij=P (qt+1=
Nj|qt=Ni)1≤i,j≤n;
qtThe hidden state of expression system t moment, qt+1The hidden state at expression system t+1 moment, qt=NiExpression system
Belong to i-th of hidden state in t moment, p () indicates system in t moment from NiState is to NjThe probability of state transfer, n is system
Hidden state number.
(5) B is observation probability matrix, indicates the hidden state of equipment to the transition probability of observation state, B=
(bjk)r×n, bjkIt indicates the transition probability of hidden state k to observation state j, remembers bjk=bj(k), bj(k)=P (Ot=Mk|qt=
Nj), 1≤j≤n, 1≤k≤r, QtThe observation state of expression system t moment, MkExpression system t moment belongs to k-th of observation shape
State, qtThe hidden state of expression system t moment, j are to hide layer state order, and n is the total status number of hidden layer, and k is observation state
Order, r are the total status number of observation layer, and p () indicates that system t moment goes to the general of k-th of observation state from j-th of hidden state
Rate.Step 3.2 establishes fault model
The acquisition data of selected equipment different conditions, including equipment normal operating condition, 4 kinds of equipment is not under non-failure conditions
With the wear degradation state and malfunction of degree, establishing Hidden Markov has model, using Forward-backward algorithm to collecting
Device status data carry out model training, determine the state-transition matrix of equipment hidden state, calculate that steps are as follows:
(1) to hidden Markov model matrix initialisation: π=(π1,π2,...,πn), A=(aij)n×n, B=(bjk)r×n;
(2) observation state sequence of the T group measurement data as model is taken from sample data;
(3) hidden layer of neural network is mapped data into according to the calculated result of wavelet neural network, Data Dimensionality Reduction is defeated
Observation sequence O=[O out1,O2,...OT];
(4) to probability a before definitiont(i), indicate that t moment (t < T) hidden state is Ni, observation sequence is [O1,O2,
...Ot] probability:
a1(i)=πibi(O1) (16)
Wherein, a1(i) the forward direction probability of i-th of hidden state of system initial time is indicated;πiIndicate probability matrix
The probability matrix of i hidden state;bi(O1) indicate system initial time hidden state be NiObserve O1Probability;Nj
J-th of hidden state of expression system;λ indicates hidden Markov model;at(j)ajiExpression moment t hidden state is Nj, observe sequence
It is classified as [O1,O2,...Ot], moment t+1 hidden state is NiProbability;bi(Ot+1) expression hidden state be NiObserve Ot+1's
Probability;P () indicates that in t+1 moment observation sequence be [O1,O2,...Ot,Ot+1], hidden state NiProbability.
(5) backward probability β is definedt(i), indicate that t moment (t < T) hidden state is Ni, the t+1 moment to T moment observes sequence
It is classified as [Ot+1,Ot+2,...OT] probability:
βT(i)=1 (18)
Wherein, qt=NiExpression t moment hidden state is Ni;λ indicates hidden Markov model;βt+1(j) the t+1 moment is indicated
Hidden state is NjBackward probability;aijIndicate the probability that state j is transferred to by state i;aijβt+1(j) indicate that the t+1 moment hides
State is Nj, t moment hidden state is NiProbability;aijbj(Ot+1)βt+1(j) indicate that observation sequence is [Ot+1,Ot+2,...OT],
T+1 moment hidden state is Nj, t moment hidden state is NiProbability;It is N that p (), which indicates that t moment hides layer state,iProbability.
(6) the sum of forward direction probability and the backward probability of current observation sequence are calculated
at(i) indicate that t moment hidden layer is NiForward direction probability, βt(i) indicate that t moment hidden layer is NiBackward probability,
N is hidden layer status number.
Given observation sequence, is in state N in moment t equipmentiProbability:
at(i) indicate that t moment hidden layer is NiForward direction probability, βt(i) indicate that t moment hidden layer is NiBackward probability,
N is hidden layer status number.
(7) observation sequence is given, is in state N in moment t+1 equipmentiProbability
at(i) indicate that t moment hidden layer is NiForward direction probability, βt+1(j) indicate that t+1 moment hidden layer is NjIt is backward general
Rate, n are hidden layer status number, aijIndicate the probability that state j is transferred to by state i, bj(Ot+1) indicate that t moment hides layer state
For Nj, the t+1 moment observes Ot+1Probability.
(8) if P (O | λ) does not restrain, return step (2), Hidden Markov Model parameter is otherwise calculated:
Sample number assuming that P when (O | λ) convergence for calculating is D, then:
πiExpression state is the probability of i, and the average value of probability is acquired for each sample;Indicate initial time d
Sample state is NiProbability.
Indicate t moment, d-th of sample, by state NiIt is transferred to state NjProbability, T be acquisition at the time of number,
D is sample number;aijIt indicates finally by state NiIt is transferred to state NjProbability;Expression d-th of sample state of t moment is Ni
Probability.
bj(k) transition probability of hidden state k to observation state j is indicated.
(9) training terminates, and exports final Hidden Markov Model λ=(N, M, π, A, B);
Step 3.3 failure predication
When carrying out failure predication to equipment, historical perspective sequence O=[O is exported1,O2,...OT], according to trained hidden horse
Er Kefu model calculates the degenerate state of the maximum possible locating for it, and steps are as follows:
(1) state initialization:
δ1(i)=πibi(O1), i=1,2 ..., n (26)
N is the state number of hidden layer, πiIndicate initial time state NiProbability, bi(O1) indicate initial time observation
For O1, state NiProbability, δ1(i) indicate that initial time observes O1System is in NiState.
