CN112036082B - Time series data prediction method based on attention mechanism - Google Patents

Time series data prediction method based on attention mechanism Download PDF

Info

Publication number
CN112036082B
CN112036082B CN202010875958.0A CN202010875958A CN112036082B CN 112036082 B CN112036082 B CN 112036082B CN 202010875958 A CN202010875958 A CN 202010875958A CN 112036082 B CN112036082 B CN 112036082B
Authority
CN
China
Prior art keywords
alpha
data
prediction
beta
attention
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202010875958.0A
Other languages
Chinese (zh)
Other versions
CN112036082A (en
Inventor
李国瑞
武雅君
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Northeastern University Qinhuangdao Branch
Original Assignee
Northeastern University Qinhuangdao Branch
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Northeastern University Qinhuangdao Branch filed Critical Northeastern University Qinhuangdao Branch
Priority to CN202010875958.0A priority Critical patent/CN112036082B/en
Publication of CN112036082A publication Critical patent/CN112036082A/en
Application granted granted Critical
Publication of CN112036082B publication Critical patent/CN112036082B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00—Testing of machine parts
    • G01M13/04—Bearings
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/20—Administration of product repair or maintenance

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Business, Economics & Management (AREA)
  • Quality & Reliability (AREA)
  • Operations Research (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Marketing (AREA)
  • Medical Informatics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Tourism & Hospitality (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Data Mining & Analysis (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Complex Calculations (AREA)

Abstract

The invention provides a time series data prediction method based on an attention mechanism, and relates to the technical field of data prediction. The present invention takes into account the inherent uncertainty that arises over the course of time and quantifies it as a time-varying variable. A prediction model is established for time sequence data, attention distribution is added when parameters are updated to score the weight, and prediction accuracy is effectively improved. And judging whether the machine can normally operate by judging whether the predicted value reaches a fault threshold value, if the predicted data exceeds the fault threshold value, the machine possibly breaks down and needs to be maintained, and if the predicted value is smaller than the fault threshold value, the machine can normally operate. Compared with the prior art, the method can predict future data more accurately by adding attention distribution. Failure warning is provided in advance and maintenance planning is improved, thereby reducing the cost of accidental maintenance and improving the reliability, availability and safety of the machine.

