CN109615003A - A kind of power source failure prediction method based on ELM-CHMM - Google Patents
A kind of power source failure prediction method based on ELM-CHMM Download PDFInfo
- Publication number
- CN109615003A CN109615003A CN201811488243.9A CN201811488243A CN109615003A CN 109615003 A CN109615003 A CN 109615003A CN 201811488243 A CN201811488243 A CN 201811488243A CN 109615003 A CN109615003 A CN 109615003A
- Authority
- CN
- China
- Prior art keywords
- model
- elm
- voltage signal
- degraded
- failure
- 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
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/40—Testing power supplies
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
A kind of power source failure prediction method based on ELM-CHMM, the present invention relates to power source failure prediction methods.The purpose of the present invention is to solve the low problems of existing method failure predication accuracy.Process are as follows: voltage signal data is divided into trained and test data set, training dataset is pre-processed, the voltage signal matrix after being reconstructed；Establish ELM model；In test data set input ELM model, the voltage signal Jing Guo ELM model prediction is exported；Extract the characteristic parameter of training dataset；Establish CHMM State Forecasting Model；The characteristic parameter for extracting the voltage signal data Jing Guo ELM model prediction, is separately input in CHMM model；ELM-CHMM model is obtained, the state of power failure to be measured is obtained.The present invention is used for power source failure prediction field.
Description
Technical field
The present invention relates to the power source failure prediction methods based on ELM-CHMM.
Background technique
Ship'ssupply system is responsible for the electric energy supply of ship electrical equipment, ensure that various complex electronic equipments are just on ship
Often, on steady operation and ship staff life requirement, be ship " blood ".Therefore the stability of ship'ssupply system,
Reliability is most important, in order to provide stable power supply guarantee to electrical equipment on ship, needs to carry out the failure of power supply real-time
Accurately prediction.
At sea due to ship long-term work, the shadows such as adverse circumstances factor is such as made moist, burn into is vibrated, impacted be will receive
It rings, while also suffering from other equipment radiation interference, cause power failure possibility to greatly improve, seriously affect its performance
And the service life.Once ship'ssupply breaks down, it may result in part electrical equipment on ship and damage, even resulting in ship can not be just
Often work threatens ship's staff's safety.Therefore, method appropriate should be taken timely and effectively to find incipient fault, is convenient for work
Personnel take preventative maintenance in time, reduce loss.
In traditional all kinds of fittings, forecasting problem, widely applied method is that Single hidden layer feedforward neural networks utilize sample
Data are trained, and are fitted mapping function, the big and nonlinear data modeling problem suitable for data volume.Exist simultaneously many lack
Point, for example, it is computationally intensive, and training time length, conventional feed forward neural network are easily ensnared into local optimum, it is difficult to obtain the overall situation most
Excellent the problems such as solving, is sensitive to parameter selection, therefore it is not suitable for the higher failure predication scene of accuracy requirement.
In recent years, a kind of simple, effective Single hidden layer feedforward neural networks SLFNs learning algorithm --- extreme learning machine
(ELM) it is suggested and is used widely.Biological brain is analogous to according to different applications, the company of adjusting hidden layer to output layer
It fetches and realizes compression, feature learning, sparse coding, cluster, regression fit and classification.ELM algorithm is playing the single hidden layer of tradition
On the basis of feedforward neural network advantage, the bottleneck of traditional neural network is overcome, avoids the successive ignition of gradient descent method
It solves, effectively promotes calculating speed, Local Minimum problem is avoided to obtain globally optimal solution, realize as few as possible artificial do
In advance, higher accuracy is obtained.
Summary of the invention
The purpose of the present invention is to solve the low problems of existing method failure predication accuracy, and propose one kind and be based on
The power source failure prediction method of ELM-CHMM.
A kind of power source failure prediction method detailed process based on ELM-CHMM are as follows:
Step 1: acquisition ship'ssupply is slightly degenerated, the electricity of four kinds of states of gently degraded, heavy-degraded, complete failure
Signal data is pressed, the voltage signal data of each state is divided into training dataset and test data set, to training dataset
Voltage signal data pre-processed, the voltage signal matrix after being reconstructed
In formula, X, Y are the ship'ssupply output voltage signal matrix after reconstruct,For one-dimensional matrix [x1,x2...xm],For one-dimensional matrix [x2,x3...xm+1],For one-dimensional matrix [xN-m,xN-m+1...xN-1], xm+1Backward for m-th of time point
The supply voltage at 1 time point, xm+2For the supply voltage at m-th of time point, 2 time points backward, xNIt is defeated for n-th time point
Supply voltage out；
Step 2: the voltage signal matrix X after step 1 is reconstructed is as the input of ELM model, after step 1 is reconstructed
Output of the voltage signal matrix Y as ELM model；
With input and output voltage signal matrix, training ELM model obtains ELM model parameter, has established ELM model；
The corresponding voltage signal of test data set in step 1 is input in the ELM model established, ELM is passed through in output
The voltage signal of model prediction；
The ELM model is extreme learning machine model；
Step 3: being acquired using wavelet packet analysis to step 1 every according to the frequency domain characteristic of electric power output voltage signal
A kind of training dataset corresponding voltage signal data in state carry out it is decomposed and reconstituted, extract slight degeneration, gently degraded,
The characteristic parameter of four kinds of heavy-degraded, complete failure of-state voltages；
Step 4: the characteristic parameter of the four kinds of of-state voltages obtained according to step 3, training CHMM State Forecasting Model, obtain
It slightly degenerates to power supply, the model parameter under four kinds of gently degraded, heavy-degraded, complete failure states, completes four kinds of power supply shapes
The foundation of state CHMM State Forecasting Model；
The CHMM is continuous HMM；
Step 5: the voltage signal data using wavelet packet analysis to step 2 Jing Guo ELM model prediction carries out decomposing weight
Structure, extract slight degeneration, gently degraded, heavy-degraded, four kinds of of-state voltages of complete failure characteristic parameter, be separately input to
In CHMM State Forecasting Model under four kinds of power supply status；
If the consistent probability of power supply status where prediction result and test data set is greater than 85%, power source failure prediction
Accurately, ELM-CHMM power source failure prediction model is obtained, power failure voltage signal to be measured is inputted into ELM-CHMM power failure
Prediction model, obtains the state of power failure to be measured, and state includes for slight degeneration, gently degraded, heavy-degraded and event completely
Hinder four kinds of states, realizes the assessment to power failure degree；
If the consistent probability of power supply status where prediction result and test data set is less than or equal to 85%, power failure
Forecasting inaccuracy is true.