The possible state for indicating initialization system, is set to 0 entirely.
(2) state of recursion moment t:
δt(i)=max (δt-1(1),δt-1(2),...,δt-1(n))·bi(Ot) (28)
max(δt-1(1),δt-1(2),...,δt-1(n)) the maximum possible shape being in n state of etching system when t-1 is indicated
State;bi(Ot) indicate to observe Ot, system is in state NiProbability;δt(i) indicate that t moment observation sequence is O=[O1,
O2,...Ot] when system be in state NiProbability.
ankIndicate that etching system is in state N when t-1n, t moment is in state NkProbability;It indicates at t moment system
In the maximum possible probability of state k.
(3) momentMaximum value indicates the maximum possible state of equipment and the degenerate state of equipment.
In step 2, after non-failure operation time of selected equipment, temperature, humidity, voltage, maintenance frequency, maintenance without reason
Hinder neuron of the service time as neural network input layer and output layer, hidden layer neuron number is 3, initial weight 1/
7, it is biased to the random value of [- 0.25,0.25].
In step 3.1 (4) (5), model initial value B be uniformly distributed, and all parameters of B and for 1, π=(1,
0,...,0);
In step 3, using the normal condition of equipment, degenerate state 20%, 40%, 60%, 80% and malfunction are made
For the hidden state of equipment.
The present invention has the advantages that compared with the immediate prior art
The present invention is used replaces the excitation function of traditional neural network hidden layer to construct three layers of neuron using wavelet function
Network replaces traditional empirical value to choose using the scale of wavelet function and translation function as the weight of input layer to hidden layer
Mode, avoid the occurrence of local convergence, data approached using wavelet function, improve traditional neural network iteration speed
Degree;
The present invention carries out dimension-reduction treatment to data using neural network, avoid directlying adopt Hidden Markov Model to
Data are handled, and the computation complexity of Hidden Markov Model is reduced, and improve failure predication rate;
The present invention replaces initial data using wavelet neural network hidden layer neuron, reduces the phase between initial data
Guan Xing.
The present invention carries out failure predication to equipment using the operation data based on time series using Hidden Markov Model,
Hidden state using the degenerate state of equipment as equipment, improves device predicted confidence level.
Detailed description of the invention
A kind of Fig. 1 failure prediction method process based on wavelet neural network and Hidden Markov Model proposed by the present invention
Figure.
Fig. 2 is a kind of failure prediction method based on wavelet neural network and Hidden Markov Model proposed by the present invention,
Neural network dimensionality reduction model.
Fig. 3 is a kind of failure prediction method based on wavelet neural network and Hidden Markov Model proposed by the present invention,
Hidden Markov Model.
Specific embodiment
As shown in Figures 1 to 3, a kind of failure prediction method based on wavelet neural network and Hidden Markov Model, including
Following steps:
Step 1: the acquisition of sample data, including history operation data, maintenance data, environmental data, history operation data
Refer to the time of equipment failure-free operation, maintenance data refer to the time being safely operated after the number of maintenance of equipment, maintenance, environmental data
Refer to electric current, the voltage, running temperature, humidity, the vibration degree of mobile unit of pcb board;
Step 2: neuroid is established and Data Dimensionality Reduction;
It is mutually indepedent due to that can not accomplish between data collected, for example, equipment service time inherently by environment
The influence of factor, and maintenance frequency also will affect the time operated normally after maintenance of equipment, and these factors can not utilize standard
Data model distinguish, there are correlations for the data of sensor acquisition itself, to prevent the over-fitting to data, need pair
Data carry out dimension-reduction treatment, and data are indicated in the form of mutually independent.
The excitation function of traditional neural network hidden layer is replaced to construct three layers of neuroid using wavelet function, by small echo
Weight of the scale and translation function of function as input layer to hidden layer, the mode for replacing traditional empirical value to choose avoid
There is local convergence, data are approached using wavelet function, improve the iteration speed of traditional neural network.