Description

Time series data prediction method based on attention mechanism
Technical Field
The invention relates to the technical field of data prediction, in particular to a time series data prediction method based on an attention mechanism.
Background
Time series prediction analysis uses the characteristics of an event in the past period of time to predict the characteristics of the event in the future period of time. The time series prediction technology has important significance in the fields of economy, engineering, natural science and technology and the like, such as market demand prediction, power generation prediction, stock tendency prediction, regional precipitation prediction and the like.
The relevance vector machine RVM (Relevance vector machine) is a sparse probability model based on a Bayesian training framework, can process the problems of high dimension, nonlinearity, small samples and the like, and has good sparsity, generalization capability and high algorithm precision. And establishing an attention-based RVM model for prediction, wherein the model has higher prediction precision than the existing method.
In the operation process of the equipment, due to the reasons of part aging, improper operation and the like, different types of faults of the system often occur, and if effective measures are not taken timely, the whole equipment can be shut down, and even serious disasters can be caused. To avoid this, prediction of the remaining useful life of the machine has become a key technology that can provide early warning of faults and improve maintenance planning, thereby reducing the cost of accidental maintenance and increasing the reliability, availability and safety of the machine. However, the prediction is often affected by uncertainties, such as system input uncertainties, sensor measurement uncertainties, operating environment uncertainties of use conditions, and modeling uncertainties of degradation models, and how to express and quantify these uncertainties is important to improve the effectiveness and accuracy of the prediction.
From the integration of the currently mainstream technology and application research work, the method for predicting the residual life of the machine mainly comprises the following steps:
1) model-based fault prediction techniques;
2) data-driven based failure prediction techniques;
3) statistical reliability based failure prediction techniques.
The model-based fault prediction method assumes that an accurate mathematical model of the subject system can be obtained. It is often difficult to build accurate mathematical models for complex dynamic systems. Therefore, the practical application and effectiveness of the model-based fault prediction technique is greatly limited. The fault prediction technology based on data is based on the collected data, the implicit information in the data is mined through various data analysis and processing methods for prediction operation, the relevance among the data is not considered, a considerable amount of fault data is needed to train a prediction model, the cost is overhigh, and the prediction accuracy is low.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a time series data prediction method based on an attention mechanism, which considers the inherent uncertainty caused by the process along with the time and quantizes the inherent uncertainty into a time-varying variable. And a prediction model is established for the time sequence data, and attention distribution is added to score the weight when the parameters are updated, so that the prediction accuracy is effectively improved. And judging whether the machine can normally operate by judging whether the predicted value reaches a fault threshold value, if the predicted data exceeds the fault threshold value, the machine possibly breaks down and needs to be maintained, and if the predicted value is smaller than the fault threshold value, the machine can normally operate.
The technical scheme adopted by the invention is as follows:
a time series data prediction method based on an attention mechanism comprises the following steps:
step 1: collecting time series degradation data observations, { x1,…,xNIs { t }1,…,tNThe observation value set of the moment, wherein N is the number of the observation values;
step 2, based on the time series degradation data set X ═ X1,…,xN]TEstablishment of RVM regression prediction model, xn+1＝wTφ(xn) + ε, wherein xnN is the degraded data observed at time N, d, …, N; d is the embedding dimension; w ═ w0,…,wd]TAs a model weight parameter, [ phi ] - [ phi ]d(xd)T,…,φN(xN)T]TIs a kernel function matrix, phin(xn)＝[1,k(xn-d+1,xn),…,k(xn,xn)]K (x, x) is a kernel function, ε -N (0, β)-1) Random errors that follow a normal distribution; beta is a-1Is the variance of epsilon;
step 3, aim atModel weight parameter w introduces prior parameter alpha ═ alpha0,…,αd]TInstant command
Figure BDA0002652639800000021
