CN111222290A - Large-scale equipment residual service life prediction method based on multi-parameter feature fusion - Google Patents

Large-scale equipment residual service life prediction method based on multi-parameter feature fusion Download PDF

Info

Publication number
CN111222290A
CN111222290A CN202010031007.5A CN202010031007A CN111222290A CN 111222290 A CN111222290 A CN 111222290A CN 202010031007 A CN202010031007 A CN 202010031007A CN 111222290 A CN111222290 A CN 111222290A
Authority
CN
China
Prior art keywords
equipment
state
health
parameter
parameters
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
Application number
CN202010031007.5A
Other languages
Chinese (zh)
Other versions
CN111222290B (en
Inventor
彭江超
郝平
范兴刚
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.)
Zhejiang University of Technology ZJUT
Original Assignee
Zhejiang University of Technology ZJUT
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 Zhejiang University of Technology ZJUT filed Critical Zhejiang University of Technology ZJUT
Priority to CN202010031007.5A priority Critical patent/CN111222290B/en
Publication of CN111222290A publication Critical patent/CN111222290A/en
Application granted granted Critical
Publication of CN111222290B publication Critical patent/CN111222290B/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
    • 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
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/01—Probabilistic graphical models, e.g. probabilistic networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Biology (AREA)
  • Computing Systems (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Algebra (AREA)
  • Medical Informatics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

A method for predicting the residual service life of large equipment based on multi-parameter feature fusion comprises the following steps: acquiring various sensor time sequence parameters of large-scale equipment in a laboratory through a large-scale online monitoring system; carrying out regression analysis on the multi-parameter continuous values by utilizing a Relieff algorithm, and screening the characteristic weights to obtain a parameter type with high correlation with the equipment state; performing data dimension reduction and feature extraction on the screened parameters based on a principal component analysis method, and obtaining a health index representing the running state of the large-scale equipment through weight fusion; constructing an HMM model based on an expectation maximization algorithm, performing model training by taking the health index as a training set, and finding a hierarchical model for evaluating the health state of equipment corresponding to the current health index; performing exponential likelihood value calculation through a Viterbi algorithm to obtain a health index nearest to the likelihood value, and predicting the index difference by using a weighted average method to obtain a health state fitting curve; and then calculating the predicted value of the residual service life of the large-scale equipment.