Step 2.1 primary condition:
Neural network input sample is determined to the initial connection weight of hidden layer neuron and biasing, hidden layer neuron is extremely
The initial connection weight of output layer and biasing, input layer, hidden layer, output layer neuron number;
Step 2.2 hidden layer excitation function:
In formula:Indicate wavelet function;aj, bjIndicate the scaling function and translation function of j-th of neuron of hidden layer, x table
Show the signal that is input to hidden layer of the input signal after weight and biasing calculate;Then k-th of neural network output may be expressed as:
In formula, fk(x) k-th of output valve of neural network, x are indicatediIndicate xthiA sample i-th dimension input, n indicate to hide
Layer neuron number, m indicate that sample inputs number, and n < m, wkjIndicate the company that j-th of neuron of hidden layer is exported to k-th
Meet weight, wjiIndicate i-th of input sample xiTo the connection weight of j-th of neuron of hidden layer, λiIndicate input layer biasing, λj
Hidden layer biasing is indicated, by fk(x) it is divided into three parts:
(1) input of j-th of neuron of hidden layer:
(2) output of j-th of neuron of hidden layer:
(3) output of k-th of node of output layer:
Step 2.3 autocoding:
Input function is approached using the neuron output of step 2.2, defines system error function:
Ask error function E to wkj、wji、λj、λi, scale coefficient ajWith translation coefficient bjPartial derivative:
Step 2.4: above-mentioned partial derivative is directed to, using gradient descent algorithm to wkj、wji、λj、λi、ajAnd bjIt is updated, it is fixed
The Learning Step of adopted gradient descent algorithm is β, then by p+1 parameter factors of p-th of Sample Refreshment are as follows:
The next sample of step 2.5 training, return step 2.2 calculate neural network according to updated parameter factors
Output, compared with initial data, computing system error, if error amount is less than the error threshold of setting, judgement exports result at this time
Initial data is approached, deconditioning, hidden layer is the single order character representation of system at this time;
Probabilistic forecasting of the step 3 based on Hidden Markov Model:
According to step 2, from initial data XN×mExtract the failure sequence H of equipmentN×r=[h1,h2,...,hr], r is number
According to characteristic dimension, and r < m, N are number of samples,
M indicates that the data dimension for including in each sample, h indicate the required data extracted from each sample;
Step 3.1 primary condition:
Hidden Markov Model is denoted as λ=(N, M, π, A, B),
(1) N indicates the hidden state number of Hidden Markov Model, using the change procedure of device parameter as hidden state
Random process, N=(N1,N2,N3,...,Nn), the hidden state of t moment system is qt, qt∈N;
(2) M indicates the observation state of system, indicates the failure sequence that neural network is extracted, M=[M1,M2,...,Mr], t
When etching system observation state be Ot, Ot∈M;
(3) π indicates the probability matrix of initial hidden, π=(π1,π2,...,πn), πi=P (q1=Ni),1≤i≤n;
q1The original state of expression system, NiIndicate that i-th of hidden state of hidden Markov model, p () indicate system
Original state is the probability of i-th of hidden state;
(4) A is state-transition matrix, indicates that equipment is transferred to the probability square of another hidden state by current hidden state
Battle array, A=(aij)n×n, n × n representing matrix dimension, wherein aijIndicate the probability that state j is transferred to by state i, aij=P (qt+1=
Nj|qt=Ni)1≤i,j≤n;
qtThe hidden state of expression system t moment, qt+1The hidden state at expression system t+1 moment, qt=NiExpression system
Belong to i-th of hidden state in t moment, p () indicates system in t moment from NiState is to NjThe probability of state transfer, n is system
Hidden state number.
(5) B is observation probability matrix, indicates the hidden state of equipment to the transition probability of observation state, B=
(bjk)r×n, bjkIt indicates the transition probability of hidden state k to observation state j, remembers bjk=bj(k), bj(k)=P (Ot=Mk|qt=
Nj), 1≤j≤n, 1≤k≤r, QtThe observation state of expression system t moment, MkExpression system t moment belongs to k-th of observation shape
State, qtThe hidden state of expression system t moment, j are to hide layer state order, and n is the total status number of hidden layer, and k is observation state
Order, r are the total status number of observation layer, and p () indicates that system t moment goes to the general of k-th of observation state from j-th of hidden state
Rate.Step 3.2 establishes fault model
The acquisition data of selected equipment different conditions, including equipment normal operating condition, 4 kinds of equipment is not under non-failure conditions
With the wear degradation state and malfunction of degree, establishing Hidden Markov has model, using Forward-backward algorithm to collecting
Device status data carry out model training, determine the state-transition matrix of equipment hidden state, calculate that steps are as follows:
(1) to hidden Markov model matrix initialisation: π=(π1,π2,...,πn), A=(aij)n×n, B=(bjk)r×n;
(2) observation state sequence of the T group measurement data as model is taken from sample data;
(3) hidden layer of neural network is mapped data into according to the calculated result of wavelet neural network, Data Dimensionality Reduction is defeated
Observation sequence O=[O out1,O2,...OT];
(4) to probability a before definitiont(i), indicate that t moment (t < T) hidden state is Ni, observation sequence is [O1,O2,
...Ot] probability:
a1(i)=πibi(O1) (16)
Wherein, a1(i) the forward direction probability of i-th of hidden state of system initial time is indicated;πiIndicate probability matrix
The probability matrix of i hidden state;bi(O1) indicate system initial time hidden state be NiObserve O1Probability;Nj
J-th of hidden state of expression system;λ indicates hidden Markov model;at(j)ajiExpression moment t hidden state is Nj, observe sequence
It is classified as [O1,O2,...Ot], moment t+1 hidden state is NiProbability;bi(Ot+1) expression hidden state be NiObserve Ot+1's
Probability;P () indicates that in t+1 moment observation sequence be [O1,O2,...Ot,Ot+1], hidden state NiProbability.
(5) backward probability β is definedt(i), indicate that t moment (t < T) hidden state is Ni, the t+1 moment to T moment observes sequence
It is classified as [Ot+1,Ot+2,...OT] probability:
βT(i)=1 (18)
Wherein, qt=NiExpression t moment hidden state is Ni;λ indicates hidden Markov model;βt+1(j) the t+1 moment is indicated
Hidden state is NjBackward probability;aijIndicate the probability that state j is transferred to by state i;aijβt+1(j) indicate that the t+1 moment hides
State is Nj, t moment hidden state is NiProbability;aijbj(Ot+1)βt+1(j) indicate that observation sequence is [Ot+1,Ot+2,...OT],
T+1 moment hidden state is Nj, t moment hidden state is NiProbability;It is N that p (), which indicates that t moment hides layer state,iProbability.
(6) the sum of forward direction probability and the backward probability of current observation sequence are calculated
at(i) indicate that t moment hidden layer is NiForward direction probability, βt(i) indicate that t moment hidden layer is NiBackward probability,
N is hidden layer status number.
(7) observation sequence is given, is in state N in moment t equipmentiProbability:
at(i) indicate that t moment hidden layer is NiForward direction probability, βt(i) indicate that t moment hidden layer is NiBackward probability,
N is hidden layer status number.