Training the model by using a time sequence data set X, and iteratively solving parameters alpha and beta, namely commanding
Figure BDA0002652639800000022
Wherein m is the posterior mean value of w, miAn i-th element being m; a posterior variance matrix of w sigmaiiFor the i row and i column elements of the matrix Σ, an attention weight c is added to the weight parameter w [ c ] based on the correlation between the time-series data when iteratively solving0,…,cd]TK represents the number of iterations; alpha is alpha-1Is a prior variance; p (w)i) Is wiOf the probability density function, wiIs w element i; alpha is alphaiIs alpha element i; t ═ xd+1,…,xN]T；
Step 3.1, data initialization is carried out: setting prior parameter alpha as d +1 dimension random vector, setting parameter beta as random number between 0 and 1, setting maximum iteration number kthInitializing k to 1;
step 3.2, iteration updating:
step 3.2.1, updating the posterior variance matrix sigma of w: sigma ═ Z + β ΦTΦ)-1；
Step 3.2.2, updating the posterior mean m of w: beta-sigma phiTt；
Step 3.2.3, updating alpha:
Figure BDA0002652639800000031
wherein alpha isiIs α, i is 0, …, d;
step 3.2.4, updating beta:
Figure BDA0002652639800000032
step 3.2.5, let k equal to k +1, if the iteration number k is greater than or equal to kthOr | | | αk+1-αkIf the value of | is converged, the iteration is finished and alpha and beta are output-1(ii) a Otherwise, skipping to 3.2.1;
the weight parameter w adds an attention weight c ═ c0,…,cd]T：
Figure BDA0002652639800000033
Wherein,
Figure BDA0002652639800000034
p (w | c α) is w prior distribution.
The attention weight c: c. Cj＝f(s([xj,…,xN-d-1+j]T,t))；
Wherein j is 1, …, d, c is [ c ]0,…,cd]T,c0＝1；t＝[xd+1,…,xN]T(ii) a s (a, b) is an attention metric, representing the similarity of a and b; f(s) is an attention weight assignment function.
The iterative solution is an edge likelihood function:
Figure BDA0002652639800000035
let the derivative of the edge likelihood function equal 0;
Figure BDA0002652639800000036
obtaining:
Figure BDA0002652639800000037
wherein, Σ ═ Z + β ΦTΦ)-1，m＝βΣΦTt; since the weight w a posteriori p (w | t, α, β) is N (w | m, Σ), Σ is (Z + β Φ)TΦ)-1For posterior variance, m ═ beta Σ ΦTt is the posterior mean value;
step 4, predicting the degradation prediction data x at the time of N +1 according to the RVM regression prediction modelN+1＝mTφN(xN)；
Step 5, judging xN+1Judging whether the machine can normally run or not by judging whether the fault threshold is reached or not, and if x is reachedN+1And if the fault threshold value is exceeded, the machine needs to be maintained, and if the fault threshold value is smaller than the fault threshold value, the machine normally operates.
Adopt the produced beneficial effect of above-mentioned technical scheme to lie in:
compared with the prior art, the time series data prediction method based on the attention mechanism can predict future data more accurately by adding the attention distribution. Failure warning is provided in advance and maintenance planning is improved, thereby reducing the cost of accidental maintenance and improving the reliability, availability and safety of the machine.
Drawings
FIG. 1 is a flow chart of the present invention for predicting time series data based on attention mechanism;
FIG. 2 is a diagram of an example of a method for predicting time series data based on an attention mechanism according to an embodiment of the present invention.
Detailed Description
The following detailed description of embodiments of the invention refers to the accompanying drawings.
A method for predicting time series data based on attention mechanism, as shown in fig. 1, comprising the following steps:
step 1: collecting an accelerated life degradation test data set of the rolling bearing, wherein { x1,…,xNIs { t }1,…,tNObserved value at time;
in this embodiment, the rolling bearing model is ldkuuer 204, and the sample size N is 316.
Step 2, based on the rolling bearing time sequence data set X ═ X1,…,xN]TEstablishment of RVM regression prediction model xn+1＝wTφ(xn) + ε, wherein xnN is d, …, N; d is the embedding dimension of 200; w ═ w0,…,wd]TAs a model weight parameter, [ phi ] - [ phi ]d(xd)T,…,φN(xN)T]TIs a kernel function matrix, phin(xn)＝[1,k(xn-d+1,xn),…,k(xn,xn)]K (x, x) is a kernel function, ε -N (0, β)-1) Random errors that follow a normal distribution; beta is a-1Is the variance of epsilon;
step 3, introducing a prior parameter alpha ═ alpha for w0,…,αd]TI.e. p (w)i)＝N(0,αi -1) I-0, …, d, training the model with the time series data set X, and iteratively solving the parameters α and β, i.e., order
Figure BDA0002652639800000041
Figure BDA0002652639800000042
Wherein m is the posterior mean value of w, miAn i-th element being m; a posterior variance matrix of w sigmaiiI rows and i columns of elements of the matrix Σ. Adding attention weight c ═ c to weight parameter w based on correlation between time series data when iteratively solving0,…,cd]TK represents the number of iterations; alpha is alpha-1Is a prior variance; p (w)i) Is wiA probability density function; w is aiIs w element i; alpha is alphaiIs alpha element i; t ═ xd+1,…,xN]T；