The invention has the benefit that
The present invention proposes a kind of power source failure prediction method based on ELM-CHMM.For power data, carried out using ELM
Data Tendency Forecast Based；For the data that prediction obtains, state is carried out using established continuous HMM (CHMM)
Assessment, judges the state of prediction data.This method combines that ELM arithmetic speed is fast, generalization ability is strong, can get the overall situation
The advantages that optimal solution and CHMM preferably classification capacity, corresponding relationship between observation data and hidden state can be established, can be carried out
The advantages that accurate log-likelihood probability calculation, improves power source failure prediction accuracy and speed, shortens predicted time, solves part
It is optimal, to parameter selection tender subject.Wherein, CHMM model is divided into minor failure state, moderate fault case, again by fault degree
Spend fault case, complete failure state.This makes it possible to the predictions completed to power down mode, and with according to sampling interval conversion energy
Enough provide the approximate time that different degrees of failure occurs in power supply.
In power source failure prediction algorithm proposed by the present invention, since the type of hidden layer activation primitive is pre- to ELM model
It surveys result and there is certain influence, it is relatively more preferable by comparing Sigmoid function effect, it can achieve faster speed and precision,
Therefore select Sigmoid function as the activation primitive of ELM model.ELM and BP neural network are instructed using training data
Practice, trained model is tested using test data, calculates root-mean-square error (RMSE) and predicted time, compare
To power source failure prediction precision improvement about 10.8%, calculating speed promotes about 78.8%, while by comparing initial data and in advance
Measured data, corresponding degenerate state is consistent, i.e., prediction result is consistent with legitimate reading.Therefore failure predication proposed by the present invention is calculated
Method can significantly improve the real-time and accuracy of power source failure prediction.
Detailed description of the invention
Fig. 1 is that ELM-CHMM of the present invention predicts block diagram.
Specific embodiment
Specific embodiment 1: embodiment is described with reference to Fig. 1, a kind of power supply based on ELM-CHMM of present embodiment
Failure prediction method detailed process are as follows:
Step 1: acquisition ship'ssupply is slightly degenerated, the electricity of four kinds of states of gently degraded, heavy-degraded, complete failure
Signal data is pressed, the voltage signal data of each state is divided into training dataset and test data set, to training dataset
Voltage signal data pre-processed, the voltage signal matrix after being reconstructed
In formula, X, Y are the ship'ssupply output voltage signal matrix after reconstruct,For one-dimensional matrix [x1,x2...xm],For one-dimensional matrix [x2,x3...xm+1],For one-dimensional matrix [xN-m,xN-m+1...xN-1], xm+1Backward for m-th of time point
The supply voltage at 1 time point, xm+2For the supply voltage at m-th of time point, 2 time points backward, xNIt is defeated for n-th time point
Supply voltage out；
Power supply slightly degenerate for power supply the loss of power be less than or equal to 20% when；
Power supply gently degraded is that the loss of power of power supply is greater than 20%, when being less than or equal to 30%；
Power supply heavy-degraded is that the loss of power of power supply is greater than 30%, when being less than or equal to 50%；
When power supply complete failure is that the loss of power of power supply is greater than 50%；
Step 2: the voltage signal matrix X after step 1 is reconstructed is as the input of ELM model, after step 1 is reconstructed
Output of the voltage signal matrix Y as ELM model；
With input and output voltage signal matrix, training ELM model obtains ELM model parameter, has established ELM model；
The corresponding voltage signal of test data set in step 1 is input in the ELM model established, ELM is passed through in output
The voltage signal of model prediction；
The ELM model is extreme learning machine model；
Step 3: being acquired using wavelet packet analysis to step 1 every according to the frequency domain characteristic of electric power output voltage signal
A kind of training dataset corresponding voltage signal data in state carry out it is decomposed and reconstituted, extract slight degeneration, gently degraded,
The characteristic parameter of four kinds of heavy-degraded, complete failure of-state voltages；
Step 4: the characteristic parameter of the four kinds of of-state voltages obtained according to step 3, training CHMM State Forecasting Model, obtain
It slightly degenerates to power supply, the model parameter under four kinds of gently degraded, heavy-degraded, complete failure states, completes four kinds of power supply shapes
The foundation of state CHMM State Forecasting Model；
The CHMM is continuous HMM；
Step 5: the voltage signal data using wavelet packet analysis to step 2 Jing Guo ELM model prediction carries out decomposing weight
Structure, extract slight degeneration, gently degraded, heavy-degraded, four kinds of of-state voltages of complete failure characteristic parameter, be separately input to
In CHMM State Forecasting Model under four kinds of power supply status；
If the consistent probability of power supply status where prediction result and test data set is greater than 85%, power source failure prediction
Accurately, ELM-CHMM power source failure prediction model is obtained, power failure voltage signal to be measured is inputted into ELM-CHMM power failure
Prediction model, obtains the state of power failure to be measured, and state includes for slight degeneration, gently degraded, heavy-degraded and event completely
Hinder four kinds of states, realizes the assessment to power failure degree；
If the consistent probability of power supply status where prediction result and test data set is less than or equal to 85%, power failure
Forecasting inaccuracy is true.