(8) observation sequence is given, is in state N in moment t+1 equipmentiProbability
at(i) indicate that t moment hidden layer is NiForward direction probability, βt+1(j) indicate that t+1 moment hidden layer is NjIt is backward general
Rate, n are hidden layer status number, aijIndicate the probability that state j is transferred to by state i, bj(Ot+1) indicate that t moment hides layer state
For Nj, the t+1 moment observes Ot+1Probability.
(9) if P (O | λ) does not restrain, return step (2), Hidden Markov Model parameter is otherwise calculated:
Sample number assuming that P when (O | λ) convergence for calculating is D, then:
πiExpression state is the probability of i, and the average value of probability is acquired for each sample;Indicate initial time d
Sample state is NiProbability.
Indicate t moment, d-th of sample, by state NiIt is transferred to state NjProbability, T be acquisition at the time of number,
D is sample number;aijIt indicates finally by state NiIt is transferred to state NjProbability;Expression d-th of sample state of t moment is Ni
Probability.
bj(k) transition probability of hidden state k to observation state j is indicated.
(10) training terminates, and exports final Hidden Markov Model λ=(N, M, π, A, B);
Step 3.3 failure predication
When carrying out failure predication to equipment, historical perspective sequence O=[O is exported1,O2,...OT], according to trained hidden horse
Er Kefu model calculates the degenerate state of the maximum possible locating for it, and steps are as follows:
(1) state initialization:
δ1(i)=πibi(O1), i=1,2 ..., n (26)
N is the state number of hidden layer, πiIndicate initial time state NiProbability, bi(O1) indicate initial time observation
For O1, state NiProbability, δ1(i) indicate that initial time observes O1System is in NiState.
The possible state for indicating initialization system, is set to 0 entirely.
(2) state of recursion moment t:
δt(i)=max (δt-1(1),δt-1(2),...,δt-1(n))·bi(Ot) (28)
max(δt-1(1),δt-1(2),...,δt-1(n)) the maximum possible shape being in n state of etching system when t-1 is indicated
State;bi(Ot) indicate to observe Ot, system is in state NiProbability;δt(i) indicate that t moment observation sequence is O=[O1,
O2,...Ot] when system be in state NiProbability.
ankIndicate that etching system is in state N when t-1n, t moment is in state NkProbability;It indicates at t moment system
In the maximum possible probability of state k.
(3) momentMaximum value indicates the maximum possible state of equipment and the degenerate state of equipment.
In step 3.1 (4) (5), model initial value B be uniformly distributed, and all parameters of B and for 1, π=(1,
0,...,0);
In step 3.2, using the normal condition of equipment, degenerate state 20%, 40%, 60%, 80% and malfunction
Hidden state as equipment.
The foregoing is only a preferred embodiment of the present invention, is not restricted to the present invention, for the technology of this field
For personnel, the invention may be variously modified and varied.All within the spirits and principles of the present invention, made any to repair
Change, equivalent replacement, improvement etc., should be included within scope of the presently claimed invention.
Claims (4)
1. a kind of failure prediction method based on wavelet neural network and Hidden Markov Model, which is characterized in that including following
Step:
Step 1: the acquisition of sample data, including history operation data, maintenance data, environmental data, history operation data, which refers to, to be set
The time of standby failure-free operation, maintenance data refer to the time being safely operated after the number of maintenance of equipment, maintenance, and environmental data refers to
The electric current of pcb board, voltage, running temperature, humidity, the vibration degree of mobile unit;
Step 2: neuroid is established and Data Dimensionality Reduction;
The excitation function of traditional neural network hidden layer is replaced to construct three layers of neuroid using wavelet function, by wavelet function
Weight as input layer to hidden layer of scale and translation function, the mode for replacing traditional empirical value to choose avoids the occurrence of
Local convergence approaches data using wavelet function, improves the iteration speed of traditional neural network;
Step 2.1 primary condition:
Determine neural network input sample to the initial connection weight of hidden layer neuron and biasing, hidden layer neuron to output
Layer initial connection weight and biasing, input layer, hidden layer, output layer neuron number;
Step 2.2 hidden layer excitation function:
In formula:Indicate wavelet function;aj, bjIndicate that the scale coefficient and translation coefficient of j-th of neuron of hidden layer, x indicate defeated
Enter the signal that is input to hidden layer of the signal after weight and biasing calculate;Then k-th of neural network output may be expressed as:
In formula, fk(x) k-th of output valve of neural network, x are indicatediIndicate xthiA sample i-th dimension input, n indicate hidden layer mind
Through first number, m indicates that sample inputs number, and n < m, wkjIndicate the connection weight that j-th of neuron of hidden layer is exported to k-th
Value, wjiIndicate i-th of input sample xiTo the connection weight of j-th of neuron of hidden layer, λiIndicate input layer biasing, λjIt indicates
Hidden layer biasing, by fk(x) it is divided into three parts:
(1) input of j-th of neuron of hidden layer:
(2) output of j-th of neuron of hidden layer:
(3) output of k-th of node of output layer:
Step 2.3 autocoding:
Input function is approached using the neuron output of step 2.2, defines system error function:
Ask error function E to wkj、wji、λj、λi, scale coefficient ajWith translation coefficient bjPartial derivative:
Step 2.4: above-mentioned partial derivative is directed to, using gradient descent algorithm to wkj、wji、λj、λi、ajAnd bjIt is updated, definition ladder
The Learning Step for spending descent algorithm is β, then by p+1 parameter factors of p-th of Sample Refreshment are as follows:
The next sample of step 2.5 training, return step 2.2 calculate the output of neural network according to updated parameter factors,
Compared with initial data, computing system error determines that output result approaches at this time if error amount is less than the error threshold of setting
Initial data, deconditioning, hidden layer is the single order character representation of system at this time;
Probabilistic forecasting of the step 3 based on Hidden Markov Model:
According to step 2, from initial data XN×mExtract the failure sequence H of equipmentN×r=[h1,h2,...,hr], r is the spy of data
Dimension is levied, and r < m, N are number of samples, m indicates that the data dimension for including in each sample, h are indicated from each sample
The required data extracted;
Step 3.1 primary condition:
Hidden Markov Model is denoted as λ=(N, M, π, A, B),
(1) N indicates the hidden state number of Hidden Markov Model, using the change procedure of device parameter as the random of hidden state
Process, N=(N1,N2,N3,...,Nn), the hidden state of t moment system is qt, qt∈N;
(2) M indicates the observation state of system, indicates the failure sequence that neural network is extracted, M=[M1,M2,...,Mr], t moment
The observation state of system is Ot, Ot∈M;
(3) π indicates the probability matrix of initial hidden, π=(π1,π2,...,πn), πi=P (q1=Ni),1≤i≤n;q1Table
Show the original state of system, NiIndicate that i-th of hidden state of hidden Markov model, p () indicate that system initial state is i-th
The probability of a hidden state;
(4) A is state-transition matrix, indicates that equipment is transferred to the probability matrix of another hidden state, A by current hidden state
=(aij)n×n, n × n representing matrix dimension, wherein aijIndicate the probability that state j is transferred to by state i,
aij=P (qt+1=Nj|qt=Ni)1≤i,j≤n;
qtThe hidden state of expression system t moment, qt+1The hidden state at expression system t+1 moment, qt=NiExpression system is in t
Belong to i-th of hidden state quarter, p () indicates system in t moment from NiState is to NjThe probability of state transfer, n are the hidden of system
Hide status number.
(5) B is observation probability matrix, indicates the hidden state of equipment to the transition probability of observation state, B=(bjk)r×n, bjk
It indicates the transition probability of hidden state k to observation state j, remembers bjk=bj(k), bj(k)=P (Ot=Mk|qt=Nj),1≤j≤n,
1≤k≤r, QtThe observation state of expression system t moment, MkExpression system t moment belongs to k-th of observation state, qtExpression system t
The hidden state at moment, j are to hide layer state order, and n is the total status number of hidden layer, and k is observation state order, and r is observation layer
Total status number, p () indicate that system t moment goes to the probability of k-th of observation state from j-th of hidden state;
Step 3.2 establishes fault model
The acquisition data of selected equipment different conditions, including equipment normal operating condition, 4 kinds of equipment different journeys under non-failure conditions
The wear degradation state and malfunction of degree, establishing Hidden Markov has model, is set using Forward-backward algorithm to collected
Standby status data carries out model training, determines the state-transition matrix of equipment hidden state, steps are as follows for calculating:
(1) to hidden Markov model matrix initialisation: π=(π1,π2,...,πn), A=(aij)n×n, B=(bjk)r×n;
(2) observation state sequence of the T group measurement data as model is taken from sample data;
(3) hidden layer of neural network, Data Dimensionality Reduction are mapped data into according to the calculated result of wavelet neural network, output is seen
Sequencing column O=[O1,O2,...OT];
(4) to probability a before definitiont(i), indicate that t moment (t < T) hidden state is Ni, observation sequence is [O1,O2,...Ot]
Probability:
a1(i)=πibi(O1) (16)
Wherein, a1(i) the forward direction probability of i-th of hidden state of system initial time is indicated;πiIndicate i, probability matrix
The probability matrix of hidden state;bi(O1) indicate system initial time hidden state be NiObserve O1Probability;NjTable
Show j-th of hidden state of system;λ indicates hidden Markov model;at(j)ajiExpression moment t hidden state is Nj, observation sequence
For [O1,O2,...Ot], moment t+1 hidden state is NiProbability;bi(Ot+1) expression hidden state be NiObserve Ot+1It is general
Rate;P () indicates that in t+1 moment observation sequence be [O1,O2,...Ot,Ot+1], hidden state NiProbability;
(5) backward probability β is definedt(i), indicate that t moment (t < T) hidden state is Ni, the t+1 moment to T moment observation sequence is
[Ot+1,Ot+2,...OT] probability:
βT(i)=1 (18)
Wherein, qt=NiExpression t moment hidden state is Ni;λ indicates hidden Markov model;βt+1(j) indicate that the t+1 moment hides
State is NjBackward probability;aijIndicate the probability that state j is transferred to by state i;aijβt+1(j) t+1 moment hidden state is indicated