Description

Large-scale equipment residual service life prediction method based on multi-parameter feature fusion
The technical field is as follows:
the invention relates to a method for predicting the residual service life of large equipment, in particular to a method for predicting the residual service life of the large equipment based on multi-parameter feature fusion.
Background art:
as chemical equipment used for polymer material processing or chemical research in colleges and universities laboratories, the reliability of the equipment is improved, the maintenance cost is reduced, and the establishment of a health maintenance management technology for large-scale equipment with the residual service life as the core is very important. The prediction of the residual life of the large-scale equipment refers to predicting the residual life of the equipment until the fault threshold value is used for health by analyzing the process of declining historical operating performance of the equipment, so that the maintenance is realized in advance, and the loss caused by fault shutdown is avoided.
Although the common service life prediction-based method considers the differences among different large-scale equipment individuals, the common service life prediction-based method is directed at single sensor performance parameters, and meanwhile, the similarity of the influences of different sensor parameters on the equipment is not considered. The performance parameters of the various sensors after the relevance screening can more comprehensively represent the health state of the equipment. The service life prediction is carried out aiming at multiple parameters, namely, multiple performance parameters are fused into a comprehensive health index, and then the service life of the equipment is predicted through an HMM state model based on an expectation maximization algorithm.
The invention content is as follows:
in order to solve the problem of single parameter in the traditional method for predicting the service life of the large-scale equipment, the method improves the existing method for predicting the health state of the large-scale equipment, fuses various sensor parameters into the health indexes representing the performance state of the equipment through a Relieff-PCA algorithm, constructs an HMM state training model based on an EM algorithm by substituting the health indexes into a time sequence, finds out the corresponding health state, calculates the likelihood value by using a Viterbi algorithm, and performs curve fitting on the health indexes by using a weighted average method so as to obtain the predicted value of the health state of the equipment.
The invention adopts the following technical scheme: a method for predicting the residual service life of large equipment based on multi-parameter feature fusion comprises the following steps:
1, acquiring various sensor time sequence parameters of instrument equipment in a laboratory through a large-scale online monitoring system;
2, carrying out regression analysis on the multi-parameter continuous values by utilizing a Relieff algorithm, and screening through feature weights to obtain a parameter type with high correlation with the equipment state;
3, performing data dimension reduction and feature extraction on the screened parameters based on a Principal Component Analysis (PCA), and obtaining a health index representing the running state of the large-scale equipment through weight fusion;
4, constructing an HMM model based on an expectation-maximization algorithm, performing model training by taking the health index as a training set, and finding a hierarchical model for evaluating the health state of equipment corresponding to the current health index;
5, calculating a likelihood value through a Viterbi algorithm to obtain a nearest neighbor health index of the likelihood value, and then performing fitting prediction on a health state curve by using a weighted average method;
and 6, calculating a predicted value of the residual service life of the large-scale equipment.
The invention has the following optimization effects: the invention discloses a method for predicting the residual service life of large equipment based on multi-parameter feature fusion, which belongs to a data driving method and is used for obtaining the change trend of the use state of the equipment according to multi-sensor parameters of the equipment. The multi-parameter fusion process can realize the fusion of data and the fusion of characteristics, so that the use state track of the large-scale equipment is represented practically. Therefore, the method can utilize more comprehensive sensor data monitoring information, achieves a better prediction result by means of a machine learning algorithm based on the data characteristic diversity of the equipment, and has practical application value.