Specific embodiment 2: the present embodiment is different from the first embodiment in that, acquisition vessel in the step 1
It is slightly degenerated with power supply, the output voltage signal data of four kinds of states of gently degraded, heavy-degraded, complete failure, to output
Voltage signal data is pre-processed, output voltage signal matrix X, Y after being reconstructed；Detailed process are as follows:
When power supply generates degeneration or failure, output voltage signal can generate corresponding change.The present invention is peculiar to vessel by monitoring
Electric power output voltage realizes the prediction to its gradual failure.By minor failure state, moderate fault case, severe fault case, completely
The output voltage of fault case power supply is modeled by the training to extreme learning machine (ELM) model as initial data, predicts five kinds
The output voltage trend of power supply.
In order to preferably support ELM modeling and forecasting, original data sequence is carried out according to the prediction model structure of ELM mutually empty
Between reconstruct, obtain higher-dimension input data, excavate information content as big as possible to obtain the incidence relation between data；Specific mistake
Journey are as follows:
If the ship'ssupply of acquisition is slightly degenerated, the output of four kinds of states of gently degraded, heavy-degraded, complete failure electricity
Pressure signal sequence is XN={ x1,x2,…,xN}；
The sequential value x at a given one-dimensional time pointn, it is assumed that there are Nonlinear Mapping passes between preceding m sequential value
System:
xn=f (xn-m,...,xn-2,xn-1),n∈(1,...,N),m∈(0,...,N-1) (1)
In formula, xnFor the supply voltage (any one voltage value in section) of n-th of time point output, xn-mWhen being n-th
Between put the supply voltage at m time point forward, xn-2For the supply voltage at n-th of time point, 2 time points forward, xn-1It is n-th
A time point is the sequential value label at time point to previous supply voltage, n, meets n=N-m；N is the original number of output voltage
According to number, m is Embedded dimensions, and f () is Nonlinear Mapping model；
By the sequential value x at one-dimensional time pointnIt is reconstructed into following matrix form,
In formula, X, Y are the ship'ssupply output voltage signal matrix after reconstruct,For one-dimensional matrix [x1,x2...xm],For one-dimensional matrix [x2,x3...xm+1],For one-dimensional matrix [xN-m,xN-m+1...xN-1], xm+1Backward for m-th of time point
The supply voltage at 1 time point, xm+2For the supply voltage at m-th of time point, 2 time points backward, xNIt is defeated for n-th time point
Supply voltage out.
Other steps and parameter are same as the specific embodiment one.
Specific embodiment 3: the present embodiment is different from the first and the second embodiment in that, it will in the step 2
Input of the voltage signal matrix X as ELM model after step 1 reconstruct, the voltage signal matrix Y after step 1 is reconstructed make
For the output of ELM model；
With input and output voltage signal matrix, training ELM model obtains ELM model parameter, has established ELM model；
The corresponding voltage signal of test data set in step 1 is input in the ELM model established, ELM is passed through in output
The voltage signal of model prediction；
Detailed process are as follows:
For M group training sample (Xj,Yj), if ELM prototype network reality output is equal to desired output, activation primitive
Mathematical model expression-form g (x) it is as follows:
Wherein, XjAnd YjMatrix is output and input for any one group of ELM model,For ELM prototype network hidden layer node
Number, M are training sample group number, βiFor the weight for connecting i-th of hidden layer node, biIt is inclined for ELM prototype network hidden layer node
It sets, ωiFor the connection weight of ELM prototype network input layer and hidden layer；
Formula (3) can be write a Chinese character in simplified form into: H β=Y；
Wherein H is ELM prototype network hidden layer output matrix；
Hidden node parameter (ω is randomly generatedi,bi), hidden layer output matrix H is calculated, then calculate weight beta, β=H-1Y；
Using β as ELM model parameter, ELM model has been established；
The corresponding voltage signal of test data set in step 1 is input in the ELM model established, ELM is passed through in output
The voltage signal of model prediction.
Other steps and parameter are the same as one or two specific embodiments.
Specific embodiment 4: unlike one of present embodiment and specific embodiment one to three, the step 3
In
As power supply generates degeneration or failure, the corresponding change that output voltage signal generates can be embodied in Energy distribution.