For Nj, t moment hidden state is NiProbability;aijbj(Ot+1)βt+1(j) indicate that observation sequence is [Ot+1,Ot+2,...OT], t+1
Moment hidden state is Nj, t moment hidden state is NiProbability;It is N that p (), which indicates that t moment hides layer state,iProbability;
(6) the sum of forward direction probability and the backward probability of current observation sequence are calculated by formula (17) and (19)
at(i) indicate that t moment hidden layer is NiForward direction probability, βt(i) indicate that t moment hidden layer is NiBackward probability, n be it is hidden
Hide layer state number;Given observation sequence O=[O1,O2,...Ot], state N is in moment t equipmentiProbability γt(i) are as follows:
at(i) indicate that t moment hidden layer is NiForward direction probability, βt(i) indicate that t moment hidden layer is NiBackward probability, n be it is hidden
Hide layer state number;
(7) observation sequence O=[O is given1,O2,...Ot,Ot+1], in moment t+1 equipment by state NiIt is transferred to state NjProbability
at(i) indicate that t moment hidden layer is NiForward direction probability, βt+1(j) indicate that t+1 moment hidden layer is NjBackward probability, n
For hidden layer status number, aijIndicate the probability that state j is transferred to by state i, bj(Ot+1) indicate that t moment hides layer state for Nj,
The t+1 moment observes Ot+1Probability;
(8) if P (O | λ) does not restrain, return step (2), Hidden Markov Model parameter is otherwise calculated:
Sample number assuming that P when (O | λ) convergence for calculating is D, then:
πiExpression state is the probability of i, and the average value of probability is acquired for each sample;Indicate initial time d sample
State is NiProbability;
Indicate t moment, d-th of sample, by state NiIt is transferred to state NjProbability, T be acquisition at the time of number, D is sample
This number;aijIt indicates finally by state NiIt is transferred to state NjProbability;Expression d-th of sample state of t moment is NiIt is general
Rate;
bj(k) transition probability of hidden state k to observation state j is indicated;
(9) training terminates, and exports final Hidden Markov Model λ=(N, M, π, A, B);
Step 3.3 failure predication
When carrying out failure predication to equipment, historical perspective sequence O=[O is exported1,O2,...OT], according to trained hidden Ma Erke
Husband's model calculates the degenerate state of the maximum possible locating for it, and steps are as follows:
(1) state initialization:
δ1(i)=πibi(O1), i=1,2 ..., n (26)
N is the state number of hidden layer, πiIndicate initial time state NiProbability, bi(O1) indicate that initial time is observed O1,
State is NiProbability, δ1(i) indicate that initial time observes O1System is in NiState;
The possible state for indicating initialization system, is set to 0 entirely;
(2) state of recursion moment t:
δt(i)=max (δt-1(1),δt-1(2),...,δt-1(n))·bi(Ot) (28)
max(δt-1(1),δt-1(2),...,δt-1(n)) the maximum possible state being in n state of etching system when t-1 is indicated;bi
(Ot) indicate to observe Ot, system is in state NiProbability;δt(i) indicate that t moment observation sequence is O=[O1,O2,...Ot]
When system be in state NiProbability;
ankIndicate that etching system is in state N when t-1n, t moment is in state NkProbability;Indicate that t moment system is in shape
The maximum possible probability of state k;
(3) momentMaximum value indicates the maximum possible state of equipment and the degenerate state of equipment.
2. a kind of failure prediction method based on wavelet neural network and Hidden Markov Model according to claim 1,
It is characterized in that, in step 2, after non-failure operation time of selected equipment, temperature, humidity, voltage, maintenance frequency, maintenance
Neuron of the fault-free service time as neural network input layer and output layer, hidden layer neuron number are 3, initial weight
It is 1/7, is biased to the random value of [- 0.25,0.25].
3. a kind of failure prediction method based on wavelet neural network and Hidden Markov Model according to claim 1,
It is characterized in that, in step 3, model initial value B be uniformly distributed, and all parameters of B and be 1, π=(1,0 ..., 0);
4. a kind of failure prediction method based on wavelet neural network and Hidden Markov Model according to claim 1,
It is characterized in that, in step 3, using the normal condition of equipment, degenerate state 20%, 40%, 60%, 80% and malfunction
Hidden state as equipment.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910587643.3A CN110288046B (en) | 2019-07-02 | 2019-07-02 | Fault prediction method based on wavelet neural network and hidden Markov model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910587643.3A CN110288046B (en) | 2019-07-02 | 2019-07-02 | Fault prediction method based on wavelet neural network and hidden Markov model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110288046A true CN110288046A (en) | 2019-09-27 |
CN110288046B CN110288046B (en) | 2022-11-18 |
Family
ID=68021625
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910587643.3A Active CN110288046B (en) | 2019-07-02 | 2019-07-02 | Fault prediction method based on wavelet neural network and hidden Markov model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110288046B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111223574A (en) * | 2020-01-14 | 2020-06-02 | 宁波市海洋与渔业研究院 | Penaeus vannamei boone enterohepatic sporulosis early warning method based on big data mining |