Step 3.1, data initialization is carried out: setting a prior parameter alpha as a random vector with the dimension of 201, setting a parameter beta as a random number between 0 and 1, and setting the maximum iteration number kth3000, initialize k 1;
step 3.2, performing iterative updating;
step 3.2.1, updating the posterior variance matrix sigma of w: sigma ═ Z + β ΦTΦ)-1；
Step 3.2.2, updating the posterior mean m of w: beta-sigma phiTt；
Step 3.2.3, updating alpha:
Figure BDA0002652639800000051
wherein alpha isiIs α, i is 0, …, d;
step 3.2.4, updating beta:
Figure BDA0002652639800000052
step 3.2.5, let k equal to k +1, if the iteration number k is greater than or equal to kthOr | | | αk+1-αkIf the value of | is converged, the iteration is finished and alpha and beta are output-1(ii) a Otherwise, skipping to 3.2.1;
the attention weight c: c. Cj＝f(s([xj,…,xN-d-1+j]T,t))；
Wherein j is 1, …, d, c is [ c ]0,…,cd]T,c0＝1；t＝[xd+1,…,xN]T；s(a,b)＝aTb is an attention metric used to represent the similarity of a and b;
Figure BDA0002652639800000053
a function is assigned to the attention weight.
s (a, b) is an attention metric, and can use, but is not limited to:
(1) dot product model: a isTb；
(2) Scaling the dot product model:
Figure BDA0002652639800000054
(3) cosine model: cos (a, b).
f(s) an attention weight assignment function, which may be used but is not limited to:
(1) softmax function:
Figure BDA0002652639800000055
(2) sigmoid function:
Figure BDA0002652639800000056
the weight parameter w adds an attention weight c ═ c0,…,cd]T：
Figure BDA0002652639800000057
Wherein, Z is AB,
Figure BDA0002652639800000061
the iterative solution method comprises the following steps: solving the maximum likelihood estimates of the following mathematical model:
Figure BDA0002652639800000062
wherein t ═ xd+1,…,xN]T。
The edge likelihood function of the mathematical model is: p (t | α, β) ═ p (t | w, β) p (w | c α) dw;
where p (w | c α) is the prior distribution of w and c is the attention weight.
The edge likelihood function maximum likelihood estimate is:
Figure BDA0002652639800000063
let the derivative of the edge likelihood be equal to 0:
Figure BDA0002652639800000064
Figure BDA0002652639800000065
obtaining:
Figure BDA0002652639800000066
wherein, Σ ═ Z + β ΦTΦ)-1，m＝βΣΦTt；
The probability distribution of the weight w is p (w | t, α, β) ═ N (w | m, Σ);
wherein, Σ ═ Z + β ΦTΦ)-1Is the variance of the received signal and the received signal,m＝βΣΦTt is the mean value.
Step 4, predicting the degradation prediction data x at the time of N +1 according to the RVM regression prediction modelN+1＝mTφN(xN)＝0.6175；
Step 5, judging xN+1Judging whether the machine can normally run or not by judging whether the fault threshold is reached or not, and if x is reachedN+1And if the fault threshold value is exceeded, the machine needs to be maintained, and if the fault threshold value is smaller than the fault threshold value, the machine normally operates.
In the above time series data prediction method based on the attention mechanism, the data prediction accuracy can be further improved by the existing data analysis and a large number of experimental comparisons. Failure warning is provided in advance and maintenance planning is improved, thereby reducing the cost of accidental maintenance and improving the reliability, availability and safety of the machine.
As shown in fig. 2, in the case of using real data (the data set of the XJTU-SY rolling bearing accelerated life test in 2019), the current algorithm (i.e., the attention distribution of the data is not considered) is applied to the training data N of 316, the training step d of 200, the root mean square error of the current algorithm is 0.2106, and the root mean square error of the method of the present invention is 0.0268 in the case of the training model; for the (N + 1) th data with the true value of 0.6222, the current algorithm predicted value is 0.5035, and the method predicted value is 0.6175; for the (N + 2) th data with the true value of 0.2124, the predicted value of the current algorithm is 0.341, and the predicted value of the method is 0.3151; compared with the former data prediction, the prediction error is reduced by 18.3 percent and is closer to the true value. The broken line graph in FIG. 2 represents real data, star points are prediction data, and the graph can show that the error between the prediction data and the real data is small, so that the prediction precision is improved.
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; such modifications and substitutions do not depart from the spirit of the corresponding technical solutions and scope of the present invention as defined in the appended claims.