Description of the drawings:
fig. 1 is a flow step diagram of a method for predicting the remaining service life of a large-scale device based on multi-parameter feature fusion, which is implemented by the invention.
FIG. 2 is a flow chart of data flow within the process of predicting health status of a device according to the present invention.
Fig. 3 is a fitted trend plot of the remaining useful life of the equipment used for the study.
Fig. 4 is a health status grade corresponding to the health index of the large-scale equipment in the present invention.
The specific implementation mode is as follows:
referring to fig. 1, the method for predicting the remaining service life of a large-scale device based on multi-parameter feature fusion mainly includes two parts, namely data processing and state prediction. The data processing section includes: parameter screening, which is used for removing parameters with similar influence factors in equipment use, so that the parameter types have diversity; the characteristic extraction is used for analyzing the linear change of the time series data of different types of parameters and finding out corresponding characteristic items by utilizing matrix decomposition; and the weight fusion is used for fusing various performance parameters into a comprehensive health index, so that the subsequent model data can be conveniently substituted. The state prediction section includes: establishing a model, namely establishing a state model used by equipment based on historical performance data by adopting a machine learning algorithm on the basis of multi-parameter feature fusion; and updating and predicting parameters, updating and predicting model parameters by adopting a Verbite algorithm in combination with historical monitoring data, and finally calculating a predicted value of the residual service life of the equipment.
Referring to fig. 2, the method for predicting the remaining service life of the large-scale device based on multi-parameter feature fusion of the present invention includes the following steps:
step 1, acquiring various sensor time sequence parameters of large-scale equipment in a laboratory through a large-scale online monitoring system;
1.1, firstly, acquiring time sequence parameters of a plurality of sensors of large-scale laboratory instruments and equipment through an online monitoring system of the large-scale equipment, wherein the parameters of a single sensor cannot completely represent the use state information of the equipment and only contain partial working information of the equipment.
1.2, the prediction accuracy of the residual service life value of the equipment is improved by fusing the characteristic weights of various parameters.
And 2, carrying out regression analysis on the multi-parameter continuous values by utilizing a Relieff algorithm, and screening through feature weights to obtain a parameter type with high correlation with the equipment state, wherein the method specifically comprises the following steps:
2.1, performing relevance screening on sensor parameters of time sequence data through a Relieff algorithm, wherein the larger the characteristic weight of the parameter is, the larger the influence factor of the parameter on equipment use is, the smaller the weight is, and the influence factor of the relevance is small.
2.2 method for finding feature difference of neighboring samples, Di,DjThe difference value diff (D) of the parameters at the time i, j for the acquisition point of the v-th sensor time sequence datai,v,Dj,v) Comprises the following steps:
diff(Di,v,Dj,v)＝(Di,v,Dj,v)/Sv(1)
2.3 wherein Di,vRepresenting the value of a parameter, S, of the v-th sensor data set at time ivIs the normalized unit of the sensor data characteristic v.
2.4 the specific steps of the Relieff algorithm in this study are as follows:
2.4.1 determining a total data sample set D and a feature set M, and initializing according to the sampling times M and the selected k adjacent samples;
2.4.2 randomly selecting a sample R in a training sample set D, and finding k nearest neighbor samples H of R from a similar sample set of Rj(j ═ 1,2, 3, …, k), and then k nearest neighbor samples M are found from each sample set that is not classified as Rj(C) And performing m times of repeated sampling;
2.4.3 calculating the weight matrix W (A) for each parameter:
Figure BDA0002364293870000031
wherein p (C) represents the probability of the occurrence of the class C;