The present invention realizes the feature extraction of electric power output voltage using wavelet packet analysis.First according to the frequency domain of electric power output voltage signal
Characteristic, the corresponding voltage signal data of training dataset in each state acquired using wavelet packet analysis to step 1 into
Row is decomposed and reconstituted, extract slight degeneration, gently degraded, heavy-degraded, four kinds of of-state voltages of complete failure characteristic parameter；
Detailed process are as follows:
The corresponding voltage signal data of training dataset in each state for enabling step 1 acquire is S, utilizes small echo
Packet reconstructs to obtain discrete point signal S at pth layer q node after decomposingpq(k), k=1,2 ..., L, L is after signal reconstructions
Points；
The then signal energy E each of after WAVELET PACKET DECOMPOSITION reconstructs in equidistant frequency rangepqWith gross energy E calculation expression
Formula are as follows:
The frequency domain character vector T of wavelet packet extraction is obtained,
The frequency domain character vector T that wavelet packet extracts is exactly that each frequency band energy decomposed accounts for the ratio of gross energy, is obtained slight
Degeneration, gently degraded, heavy-degraded, complete failure power supply training dataset voltage characteristic parameter.
Other steps and parameter are identical as one of specific embodiment one to three.
Specific embodiment 5: unlike one of present embodiment and specific embodiment one to four, the step 4
The characteristic parameter of the middle four kinds of of-state voltages obtained according to step 3, training CHMM status assessment model, obtains power supply and slightly moves back
Model parameter under four kinds of change, gently degraded, heavy-degraded, complete failure states is completed four kinds of power supply status CHMM states and is commented
Estimate the foundation of model；Detailed process are as follows:
Hidden Markov Model (HMM) is a kind of special markoff process, described by way of probabilistic model with
The statistical property of machine process.It is a dual random process, and hidden Markov chain and display with finite state number are random
Collection of functions, the former is used to describe the transfer of state, and the latter is used to describe the corresponding relationship between observed value and each state.
Hidden Markov Model has following five kinds of elements:
U, U are the state sum in Hidden Markov Model, i.e., slight degeneration, gently degraded, heavy-degraded, complete failure
Four kinds of states；
V, V are the observed value sum observed under each state in Hidden Markov Model, and observed value sum is step 3
The characteristic parameter number of four kinds of obtained of-state voltages；
π={ πξ, π is initial probability distribution over states, and the ξ shape probability of state is in when indicating to start, and ξ is slightly to move back
Change, gently degraded, heavy-degraded or complete failure；
A={ aξζ, A is the transfering probability distribution matrix of state, and expression state is transferred to the probability of ζ by ξ, and ζ is slightly to move back
Change, gently degraded, heavy-degraded or complete failure；
B={ bζ(r) }, B is observed value probability distribution matrix, and when indicating in state ζ, observation vector is the r articles general
Rate (observation vector inputted below have it is many it is said herein be the r articles, an observation when observation vector represents a kind of state to
Amount)；
Generally, a Hidden Markov Model is indicated with λ=(U, V, π, A, B), can simplify as λ=(π, A, B)；
The characteristic parameter for four kinds of of-state voltages that step 3 is obtained is as observation vector OIt is original；
By observation vector OIt is originalAs the input of Hidden Markov Model HMM, λ=(π, A, B) is used as Hidden Markov Model
The parameter of HMM is trained Hidden Markov Model HMM, respectively obtains slight degradation filture HMM1, gently degraded failure
The model parameter λ of HMM2, heavy-degraded failure HMM3, complete failure HMM41、λ2、λ3、λ4, complete four kinds of power supply status CHMM shapes
The foundation of state assessment models.
Other steps and parameter are identical as one of specific embodiment one to four.
Specific embodiment 6: unlike one of present embodiment and specific embodiment one to five, the step 5
In
Voltage signal data using wavelet packet analysis to step 2 Jing Guo ELM model prediction carries out decomposed and reconstituted, extraction
To slight degeneration, gently degraded, heavy-degraded, four kinds of of-state voltages of complete failure characteristic parameter, be separately input to four kinds of electricity
In CHMM State Forecasting Model under the state of source；
If the consistent probability of power supply status where prediction result and test data set is greater than 85%, power source failure prediction
Accurately, ELM-CHMM power source failure prediction model is obtained, power failure voltage signal to be measured is inputted into ELM-CHMM power failure
Prediction model, obtains the state of power failure to be measured, and state includes for slight degeneration, gently degraded, heavy-degraded and event completely
Hinder four kinds of states, realizes the assessment to power failure degree；
If the consistent probability of power supply status where prediction result and test data set is less than or equal to 85%, power failure
Forecasting inaccuracy is true.
Detailed process are as follows:
Voltage signal data using wavelet packet analysis to step 2 Jing Guo ELM model prediction carries out decomposed and reconstituted, extraction
To slight degeneration, gently degraded, heavy-degraded, four kinds of of-state voltages of complete failure characteristic parameter, the feature of each state
Parameter passes through each group respectively with there are four types of model parameter λ as one group1、λ2、λ3、λ4HMM1, HMM2, HMM3, HMM4
Model makes each group to obtain four log-likelihood probability, respectively P (O using the Forward-backward algorithm of CHMMPrediction|λ1)、
P(OPrediction|λ2)、P(OPrediction|λ3)、P(OPrediction|λ4)；
The corresponding Status Type of each group of maximum log-likelihood probability value is power source failure prediction as a result, if prediction knot
The consistent probability of power supply status where fruit and test data set is greater than 85%, then power source failure prediction is accurate, obtains ELM-
Power failure voltage signal to be measured is inputted the power source failure prediction model of ELM-CHMM by the power source failure prediction model of CHMM,
The state of power failure to be measured is obtained, state includes for slight four kinds of degeneration, gently degraded, heavy-degraded and complete failure shapes
State realizes the assessment to power failure degree；
If the consistent probability of power supply status where prediction result and test data set is less than or equal to 85%, power failure
Forecasting inaccuracy is true.