Claims (1)

1. A time series data prediction method based on an attention mechanism is characterized in that: the method comprises the following steps:
step 1: collecting an accelerated life degradation test data set of the rolling bearing, wherein { x1，…，xNIs { t }1，…，tNThe observation value set at the moment is multiplied, and N is the number of the observation values;
step 2, based on the rolling bearing time series degradation data set X ═ X1，…，xN]TEstablishment of RVM regression prediction model, xn+1＝wTφ(xn) + ε, wherein xnN is the degraded data observed at time N, d, …, N; d is the embedding dimension; w ═ w0，…，wd]TAs a model weight parameter, [ phi ] - [ phi ]d(xd)T，…，φN(xN)T]TIs a kernel function matrix, phin(xn)＝[1，k(xn-d+1，xn)，…，k(xn，xn)]K (x, x) is a kernel function, ε -N (0, β)-1) Random errors that follow a normal distribution; beta is a-1Is the variance of epsilon;
step 3, introducing prior parameter alpha ═ alpha to model weight parameter w0，…，αd]TInstant command
Figure FDA0003487434940000011
Figure FDA0003487434940000012
Training a model by using a time series degradation data set X, and iteratively solving parameters alpha and beta, namely ordering
Figure FDA0003487434940000013
Figure FDA0003487434940000014
Wherein m is the posterior mean value of w, miAn i-th element being m; Σ is the a posteriori variance matrix of w,
Figure FDA0003487434940000015
adding attention weight c to [ c ] for weight parameter w based on correlation between time series degradation data in iterative solution for i rows and i columns of elements of matrix sigma0，…，cd]TK represents the number of iterations; alpha is alpha-1Is a prior variance; p (w)i) Is wiProbability density function, wiIs w element i; alpha is alphaiIs alpha element i; t ═ xd+1，…，xN]T；
Step 3.1, data initialization is carried out: setting prior parameter alpha as d +1 dimension random vector, setting parameter beta as random number between 0 and 1, setting maximum iteration number kthInitializing k to 1;
step 3.2, performing iterative updating;
step 3.2.1, update w posterior variance matrix sigma: sigma ═ Z + β ΦTΦ)-1
Step 3.2.2, updating the posterior mean m of w: where m is beta sigma phiTt；
Step 3.2.3, updating alpha:
Figure FDA0003487434940000016
wherein alpha isiIs α th element, i is 0, …, d.
Step 3.2.4, updating beta:
Figure FDA0003487434940000017
step 3.2.5, let k equal to k +1, if the iteration number k is greater than or equal to kthOr | | | αk+1-αkIf the value of | is converged, the iteration is finished and alpha and beta are output-1(ii) a Otherwise, skipping to 3.2.1;
step 4, predicting the degradation prediction data x at the time of N +1 according to the RVM regression prediction modelN+1＝mTφN[xN)；
Step 5, judging xN+1Judging whether the machine can normally run or not by judging whether the fault threshold is reached or not, and if x is reachedN+1If the fault threshold value is exceeded, the machine needs to be maintained, and if the fault threshold value is smaller than the fault threshold value, the machine normally operates;
wherein, the iterative solution in step 3 is: edge likelihood function
Figure FDA0003487434940000018
Figure FDA0003487434940000019
Let the derivative of the edge likelihood function equal 0;
Figure FDA00034874349400000110
obtaining:
Figure FDA0003487434940000021
where, sigma ═ Z + β ΦTΦ)-1，m＝β∑ΦTt; since the weight w a posteriori p (w | t, α, β) is N (w | m, Σ), Σ is (Z + β Φ)TΦ)-1For the posterior variance, m ═ beta ∑ ΦTt is the posterior mean value;
wherein, in step 3, the weight parameter w adds an attention weight c ═ c0，…，cd]T：
Figure FDA0003487434940000022
Wherein, Z is AB,
Figure FDA0003487434940000023
p (w | c α) is w prior distribution;
the attention weight cj＝f(s([xj，…，xN-d-1+j]T，t))
Wherein j is 1, …, d, c is [ c ]0，…，cd]T，c0＝1；t＝[xd+1，…，xN]T(ii) a s (a, b) is an attention metric, representing the similarity of a and b; f(s) is an attention weight assignment function.
CN202010875958.0A 2020-08-27 2020-08-27 Time series data prediction method based on attention mechanism Active CN112036082B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010875958.0A CN112036082B (en) 2020-08-27 2020-08-27 Time series data prediction method based on attention mechanism

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010875958.0A CN112036082B (en) 2020-08-27 2020-08-27 Time series data prediction method based on attention mechanism

Publications (2)

Publication Number Publication Date
CN112036082A CN112036082A (en) 2020-12-04
CN112036082B true CN112036082B (en) 2022-03-08

Family

ID=73581674

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010875958.0A Active CN112036082B (en) 2020-08-27 2020-08-27 Time series data prediction method based on attention mechanism

Country Status (1)

Country Link
CN (1) CN112036082B (en)

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102636740A (en) * 2012-04-18 2012-08-15 南京航空航天大学 Method for predicting faults of power electronic circuit based on FRM-RVM (fuzzy rough membership-relevant vector machine)
CN103942457A (en) * 2014-05-09 2014-07-23 浙江师范大学 Water quality parameter time series prediction method based on relevance vector machine regression
CN107610464A (en) * 2017-08-11 2018-01-19 河海大学 A kind of trajectory predictions method based on Gaussian Mixture time series models
CN107766968A (en) * 2017-09-26 2018-03-06 河海大学 Short-term wind speed forecasting method based on CAPSO RVM built-up patterns
CN108776694A (en) * 2018-06-05 2018-11-09 哈尔滨工业大学 A kind of time series abnormal point detecting method and device
CN110232416A (en) * 2019-06-13 2019-09-13 中国人民解放军空军工程大学 Equipment failure prediction technique based on HSMM-SVM
CN110929768A (en) * 2019-11-14 2020-03-27 国电大渡河检修安装有限公司 Prediction method for machine fault