and 2.4.4, sequentially calculating weights of all the sensor feature sets M according to the step (2.4.3) to obtain a weight matrix W of the parameters, and reordering according to the weight.
2.5 where the feature weight threshold δ ═ avg (w) is used, the algorithm is substituted to screen out the n-dimensional sensor parameter set, and a new device state feature data set F ═ F is obtained1,f2,…，fnAnd the new characteristic data set can represent the health state of the large-scale equipment more simply and accurately.
And 3, performing data dimension reduction and feature extraction on the screened parameters based on a Principal Component Analysis (PCA), and obtaining a health index representing the running state of the large equipment through weight fusion, wherein the process is as follows:
and 3.1, carrying out data characteristic dimension reduction processing on the screened sensor parameters by using the optimized sensor parameter set obtained in the step 2 through a principal component analysis method, and carrying out assignment fusion through corresponding weights to obtain a new fusion health index representing the health state of the equipment.
3.2 assume that device sensor dataset F after having been subjected to the ReliefF algorithm feature screening includes an n-dimensional sensor state parameter set, i.e., F ═ F (F ═ F)1，f2，…,fn)T∈Rt×n，fi＝{fi,1,fi,2,…,fi,tAnd expressing the time sequence data corresponding to the ith sensor parameter.
3.3 first, perform data centralization process on the data matrix F (i.e. mean value returns to 0, variance returns to 1), and the matrix after the centralization is F*(ii) a Calculating F*Corresponding to eigenvalues λ of covariance matrix (n × n)i(i ═ 1,2, 3, …, n) and eigenvector ξ;
3.4 dividing the eigenvalues λiIn order of magnitude, i.e. λ1≥λ2≥…≥λnAnd calculating the accumulated influence rates of different quantities of characteristic parameters
Figure BDA0002364293870000032
Taking eigenvectors corresponding to the first P eigenvalues exceeding the influence threshold value to form a projection matrix in sequence
Figure BDA0002364293870000033
3.5 multiplying the processed sample matrix by the projection matrix:
Figure BDA0002364293870000034
the product is a reduced-dimension principal component dataset X, whichWherein X ═ X1,x2,…,xP)TWherein X is defined as a principal component element of the original data F, X1,x2,… referred to as first principal component, second principal component …; as known from the calculation formula, the first P principal components finally retained by the algorithm are original data
Figure BDA0002364293870000035
The linear combination result of the converted sensor parameters used to characterize the health of the device.
3.6 normalization processing is carried out on the characteristic parameters of the sensor by using dispersion standardization, and multiple performance parameters are fused into a single performance index, namely the equipment health index O, through parameter fusioniThe calculation formula is as follows:
Figure BDA0002364293870000036
Figure BDA0002364293870000037
where rhojFor the weighted value of the performance parameter of P after dimension reduction,
Figure BDA0002364293870000038
in order to perform the parameter value after dispersion standardization, the specific calculation method comprises the following steps:
Figure BDA0002364293870000039
wherein xmax，xminThe maximum value and the minimum value in the corresponding sensor parameter sequence are respectively.
Step 4, constructing an HMM model based on an expectation-maximization algorithm, performing model training by taking the health index as a training set, and finding a hierarchical model for evaluating the health state of equipment corresponding to the current health index;
4.1 the health status of the device is classified first, and the health status of the device is divided into 5 classes of health, good, normal, abnormal and fault, corresponding to 5 hidden states of the HMM model, for the training and evaluation of the model.
4.2 index of health of the plant O in step 3iAs an observation sequence O ═ O1，O2，…，Ot}, to estimate model HMM: λ ═ values of the respective parameters of (a, B, pi), so that under this model, the probability of observation sequence P (O | λ) is maximized.
4.3 based on the Forward-Backward Algorithm, the Forward probability variables α are first definedt(j) Backward probability variable βt(j) And a probability variable gammat(j) The following are:
αt(j)＝p(O1，O2，…,Ot,it＝qj|λ) (6)
βt(j)＝p(Ot+1，Ot+2,…,OT|it＝qj，λ) (7)
γt(j)＝p(it＝qj|O，λ) (8)