Other steps and parameter are identical as one of specific embodiment one to five.
Beneficial effects of the present invention are verified using following embodiment:
Embodiment one:
The present embodiment is specifically to be prepared according to the following steps:
ELM model training
CHMM model training
The present invention can also have other various embodiments, without deviating from the spirit and substance of the present invention, this field
Technical staff makes various corresponding changes and modifications in accordance with the present invention, but these corresponding changes and modifications all should belong to
The protection scope of the appended claims of the present invention.
Claims (6)
1. a kind of power source failure prediction method based on ELM-CHMM, it is characterised in that: the method detailed process are as follows:
Step 1: acquisition ship'ssupply is slightly degenerated, the voltage of four kinds of states of gently degraded, heavy-degraded, complete failure letter
The voltage signal data of each state is divided into training dataset and test data set, to the electricity of training dataset by number
Pressure signal data is pre-processed, the voltage signal matrix after being reconstructed
In formula, X, Y are the ship'ssupply output voltage signal matrix after reconstruct,For one-dimensional matrix [x1,x2...xm],
For one-dimensional matrix [x2,x3...xm+1],For one-dimensional matrix [xN-m,xN-m+1...xN-1], xm+1It is m-th of time point 1 backward
The supply voltage at time point, xm+2For the supply voltage at m-th of time point, 2 time points backward, xNFor the output of n-th time point
Supply voltage；
Step 2: electricity of the voltage signal matrix X as the input of ELM model, after step 1 is reconstructed after step 1 is reconstructed
Press output of the signal matrix Y as ELM model；
With input and output voltage signal matrix training ELM model, ELM model parameter is obtained, ELM model has been established；
The ELM model is extreme learning machine model；
The corresponding voltage signal of test data set in step 1 is input in the ELM model established, ELM model is passed through in output
The voltage signal of prediction；
Step 3: according to the frequency domain characteristic of electric power output voltage signal, step 1 is acquired using wavelet packet analysis each
The corresponding voltage signal data progress of training dataset in state is decomposed and reconstituted, extracts slight degeneration, gently degraded, severe
It degenerates, the characteristic parameter of four kinds of of-state voltages of complete failure；
Step 4: the characteristic parameter of the four kinds of of-state voltages obtained according to step 3, training CHMM State Forecasting Model, obtain electricity
Source slightly degenerates, the model parameter under four kinds of gently degraded, heavy-degraded, complete failure states, completes four kinds of power supply status
The foundation of CHMM State Forecasting Model；
The CHMM is continuous HMM；
Step 5: the voltage signal data progress using wavelet packet analysis to step 2 Jing Guo ELM model prediction is decomposed and reconstituted, mention
Get slight degeneration, gently degraded, heavy-degraded, four kinds of of-state voltages of complete failure characteristic parameter, be separately input to four kinds
In CHMM State Forecasting Model under power supply status；
If the consistent probability of power supply status where prediction result and test data set is greater than 85%, power source failure prediction is quasi-
Really, ELM-CHMM power source failure prediction model is obtained, power failure voltage signal to be measured input ELM-CHMM power failure is pre-
Model is surveyed, the state of power failure to be measured is obtained, state includes for slight degeneration, gently degraded, heavy-degraded and complete failure
Four kinds of states realize the assessment to power failure degree；
If the consistent probability of power supply status where prediction result and test data set is less than or equal to 85%, power source failure prediction
Inaccuracy.
2. a kind of power source failure prediction method based on ELM-CHMM according to claim 1, it is characterised in that: the step
In one acquire ship'ssupply slightly degenerate, the output voltage signal number of four kinds of states of gently degraded, heavy-degraded, complete failure
According to being pre-processed to output voltage signal data, output voltage signal matrix X, Y after being reconstructed；Detailed process are as follows:
If the ship'ssupply of acquisition is slightly degenerated, the output voltage of four kinds of states of gently degraded, heavy-degraded, complete failure letter
Number sequence is XN={ x1,x2,…,xN}；
The sequential value x at a given one-dimensional time pointn, it is assumed that there are Nonlinear Mapping relationships between preceding m sequential value:
xn=f (xn-m,...,xn-2,xn-1),n∈(1,...,N),m∈(0,...,N-1) (1)
In formula, xnFor the supply voltage of n-th of time point output, xn-mFor the power supply electricity at m time point forward n-th of time point
Pressure, xn-2For the supply voltage at n-th of time point, 2 time points forward, xn-1It is n-th of time point to previous supply voltage, n
For the sequential value label at time point, meet n=N-m；N is the initial data number of output voltage, and m is Embedded dimensions, and f () is
Nonlinear Mapping model；
By the sequential value x at one-dimensional time pointnIt is reconstructed into following matrix form,
In formula, X, Y are the ship'ssupply output voltage signal matrix after reconstruct,For one-dimensional matrix [x1,x2…xm],For
One-dimensional matrix [x2,x3…xm+1],For one-dimensional matrix [xN-m,xN-m+1…xN-1], xm+1For m-th of time point, 1 time backward
The supply voltage of point, xm+2For the supply voltage at m-th of time point, 2 time points backward, xNFor the electricity of n-th time point output
Source voltage.