CN111259261A (en) * | 2020-01-02 | 2020-06-09 | 中国铁道科学研究院集团有限公司通信信号研究所 | High-speed rail driving network collaborative alarm optimization method based on state transition prediction |
CN111565118A (en) * | 2020-04-17 | 2020-08-21 | 烽火通信科技股份有限公司 | Virtualized network element fault analysis method and system based on multi-observation dimension HMM |
CN111882078A (en) * | 2020-06-28 | 2020-11-03 | 北京交通大学 | Method for optimizing state maintenance strategy of running part of rail transit train |
CN112069045A (en) * | 2020-08-14 | 2020-12-11 | 西安理工大学 | Cloud platform software performance prediction method based on hidden Markov |
CN112257777A (en) * | 2020-10-21 | 2021-01-22 | 平安科技(深圳)有限公司 | Off-job prediction method based on hidden Markov model and related device |
CN113053536A (en) * | 2021-01-15 | 2021-06-29 | 中国人民解放军军事科学院军事医学研究院 | Infectious disease prediction method, system and medium based on hidden Markov model |
CN113298240A (en) * | 2021-07-27 | 2021-08-24 | 北京科技大学 | Method and device for predicting life cycle of servo drive system |
CN116020879A (en) * | 2023-02-15 | 2023-04-28 | 北京科技大学 | Technological parameter-oriented strip steel hot continuous rolling space-time multi-scale process monitoring method and device |
CN117114352A (en) * | 2023-09-15 | 2023-11-24 | 北京阿帕科蓝科技有限公司 | Vehicle maintenance method, device, computer equipment and storage medium |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH10307186A (en) * | 1997-05-07 | 1998-11-17 | Mitsubishi Electric Corp | Signal processor |
US20060064291A1 (en) * | 2004-04-21 | 2006-03-23 | Pattipatti Krishna R | Intelligent model-based diagnostics for system monitoring, diagnosis and maintenance |
CN102867132A (en) * | 2012-10-16 | 2013-01-09 | 南京航空航天大学 | Aviation direct-current converter online fault combined prediction method based on fractional order wavelet transformation |
CN104504296A (en) * | 2015-01-16 | 2015-04-08 | 湖南科技大学 | Gaussian mixture hidden Markov model and regression analysis remaining life prediction method |
CN105841961A (en) * | 2016-03-29 | 2016-08-10 | 中国石油大学(华东) | Bearing fault diagnosis method based on Morlet wavelet transformation and convolutional neural network |
CN105834835A (en) * | 2016-04-26 | 2016-08-10 | 天津大学 | Method for monitoring tool wear on line based on multiscale principal component analysis |
CN106405384A (en) * | 2016-08-26 | 2017-02-15 | 中国电子科技集团公司第十研究所 | Simulation circuit health state evaluation method |
CN106599920A (en) * | 2016-12-14 | 2017-04-26 | 中国航空工业集团公司上海航空测控技术研究所 | Aircraft bearing fault diagnosis method based on coupled hidden semi-Markov model |
CN107122802A (en) * | 2017-05-02 | 2017-09-01 | 哈尔滨理工大学 | A kind of fault detection method based on the rolling bearing for improving wavelet neural network |
CN108090427A (en) * | 2017-12-07 | 2018-05-29 | 上海电机学院 | Fault Diagnosis of Gear Case method based on flock of birds algorithm and Hidden Markov Model |
CN108490807A (en) * | 2018-05-09 | 2018-09-04 | 南京恩瑞特实业有限公司 | Train fault analogue system and test method |
CN108763654A (en) * | 2018-05-03 | 2018-11-06 | 国网江西省电力有限公司信息通信分公司 | A kind of electrical equipment fault prediction technique based on Weibull distribution and hidden Semi-Markov Process |
-
2019
- 2019-07-02 CN CN201910587643.3A patent/CN110288046B/en active Active
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH10307186A (en) * | 1997-05-07 | 1998-11-17 | Mitsubishi Electric Corp | Signal processor |
US20060064291A1 (en) * | 2004-04-21 | 2006-03-23 | Pattipatti Krishna R | Intelligent model-based diagnostics for system monitoring, diagnosis and maintenance |
CN102867132A (en) * | 2012-10-16 | 2013-01-09 | 南京航空航天大学 | Aviation direct-current converter online fault combined prediction method based on fractional order wavelet transformation |
CN104504296A (en) * | 2015-01-16 | 2015-04-08 | 湖南科技大学 | Gaussian mixture hidden Markov model and regression analysis remaining life prediction method |
CN105841961A (en) * | 2016-03-29 | 2016-08-10 | 中国石油大学(华东) | Bearing fault diagnosis method based on Morlet wavelet transformation and convolutional neural network |
CN105834835A (en) * | 2016-04-26 | 2016-08-10 | 天津大学 | Method for monitoring tool wear on line based on multiscale principal component analysis |
CN106405384A (en) * | 2016-08-26 | 2017-02-15 | 中国电子科技集团公司第十研究所 | Simulation circuit health state evaluation method |
CN106599920A (en) * | 2016-12-14 | 2017-04-26 | 中国航空工业集团公司上海航空测控技术研究所 | Aircraft bearing fault diagnosis method based on coupled hidden semi-Markov model |
CN107122802A (en) * | 2017-05-02 | 2017-09-01 | 哈尔滨理工大学 | A kind of fault detection method based on the rolling bearing for improving wavelet neural network |
CN108090427A (en) * | 2017-12-07 | 2018-05-29 | 上海电机学院 | Fault Diagnosis of Gear Case method based on flock of birds algorithm and Hidden Markov Model |
CN108763654A (en) * | 2018-05-03 | 2018-11-06 | 国网江西省电力有限公司信息通信分公司 | A kind of electrical equipment fault prediction technique based on Weibull distribution and hidden Semi-Markov Process |
CN108490807A (en) * | 2018-05-09 | 2018-09-04 | 南京恩瑞特实业有限公司 | Train fault analogue system and test method |
Non-Patent Citations (5)
Title |
---|
R NAIR PRAVIN ETC.: "Performance evaluation of HMM and neural network in motorbike fault detection system", 《2011 INTERNATIONAL CONFERENCE ON RECENT TRENDS IN INFORMATION TECHNOLOGY (ICRTIT)》 * |
崔建国等: "LSSVM与HMM在航空发动机状态预测中的应用研究", 《计算机工程》 * |
张子璇: "太阳能发电多维随机过程动态模型研究", 《中国优秀硕士学位论文全文 工程科技Ⅱ辑》 * |
柳楠: "基于隐马尔可夫模型的航空机械系统故障诊断算法设计", 《现代工业经济和信息化》 * |