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102636740A (en) * 2012-04-18 2012-08-15 南京航空航天大学 Method for predicting faults of power electronic circuit based on FRM-RVM (fuzzy rough membership-relevant vector machine)
CN103942457A (en) * 2014-05-09 2014-07-23 浙江师范大学 Water quality parameter time series prediction method based on relevance vector machine regression
CN107610464A (en) * 2017-08-11 2018-01-19 河海大学 A kind of trajectory predictions method based on Gaussian Mixture time series models
CN107766968A (en) * 2017-09-26 2018-03-06 河海大学 Short-term wind speed forecasting method based on CAPSO RVM built-up patterns
CN108776694A (en) * 2018-06-05 2018-11-09 哈尔滨工业大学 A kind of time series abnormal point detecting method and device
CN110232416A (en) * 2019-06-13 2019-09-13 中国人民解放军空军工程大学 Equipment failure prediction technique based on HSMM-SVM
CN110929768A (en) * 2019-11-14 2020-03-27 国电大渡河检修安装有限公司 Prediction method for machine fault

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
"Dynamic Topology Based Network Energy-saving Mechanism";Wang Junwei等;《2015 27th Chinese Control and Decision Conference(CCDC)》;20151231;全文 *
"一种基于改进RVM变压器故障诊断新方法";付华等;《辽宁工程技术大学学报（自然科学版）》;20170930;第36卷(第9期);全文 *
"基于改进RVM算法的三元锂离子电池SOC估计";刘金江等;《四川师范大学学报（自然科学版）》;20181030;第41卷(第3期);全文 *
"基于注意力机制的车辆行为预测";蔡英凤等;《江苏大学学报（自然科学版）》;20200330;第41卷(第2期);全文 *
"相关向量机及其在故障诊断与预测中的应用";马登武等;《海军航空工程学院学报》;20130228;第28卷(第2期);全文 *

Also Published As

Publication number Publication date
CN112036082A (en) 2020-12-04

Similar Documents

Publication Publication Date Title
Chen et al. Anomaly detection and critical SCADA parameters identification for wind turbines based on LSTM-AE neural network
Xiang et al. LSTM networks based on attention ordered neurons for gear remaining life prediction
CN112926273B (en) Method for predicting residual life of multivariate degradation equipment
CN110955963B (en) Aviation cable residual life prediction method
Caceres et al. A probabilistic Bayesian recurrent neural network for remaining useful life prognostics considering epistemic and aleatory uncertainties
CN111382542B (en) Highway electromechanical device life prediction system facing full life cycle
CN112611584B (en) Fatigue failure detection method, device, equipment and medium for wind generating set
US12001180B2 (en) Condition-based method for malfunction prediction
CN111610466A (en) Transformer state prediction method and device and storage medium
CN115526241B (en) Real-time abnormality detection method, device, equipment and medium for aviation hydraulic pump station
Osman et al. Soft Sensor Modeling of Key Effluent Parameters in Wastewater Treatment Process Based on SAE‐NN
CN116701877A (en) Fan abnormal value repairing method based on conditional diffusion model
Lin et al. Attention-based Gate Recurrent Unit for remaining useful life prediction in prognostics
Li et al. Prognosis for stochastic degrading systems with massive data: A data-model interactive perspective
CN111783242A (en) RVM-KF-based rolling bearing residual life prediction method and device
CN114254904B (en) Method and device for evaluating operation health degree of engine room of wind turbine generator
CN116562120A (en) RVE-based turbine engine system health condition assessment method and RVE-based turbine engine system health condition assessment device
CN118188342A (en) Fan-oriented fault early warning and life prediction method
CN111160666A (en) Health state and reliability assessment method for monitoring strong noise and non-periodic state
CN112036082B (en) Time series data prediction method based on attention mechanism
CN117972625A (en) Attention neural network data assimilation method based on four-dimensional variation constraint
CN111930728A (en) Method and system for predicting characteristic parameters and fault rate of equipment
Juesas et al. Ascertainment-adjusted parameter estimation approach to improve robustness against misspecification of health monitoring methods
CN116739130A (en) Multi-time scale load prediction method of TCN-BiLSTM network
CN111461416B (en) Wind speed prediction method, system, electronic equipment and storage medium

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