3. a kind of power source failure prediction method based on ELM-CHMM according to claim 1 or claim 2, it is characterised in that: described
Voltage letter of the voltage signal matrix X after reconstructing step 1 in step 2 as the input of ELM model, after step 1 is reconstructed
Number output of the matrix Y as ELM model；
With input and output voltage signal matrix, training ELM model obtains ELM model parameter, has established ELM model；
The corresponding voltage signal of test data set in step 1 is input in the ELM model established, ELM model is passed through in output
The voltage signal of prediction；
Detailed process are as follows:
For M group training sample (Xj,Yj), if ELM prototype network reality output is equal to desired output, the number of activation primitive
It is as follows to learn model tormulation form g (x):
Wherein, XjAnd YjMatrix is output and input for any one group of ELM model,For ELM prototype network node in hidden layer, M is
Training sample group number, βiFor the weight for connecting i-th of hidden layer node, biFor the biasing of ELM prototype network hidden layer node, ωiFor
The connection weight of ELM prototype network input layer and hidden layer；
Formula (3) is write a Chinese character in simplified form into: H β=Y；
Wherein H is ELM prototype network hidden layer output matrix；
Hidden node parameter (ω is randomly generatedi,bi), hidden layer output matrix H is calculated, then calculate weight beta, β=H-1Y；β is made
For ELM model parameter, ELM model has been established；
The corresponding voltage signal of test data set in step 1 is input in the ELM model established, ELM model is passed through in output
The voltage signal of prediction.
4. a kind of power source failure prediction method based on ELM-CHMM according to claim 3, it is characterised in that: the step
According to the frequency domain characteristic of electric power output voltage signal in three, in each state acquired using wavelet packet analysis to step 1
The corresponding voltage signal data of training dataset carries out decomposed and reconstituted, extracts slight degeneration, gently degraded, heavy-degraded, complete
The characteristic parameter of four kinds of of-state voltages of total failure；
Detailed process are as follows:
The corresponding voltage signal data of training dataset in each state for enabling step 1 acquire is S, utilizes wavelet packet point
It reconstructs to obtain discrete point signal S at pth layer q node after solutionpq(k), k=1,2 ..., L, L are the points after signal reconstruction；
The then signal energy E each of after WAVELET PACKET DECOMPOSITION reconstructs in equidistant frequency rangepqWith gross energy E calculation expression
Are as follows:
The frequency domain character vector T of wavelet packet extraction is obtained,
The frequency domain character vector T that wavelet packet extracts is exactly that each frequency band energy decomposed accounts for the ratio of gross energy, slightly degenerated,
Gently degraded, heavy-degraded, complete failure power supply training dataset voltage characteristic parameter.
5. a kind of power source failure prediction method based on ELM-CHMM according to claim 4, it is characterised in that: the step
The characteristic parameter of the four kinds of of-state voltages obtained in four according to step 3, training CHMM status assessment model, it is slight to obtain power supply
Model parameter under four kinds of degeneration, gently degraded, heavy-degraded, complete failure states, completes four kinds of power supply status CHMM states
The foundation of assessment models；Detailed process are as follows:
Hidden Markov Model has following five kinds of elements:
U, U are the state sum in Hidden Markov Model, i.e., slight degeneration, gently degraded, heavy-degraded, four kinds of complete failure
State；
V, V are the observed value sum observed under each state in Hidden Markov Model, and observed value sum obtains for step 3
Four kinds of of-state voltages characteristic parameter number；
π={ πξ, π is initial probability distribution over states, be in the ξ shape probability of state when indicating to start, ξ be slight degeneration, in
Degree degeneration, heavy-degraded or complete failure；
A={ aξζ, A is the transfering probability distribution matrix of state, and expression state is transferred to the probability of ζ by ξ, ζ be it is slight degenerate, in
Degree degeneration, heavy-degraded or complete failure；
B={ bζ(r) }, B is observed value probability distribution matrix, and when indicating in state ζ, observation vector is the r articles of probability；
A Hidden Markov Model is indicated with λ=(U, V, π, A, B), is reduced to λ=(π, A, B)；
The characteristic parameter for four kinds of of-state voltages that step 3 is obtained is as observation vector OIt is original；
By observation vector OIt is originalAs the input of Hidden Markov Model HMM, λ=(π, A, B) is used as Hidden Markov Model HMM
Parameter, Hidden Markov Model HMM is trained, respectively obtain slight degradation filture HMM1, gently degraded failure HMM2,
The model parameter λ of heavy-degraded failure HMM3, complete failure HMM41、λ2、λ3、λ4, complete four kinds of power supply status CHMM states and comment
Estimate the foundation of model.