邹会荣: "基于TMS320F2812的数据采集及电力电缆故障识别", 《中国优秀硕士学位论文全文电子期刊网 工程科技Ⅱ辑》 * |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111259261A (en) * | 2020-01-02 | 2020-06-09 | 中国铁道科学研究院集团有限公司通信信号研究所 | High-speed rail driving network collaborative alarm optimization method based on state transition prediction |
CN111259261B (en) * | 2020-01-02 | 2023-09-26 | 中国铁道科学研究院集团有限公司通信信号研究所 | High-speed rail driving network collaborative alarm optimization method based on state transition prediction |
CN111223574A (en) * | 2020-01-14 | 2020-06-02 | 宁波市海洋与渔业研究院 | Penaeus vannamei boone enterohepatic sporulosis early warning method based on big data mining |
CN111565118B (en) * | 2020-04-17 | 2022-08-05 | 烽火通信科技股份有限公司 | Virtualized network element fault analysis method and system based on multi-observation dimension HMM |
CN111565118A (en) * | 2020-04-17 | 2020-08-21 | 烽火通信科技股份有限公司 | Virtualized network element fault analysis method and system based on multi-observation dimension HMM |
CN111882078A (en) * | 2020-06-28 | 2020-11-03 | 北京交通大学 | Method for optimizing state maintenance strategy of running part of rail transit train |
CN111882078B (en) * | 2020-06-28 | 2024-01-02 | 北京交通大学 | Method for optimizing state maintenance strategy of running part component of rail transit train |
CN112069045A (en) * | 2020-08-14 | 2020-12-11 | 西安理工大学 | Cloud platform software performance prediction method based on hidden Markov |
CN112257777B (en) * | 2020-10-21 | 2023-09-05 | 平安科技(深圳)有限公司 | Off-duty prediction method and related device based on hidden Markov model |
CN112257777A (en) * | 2020-10-21 | 2021-01-22 | 平安科技(深圳)有限公司 | Off-job prediction method based on hidden Markov model and related device |
CN113053536A (en) * | 2021-01-15 | 2021-06-29 | 中国人民解放军军事科学院军事医学研究院 | Infectious disease prediction method, system and medium based on hidden Markov model |
CN113053536B (en) * | 2021-01-15 | 2023-11-24 | 中国人民解放军军事科学院军事医学研究院 | Infectious disease prediction method, system and medium based on hidden Markov model |
CN113298240B (en) * | 2021-07-27 | 2021-11-05 | 北京科技大学 | Method and device for predicting life cycle of servo drive system |
CN113298240A (en) * | 2021-07-27 | 2021-08-24 | 北京科技大学 | Method and device for predicting life cycle of servo drive system |
CN116020879A (en) * | 2023-02-15 | 2023-04-28 | 北京科技大学 | Technological parameter-oriented strip steel hot continuous rolling space-time multi-scale process monitoring method and device |
CN117114352A (en) * | 2023-09-15 | 2023-11-24 | 北京阿帕科蓝科技有限公司 | Vehicle maintenance method, device, computer equipment and storage medium |
CN117114352B (en) * | 2023-09-15 | 2024-04-09 | 北京阿帕科蓝科技有限公司 | Vehicle maintenance method, device, computer equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN110288046B (en) | 2022-11-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110288046A (en) | A kind of failure prediction method based on wavelet neural network and Hidden Markov Model | |
Li et al. | A directed acyclic graph network combined with CNN and LSTM for remaining useful life prediction | |
Cheng et al. | Remaining useful life prognosis based on ensemble long short-term memory neural network | |
CN110188397B (en) | Model and method for predicting icing of overhead transmission line | |
Zhou et al. | New model for system behavior prediction based on belief rule based systems | |
Hu et al. | Deep bidirectional recurrent neural networks ensemble for remaining useful life prediction of aircraft engine | |
CN111539515A (en) | Complex equipment maintenance decision method based on fault prediction | |
CN113722985B (en) | Method and system for evaluating health state and predicting residual life of aero-engine | |
Kwon et al. | Accident identification in nuclear power plants using hidden Markov models | |
Wang et al. | A data-driven degradation prognostic strategy for aero-engine under various operational conditions | |
Remadna et al. | Leveraging the power of the combination of CNN and bi-directional LSTM networks for aircraft engine RUL estimation | |
CN112766603A (en) | Traffic flow prediction method, system, computer device and storage medium | |
CN110046663A (en) | A kind of complex electromechanical systems fault critical state discrimination method | |
Yao et al. | Model-based deep transfer learning method to fault detection and diagnosis in nuclear power plants | |
Back et al. | Prediction and uncertainty analysis of power peaking factor by cascaded fuzzy neural networks | |
Wu et al. | Ensemble recurrent neural network-based residual useful life prognostics of aircraft engines | |
Raptodimos et al. | An artificial neural network approach for predicting the performance of ship machinery equipment | |
Huang et al. | Research on module-level fault diagnosis of avionics system based on residual convolutional neural network | |
CN116415485A (en) | Multi-source domain migration learning residual service life prediction method based on dynamic distribution self-adaption | |
CN112560252B (en) | Method for predicting residual life of aeroengine | |
Pan et al. | Bearing condition prediction using enhanced online learning fuzzy neural networks | |
CN113837443A (en) | Transformer substation line load prediction method based on depth BilSTM | |
Xia et al. | Degradation prediction and rolling predictive maintenance policy for multi-sensor systems based on two-dimensional self-attention | |
Ge et al. | Remaining useful life prediction using deep multi-scale convolution neural networks | |
Wang et al. | An off-online fuzzy modelling method for fault prognosis with an application |
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 |