wherein, T is 1,2, 3, …, T, j is 1,2, 3, …, N, αt(j) Given model λ at time t, the output observation sequence is { O }1，O2，…，OtAnd the device is in state qjβt(j) Given the model λ at time t, the device is in state qjThe output observed value is Ot+1，Ot+2，…,OTThe probability of (d); gamma rayt(j) Given the model λ at time t, the device is in state qjThe probability of the following.
4.4 the specific calculation for solving p (O | λ) according to the substitution algorithm defined in step 3) is as follows:
Figure BDA0002364293870000041
Figure BDA0002364293870000042
Figure BDA0002364293870000043
4.5 defining the device in state q at time t under the entire sequence of observations and given modeljAt time t +1, is in state qkHas a probability of ξ (j, k) of:
Figure BDA0002364293870000044
4.6 derived from the probability formula:
p(it＝qj,it+1＝qj,O|λ)＝ajkbk(Ot+1)αt(j)βt+1(j) (13)
Figure BDA0002364293870000045
4.7 substituting the above equations (12) and (13) into equation (11), there are:
Figure BDA0002364293870000051
4.8 according to γt(j) And ξ (j, k) are defined by the following relationship:
Figure BDA0002364293870000052
4.9 calculate State expectation: assume that the observed value at time t is vlWill be gammat(j) The device directly summed with respect to T (1. ltoreq. T. ltoreq.T) is in state qjIs the average of the number of times of, i.e. the variable gammat(j) Condition expectation of
Figure BDA0002364293870000053
And gammat(j) State expectation of
Figure BDA0002364293870000054
4.10 according to the above calculation results, the parameter re-estimation formula of the HMM is:
Figure BDA0002364293870000055
4.11 calculating according to the reassessment formula to obtain a new updated model
Figure BDA0002364293870000056
Then, new iteration is carried out by the reestimation model until
Figure BDA0002364293870000057
Obtaining an optimal model
Figure BDA0002364293870000058
The iteration is terminated. Wherein, delta is a predefined convergence condition, otherwise, the training steps are repeated until convergence.
And 5, calculating a likelihood value through a Viterbi algorithm to obtain a nearest neighbor health index of the likelihood value, and then performing fitting prediction on a health state curve by using a weighted average method, wherein the method specifically comprises the following steps:
5.1, predicting the health state of the equipment, mainly calculating a likelihood value through a Viterbi algorithm to obtain a plurality of health index difference values which are closest to the likelihood value, and then predicting the difference values among the indexes by using a weighted average method to further obtain a predicted value of an equipment health state curve:
Figure BDA0002364293870000059
Figure BDA00023642938700000510
wherein O isi(1. ltoreq. i. ltoreq.N) represents the health index O at the time ttIth similar history index, WiIndicates the difference of health indexes Oi+1-OiThe occupied proportion, N represents the number of times of weighting (time sequence length T), LLtIndicating an exponential likelihood value at time t, LLiRepresentation and LLtHistory key of ith proximityA health index likelihood value.
And 6, calculating a predicted value of the residual service life of the large-scale equipment, substituting the predicted algorithm formula determined in the step 5 and the equipment health state model in the step 4 into the time sequence health index, predicting the change trend of the health index, and finally fitting and calculating the residual service life of the equipment.
The method for predicting the remaining service life of the large-scale equipment based on the multi-parameter feature fusion, provided by the embodiment of the invention, comprises the steps of screening and weighting the multi-parameter features, more accurately reflecting the running state of the equipment, finding out the hidden state corresponding to the health index by using an HMM (hidden Markov model), and realizing optimal path solution by using a Viterbi algorithm, so as to predict the curve of the remaining service life of the equipment in a period of time in the future. The method has excellent analysis, modeling and prediction capabilities on the sensor monitoring data of the large-scale equipment, effectively improves the prediction accuracy of the residual service life of the large-scale equipment, and helps equipment maintenance personnel to realize timely troubleshooting and equipment maintenance on the basis.
The embodiments described in this specification are merely illustrative of implementations of the inventive concept and the scope of the present invention should not be considered limited to the specific forms set forth in the embodiments but rather by the equivalents thereof as may occur to those skilled in the art upon consideration of the present inventive concept.