6. a kind of power source failure prediction method based on ELM-CHMM according to claim 5, it is characterised in that: the step
Five detailed processes are as follows:
Voltage signal data progress using wavelet packet analysis to step 2 Jing Guo ELM model prediction is decomposed and reconstituted, extracts light
Spend degeneration, gently degraded, heavy-degraded, four kinds of of-state voltages of complete failure characteristic parameter, the characteristic parameter of each state
As one group, passing through band respectively for each group, there are four types of model parameter λ1、λ2、λ3、λ4HMM1, HMM2, HMM3, HMM4 model,
Each group is made to obtain four log-likelihood probability, respectively P (O using the Forward-backward algorithm of CHMMPrediction|λ1)、P(OPrediction
|λ2)、P(OPrediction|λ3)、P(OPrediction|λ4)；
The corresponding Status Type of each group of maximum log-likelihood probability value be power source failure prediction as a result, if prediction result with
The consistent probability of power supply status where test data set is greater than 85%, then power source failure prediction is accurate, obtains ELM-CHMM's
Power source failure prediction model, by power failure voltage signal to be measured input ELM-CHMM power source failure prediction model, obtain to
The state of power failure is surveyed, state includes for slight four kinds of degeneration, gently degraded, heavy-degraded and complete failure states, realization
Assessment to power failure degree；
If the consistent probability of power supply status where prediction result and test data set is less than or equal to 85%, power source failure prediction
Inaccuracy.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811488243.9A CN109615003B (en) | 2018-12-06 | 2018-12-06 | Power failure prediction method based on ELM-CHMM |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811488243.9A CN109615003B (en) | 2018-12-06 | 2018-12-06 | Power failure prediction method based on ELM-CHMM |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109615003A true CN109615003A (en) | 2019-04-12 |
| CN109615003B CN109615003B (en) | 2022-11-08 |
Family
ID=66006652
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811488243.9A Active CN109615003B (en) | 2018-12-06 | 2018-12-06 | Power failure prediction method based on ELM-CHMM |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109615003B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112816898A (en) * | 2021-01-26 | 2021-05-18 | 三一重工股份有限公司 | Battery failure prediction method and device, electronic equipment and storage medium |
| CN117214766A (en) * | 2023-11-09 | 2023-12-12 | 深圳市蓝之洋科技有限公司 | Mobile power supply fault detection method, device and equipment |
Citations (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030184307A1 (en) * | 2002-02-19 | 2003-10-02 | Kozlowski James D. | Model-based predictive diagnostic tool for primary and secondary batteries |
| CN102606557A (en) * | 2012-01-16 | 2012-07-25 | 北京航空航天大学 | Health evaluation method for hydraulic system based on fault observer and SOM (self-organized mapping) |
| CN102830341A (en) * | 2012-08-28 | 2012-12-19 | 南京航空航天大学 | Online intelligent fault prediction method for power electronic circuit based on RS-CMAC (rough sets and cerebellar model articulation controller) |
| 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 |
| CN103176128A (en) * | 2013-03-28 | 2013-06-26 | 华南理工大学 | Method and system for forcasting state of wind generating set and diagnosing intelligent fault |
| CN103995237A (en) * | 2014-05-09 | 2014-08-20 | 南京航空航天大学 | Satellite power supply system online fault diagnosis method |
| CN104463347A (en) * | 2014-11-06 | 2015-03-25 | 电子科技大学 | Method for electronic product degenerate state trend prediction with singular signals |
| CN104807639A (en) * | 2015-04-23 | 2015-07-29 | 广西大学 | Fault diagnosis method and device for rolling bearing of running gear of locomotive |
| CN105095918A (en) * | 2015-09-07 | 2015-11-25 | 上海交通大学 | Multi-robot system fault diagnosis method |
| CN105334472A (en) * | 2015-10-26 | 2016-02-17 | 安徽理工大学 | Online remaining life prediction method for mining intrinsic safety power supply |
| CN105975749A (en) * | 2016-04-28 | 2016-09-28 | 上海交通大学 | Bearing health assessment and prediction method and system |
| US20170091641A1 (en) * | 2015-09-30 | 2017-03-30 | Tata Consultancy Services Limited | Methods and systems for optimizing hidden markov model based land change prediction |
| CN107169243A (en) * | 2017-06-27 | 2017-09-15 | 电子科技大学 | A kind of fuel cell Forecasting Methodology based on HSMM and empirical model |
| CN107202027A (en) * | 2017-05-24 | 2017-09-26 | 重庆大学 | A kind of large fan operation trend analysis and failure prediction method |
| 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 |
| CN108168924A (en) * | 2017-12-20 | 2018-06-15 | 东北石油大学 | A kind of reciprocating compressor life-span prediction method based on VMD and MFSS models |
| CN108256173A (en) * | 2017-12-27 | 2018-07-06 | 南京航空航天大学 | A kind of Gas path fault diagnosis method and system of aero-engine dynamic process |
| 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 |
| CN108892014A (en) * | 2018-09-19 | 2018-11-27 | 歌拉瑞电梯股份有限公司 | A kind of elevator internal contracting brake fault early warning method based on Elman neural network |
-
2018
- 2018-12-06 CN CN201811488243.9A patent/CN109615003B/en active Active
Patent Citations (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030184307A1 (en) * | 2002-02-19 | 2003-10-02 | Kozlowski James D. | Model-based predictive diagnostic tool for primary and secondary batteries |
| CN102606557A (en) * | 2012-01-16 | 2012-07-25 | 北京航空航天大学 | Health evaluation method for hydraulic system based on fault observer and SOM (self-organized mapping) |
| CN102830341A (en) * | 2012-08-28 | 2012-12-19 | 南京航空航天大学 | Online intelligent fault prediction method for power electronic circuit based on RS-CMAC (rough sets and cerebellar model articulation controller) |
| 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 |
| CN103176128A (en) * | 2013-03-28 | 2013-06-26 | 华南理工大学 | Method and system for forcasting state of wind generating set and diagnosing intelligent fault |
| CN103995237A (en) * | 2014-05-09 | 2014-08-20 | 南京航空航天大学 | Satellite power supply system online fault diagnosis method |
| CN104463347A (en) * | 2014-11-06 | 2015-03-25 | 电子科技大学 | Method for electronic product degenerate state trend prediction with singular signals |
| CN104807639A (en) * | 2015-04-23 | 2015-07-29 | 广西大学 | Fault diagnosis method and device for rolling bearing of running gear of locomotive |
| CN105095918A (en) * | 2015-09-07 | 2015-11-25 | 上海交通大学 | Multi-robot system fault diagnosis method |
| US20170091641A1 (en) * | 2015-09-30 | 2017-03-30 | Tata Consultancy Services Limited | Methods and systems for optimizing hidden markov model based land change prediction |
| CN105334472A (en) * | 2015-10-26 | 2016-02-17 | 安徽理工大学 | Online remaining life prediction method for mining intrinsic safety power supply |
| CN105975749A (en) * | 2016-04-28 | 2016-09-28 | 上海交通大学 | Bearing health assessment and prediction method and system |
| CN107202027A (en) * | 2017-05-24 | 2017-09-26 | 重庆大学 | A kind of large fan operation trend analysis and failure prediction method |
| CN107169243A (en) * | 2017-06-27 | 2017-09-15 | 电子科技大学 | A kind of fuel cell Forecasting Methodology based on HSMM and empirical model |
| 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 |
| CN108168924A (en) * | 2017-12-20 | 2018-06-15 | 东北石油大学 | A kind of reciprocating compressor life-span prediction method based on VMD and MFSS models |
| CN108256173A (en) * | 2017-12-27 | 2018-07-06 | 南京航空航天大学 | A kind of Gas path fault diagnosis method and system of aero-engine dynamic process |
| 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 |
| CN108892014A (en) * | 2018-09-19 | 2018-11-27 | 歌拉瑞电梯股份有限公司 | A kind of elevator internal contracting brake fault early warning method based on Elman neural network |
Non-Patent Citations (5)
| Title |
|---|
| KAMRANJAVED等: "SW-ELM:A summation wavelet extreme learning machine algorithm with a priori parameter initialization", 《NEUROCOMPUTING》 * |
| S. LADDADA等: "A data-driven prognostic approach based on wavelet transform and extreme learning machine", 《THE 5TH INTERNATIONAL CONFERENCE ON ELECTRICAL ENGINEERING – BOUMERDES (ICEE-B)》 * |
| 成杰: "基于ELM与HMM的序列飞机目标识别算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 林越等: "基于AP-HMM混合模型的充电桩故障诊断", 《广西师范大学学报（自然科学版）》 * |
| 许奇歆: "基于马尔科夫链的船舶输电系统短路故障预测方法研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112816898A (en) * | 2021-01-26 | 2021-05-18 | 三一重工股份有限公司 | Battery failure prediction method and device, electronic equipment and storage medium |
| CN112816898B (en) * | 2021-01-26 | 2022-03-01 | 三一重工股份有限公司 | Battery failure prediction method and device, electronic equipment and storage medium |
| CN117214766A (en) * | 2023-11-09 | 2023-12-12 | 深圳市蓝之洋科技有限公司 | Mobile power supply fault detection method, device and equipment |
| CN117214766B (en) * | 2023-11-09 | 2024-02-09 | 深圳市蓝之洋科技有限公司 | Mobile power supply fault detection method, device and equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109615003B (en) | 2022-11-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110020623B (en) | Human body activity recognition system and method based on conditional variation self-encoder | |
| Alvarez et al. | Learning the number of neurons in deep networks | |
| CN113283175B (en) | Photovoltaic power station joint fault diagnosis method based on asynchronous decentralized federal learning | |
| CN110110887A (en) | To the prediction technique of low-voltage platform area line loss per unit | |
| Yuan et al. | Learning latent interactions for event classification via graph neural networks and PMU data | |
| CN109376763A (en) | Sample classification method, system and medium based on multisample reasoning neural network | |
| CN114355240A (en) | Power distribution network ground fault diagnosis method and device | |
| CN112163106B (en) | Second-order similar-perceived image hash code extraction model establishment method and application thereof | |
| CN109615003A (en) | A kind of power source failure prediction method based on ELM-CHMM | |
| Ren et al. | DLformer: A dynamic length transformer-based network for efficient feature representation in remaining useful life prediction | |
| CN116021981A (en) | Method, device, equipment and storage medium for predicting ice coating faults of power distribution network line | |
| Manikandan et al. | Deepq: Residue analysis of localization images in large scale solid state physical environments | |
| de Miranda et al. | Recurrent neural network based on statistical recurrent unit for remaining useful life estimation | |
| Zhang et al. | Collaborative Forecasting and Analysis of Fish Catch in Hokkaido From Multiple Scales by Using Neural Network and ARIMA Model | |
| CN116007937A (en) | Intelligent fault diagnosis method and device for mechanical equipment transmission part | |
| Miao | Emotion Analysis and Opinion Monitoring of Social Network Users Under Deep Convolutional Neural Network | |
| CN115965150A (en) | Power load prediction method based on weight distribution | |
| Tong et al. | A prediction model for complex equipment remaining useful life using gated recurrent unit complex networks | |
| Zhang et al. | A Hybrid Neural Network-Based Intelligent Forecasting Approach for Capacity of Photovoltaic Electricity Generation | |
| CN114048849A (en) | Unsupervised representation learning method and device supporting time sequence social network diagram | |
| Zhang | Deep learning-based hybrid short-term solar forecast using sky images and meteorological data | |
| Wang et al. | Multivariate Time Series Imputation Based on Masked Autoencoding with Transformer | |
| CN115114345B (en) | Feature representation extraction method, device, equipment, storage medium and program product | |
| CN114154679B (en) | Spark-based PCFOA-KELM wind power prediction method and device | |
| CN116882461B (en) | Neural network evaluation optimization method and system based on neuron plasticity |
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 |