Claims (1)

1. A method for predicting the remaining service life of large equipment based on multi-parameter feature fusion is characterized by comprising the following steps:
step 1, acquiring various sensor time sequence parameters of large-scale equipment in a laboratory through a large-scale online monitoring system;
1.1, firstly, acquiring time sequence parameters of a plurality of sensors of large-scale laboratory instruments and equipment through an online monitoring system of the large-scale equipment, wherein the parameters of a single sensor cannot completely represent the use state information of the equipment and only contain partial working information of the equipment;
1.2, the prediction accuracy of the residual service life value of the equipment is improved by fusing the characteristic weights of various parameters;
and 2, carrying out regression analysis on the multi-parameter continuous values by utilizing a Relieff algorithm, and screening through feature weights to obtain a parameter type with high correlation with the equipment state, wherein the method specifically comprises the following steps:
2.1, performing relevance screening on sensor parameters of time sequence data through a Relieff algorithm, wherein the larger the characteristic weight of the parameter is, the larger the influence factor of the parameter on equipment use is, the smaller the weight is, and the smaller the influence factor of the relevance is;
2.2 method for finding feature difference of neighboring samples, Di，DjThe difference value diff (D) of the parameters at the time i, j for the acquisition point of the v-th sensor time sequence datai，v，Dj，v) Comprises the following steps:
diff(Di，v，Dj，v)＝(Di，v，Dj，v)/Sv(1)
2.3 wherein Di，vRepresenting the value of a parameter, S, of the v-th sensor data set at time ivIs a standardized unit of sensor data characteristic v;
2.4 the specific steps of the Relieff algorithm in this study are as follows:
2.4.1 determining a total data sample set D and a feature set M, and initializing according to the sampling times M and the selected k adjacent samples;
2.4.2 randomly selecting a sample R in a training sample set D, and finding k nearest neighbor samples H of R from a similar sample set of Rj(j ═ 1,2, 3.. times, k), and then k nearest neighbor samples M are found from each sample set which is not in the same class as Rj(C) And performing m times of repeated sampling;
2.4.3 calculating the weight matrix W (A) for each parameter:
Figure FDA0002364293860000011
wherein p (C) represents the probability of the occurrence of the class C;
2.4.4, calculating weights of all sensor feature sets M in sequence according to the step (2.4.3) to obtain a weight matrix W of the parameters, and reordering according to the weights;
2.5 where the feature weight threshold δ ═ avg (w) is used, the algorithm is substituted to screen out the n-dimensional sensor parameter set, and a new device state feature data set F ═ F is obtained1，f2，…，fnThe new characteristic data set can represent the health state of the large-scale equipment more simply and accurately;
and 3, performing data dimension reduction and feature extraction on the screened parameters based on a Principal Component Analysis (PCA), and obtaining a health index representing the running state of the large equipment through weight fusion, wherein the process is as follows:
3.1, using the optimized sensor parameter set obtained in the step 2, performing data feature dimension reduction processing on the screened sensor parameters by a principal component analysis method, and performing corresponding weight assignment fusion to obtain a new fusion health index representing the health state of the equipment;
3.2 assume that device sensor dataset F after having been subjected to the ReliefF algorithm feature screening includes an n-dimensional sensor state parameter set, i.e., F ═ F (F ═ F)1，f2，...，fn)T∈Rt×n，fi＝{fi，1，fi，2，...，fi，tRepresenting the time sequence data corresponding to the ith sensor parameter;
3.3 first, perform data centralization process on the data matrix F (i.e. mean value returns to 0, variance returns to 1), and the matrix after the centralization is F*(ii) a Calculating F*Corresponding to eigenvalues λ of covariance matrix (n × n)i(i ═ 1,2, 3,. n), and feature vector ξ;
3.4 dividing the eigenvalues λiIn order of magnitude, i.e. λ1≥λ2≥…≥λnAnd calculating the accumulated influence rates of different quantities of characteristic parameters
Figure FDA0002364293860000021
Taking eigenvectors corresponding to the first P eigenvalues exceeding the influence threshold value to form a projection matrix in sequence
Figure FDA0002364293860000022
3.5 multiplying the processed sample matrix by the projection matrix:
Figure FDA0002364293860000023
the product is a principal component data set X after dimensionality reduction, wherein X is (X)1，x2，...，xP)TWherein X is defined as a principal component element of the original data F, X1，x2… are referred to as first principal component, second principal component … in this order; as known from the calculation formula, the first P principal components finally retained by the algorithm are original data
Figure FDA0002364293860000024
The projection is used for representing the linear combination result of the converted sensor parameters of the equipment health degree;
3.6 normalization processing is carried out on the characteristic parameters of the sensor by using dispersion standardization, and multiple performance parameters are fused into a single performance index, namely the equipment health index O, through parameter fusioniThe calculation formula is as follows:
Figure FDA0002364293860000025
Figure FDA0002364293860000026
where rhojFor the weighted value of the performance parameter of P after dimension reduction,
Figure FDA0002364293860000027
in order to perform the parameter value after dispersion standardization, the specific calculation method comprises the following steps:
Figure FDA0002364293860000028
wherein xmax，xminRespectively corresponding to the maximum value and the minimum value in the sensor parameter sequence;
step 4, constructing an HMM model based on an expectation-maximization algorithm, performing model training by taking the health index as a training set, and finding a hierarchical model for evaluating the health state of equipment corresponding to the current health index;
4.1, firstly, grading the health state of the equipment, firstly, dividing the health state of the equipment into 5 grades of health, good, normal, abnormal and fault, corresponding to 5 hidden states of an HMM model, and using the hidden states for training and evaluating the model;
4.2 index of health of the plant O in step 3iAs an observation sequence O ═ O1，O2，…，Ot}, to estimate model HMM: λ ═ values of the parameters (a, B, pi) such that the observed sequence probability P (O | λ) is maximal under this model;
4.3 based on the Forward-Backward Algorithm, the Forward probability variables α are first definedt(j) Backward probability variable βt(j) And a probability variable gammat(j) The following are:
αt(j)＝p(O1，O2，…，Ot，it＝qj|λ) (6)
βt(j)＝p(Ot+1，Ot+2，…，OT|it＝qj，λ) (7)
γt(j)＝p(it＝qj|0，λ) (8)
wherein, T is 1,2, 3, T, and j is 1,2, 3, T, N, αt(j) Given model λ at time t, the output observation sequence is { O }1，O2，...，OtAnd the device is in state qjβt(j) Given the model λ at time t, the device is in state qjThe output observed value is Ot+1，Ot+2，...，OrThe probability of (d); gamma rayt(j) Given the model λ at time t, the device is in state qjProbability of being lower;
4.4 the specific calculation for solving p (O | λ) according to the substitution algorithm defined in step 3) is as follows:
Figure FDA0002364293860000031
Figure FDA0002364293860000032
Figure FDA0002364293860000033
4.5 defining the device in state q at time t under the entire sequence of observations and given modeljAt time t +1, is in state qkHas a probability of ξ (j, k) of:
Figure FDA0002364293860000034
4.6 derived from the probability formula:
p(it＝qj，it+1＝qj，O|λ)＝ajkbk(Ot+1)αt(j)βt+1(j) (13)
Figure FDA0002364293860000035
4.7 substituting the above equations (12) and (13) into equation (11), there are:
Figure FDA0002364293860000036
4.8 according to γt(j) And ξ (j, k) are defined by the following relationship:
Figure FDA0002364293860000037
4.9 calculate State expectation: assume that the observed value at time t is vlWill be gammat(j) The average of the number of times the device is in state qj, i.e. the variable γ, is obtained by directly summing T (1. ltoreq. t.ltoreq.T)t(j) Strip ofArticle expectation
Figure FDA0002364293860000038
And gammat(j) State expectation of
Figure FDA0002364293860000039
4.10 according to the above calculation results, the parameter re-estimation formula of the HMM is:
Figure FDA00023642938600000310
4.11 calculating according to the reassessment formula to obtain a new updated model
Figure FDA0002364293860000041
Then, new iteration is carried out by the reestimation model until
Figure FDA0002364293860000042
Obtaining an optimal model
Figure FDA0002364293860000043
Terminating the iteration; wherein, delta is a predefined convergence condition, otherwise, the training steps are repeated until convergence;
and 5, calculating a likelihood value through a Viterbi algorithm to obtain a nearest neighbor health index of the likelihood value, and then performing fitting prediction on a health state curve by using a weighted average method, wherein the method specifically comprises the following steps:
5.1, predicting the health state of the equipment, mainly calculating a likelihood value through a Viterbi algorithm to obtain a plurality of health index difference values which are closest to the likelihood value, and then predicting the difference values among the indexes by using a weighted average method to further obtain a predicted value of an equipment health state curve:
Figure FDA0002364293860000044
Figure FDA0002364293860000045
wherein O isi(1. ltoreq. i. ltoreq.N) represents the health index O at the time ttIth similar history index, WiIndicates the difference of health indexes Oi+1-OiThe occupied proportion, N represents the number of times of weighting, and the time sequence length T, LLtIndicating an exponential likelihood value at time t, LLiRepresentation and LLt(ii) an ith similar historical health index likelihood value;
and 6, calculating a predicted value of the residual service life of the large-scale equipment, substituting the predicted algorithm formula determined in the step 5 and the equipment health state model in the step 4 into the time sequence health index, predicting the change trend of the health index, and finally fitting and calculating the residual service life of the equipment.
CN202010031007.5A 2020-01-13 2020-01-13 Multi-parameter feature fusion-based method for predicting residual service life of large-scale equipment Active CN111222290B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010031007.5A CN111222290B (en) 2020-01-13 2020-01-13 Multi-parameter feature fusion-based method for predicting residual service life of large-scale equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010031007.5A CN111222290B (en) 2020-01-13 2020-01-13 Multi-parameter feature fusion-based method for predicting residual service life of large-scale equipment

Publications (2)

Publication Number Publication Date
CN111222290A true CN111222290A (en) 2020-06-02
CN111222290B CN111222290B (en) 2024-04-09

Family

ID=70829417

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010031007.5A Active CN111222290B (en) 2020-01-13 2020-01-13 Multi-parameter feature fusion-based method for predicting residual service life of large-scale equipment

Country Status (1)

Country Link
CN (1) CN111222290B (en)

Cited By (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112712129A (en) * 2021-01-11 2021-04-27 深圳力维智联技术有限公司 Multi-sensor fusion method, device, equipment and storage medium
CN112734195A (en) * 2020-12-31 2021-04-30 平安科技（深圳）有限公司 Data processing method and device, electronic equipment and storage medium
CN113159566A (en) * 2021-04-19 2021-07-23 华南理工大学 Adaptive collection method, system, apparatus and medium for equipment health stage detection
CN113280908A (en) * 2021-06-16 2021-08-20 深圳防灾减灾技术研究院 Building structure real-time diagnosis method and device based on fusion index
CN113357138A (en) * 2021-07-21 2021-09-07 中国人民解放军陆军工程大学 Method and device for predicting remaining service life of hydraulic pump and terminal equipment
CN113393488A (en) * 2021-06-08 2021-09-14 南京师范大学 Behavior track sequence multi-feature simulation method based on quantum migration
CN113449473A (en) * 2021-06-29 2021-09-28 南京工业大学 Wind power bearing degradation trend prediction method, system and device based on DFCAE-BilSTM
CN113486586A (en) * 2021-07-06 2021-10-08 新智数字科技有限公司 Equipment health state evaluation method and device, computer equipment and storage medium
CN113487084A (en) * 2021-07-06 2021-10-08 新智数字科技有限公司 Method and device for predicting service life of equipment, computer equipment and computer-readable storage medium
CN113705887A (en) * 2021-08-27 2021-11-26 浙江工业大学 Data-driven photovoltaic power generation power prediction method and system
CN113744087A (en) * 2021-09-10 2021-12-03 江苏省电子信息产品质量监督检验研究院(江苏省信息安全测评中心) Method and device for intelligently detecting running state of transformer for electronic appliance
CN113779882A (en) * 2021-09-10 2021-12-10 中国石油大学(北京) Method, device, equipment and storage medium for predicting residual service life of equipment
CN113779884A (en) * 2021-09-14 2021-12-10 慧镕电子系统工程股份有限公司 Detection method for service life of recovered chip
CN113867262A (en) * 2020-06-30 2021-12-31 华晨宝马汽车有限公司 Apparatus, method and medium for monitoring operation state of spindle device
CN114298443A (en) * 2022-03-03 2022-04-08 北京航空航天大学杭州创新研究院 Industrial equipment predictive maintenance method and device based on health state index and electronic equipment
CN114429316A (en) * 2022-04-06 2022-05-03 北京全路通信信号研究设计院集团有限公司 Equipment health state prediction method and system in centralized monitoring system
CN117609716A (en) * 2023-10-30 2024-02-27 广州发展南沙电力有限公司 Power equipment residual life remote prediction method and system based on state health information

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107908864A (en) * 2017-11-14 2018-04-13 中国科学院空间应用工程与技术中心 A kind of complex device remaining life Forecasting Methodology of feature based fusion
US20180165592A1 (en) * 2016-12-13 2018-06-14 Industrial Technology Research Institute System and method for predicting remaining lifetime of a component of equipment
CN109376401A (en) * 2018-09-29 2019-02-22 西安交通大学 A kind of adaptive multi-source information preferably with the mechanical method for predicting residual useful life that merges
CN109948860A (en) * 2019-03-26 2019-06-28 哈工大机器人（合肥）国际创新研究院 A kind of mechanical system method for predicting residual useful life and system
CN110456199A (en) * 2019-08-14 2019-11-15 四川大学 A kind of method for predicting residual useful life of multisensor syste

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180165592A1 (en) * 2016-12-13 2018-06-14 Industrial Technology Research Institute System and method for predicting remaining lifetime of a component of equipment
CN107908864A (en) * 2017-11-14 2018-04-13 中国科学院空间应用工程与技术中心 A kind of complex device remaining life Forecasting Methodology of feature based fusion
CN109376401A (en) * 2018-09-29 2019-02-22 西安交通大学 A kind of adaptive multi-source information preferably with the mechanical method for predicting residual useful life that merges
CN109948860A (en) * 2019-03-26 2019-06-28 哈工大机器人（合肥）国际创新研究院 A kind of mechanical system method for predicting residual useful life and system
CN110456199A (en) * 2019-08-14 2019-11-15 四川大学 A kind of method for predicting residual useful life of multisensor syste

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
曹惠玲 等: "基于多参数融合相似的民航发动机寿命预测", 中国机械工程, no. 07, 13 November 2019 (2019-11-13) *
阙子俊 等: "基于UKF的轴承剩余寿命预测方法研究", 仪器仪表学报, no. 09, 15 September 2016 (2016-09-15) *

Cited By (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113867262A (en) * 2020-06-30 2021-12-31 华晨宝马汽车有限公司 Apparatus, method and medium for monitoring operation state of spindle device
CN112734195A (en) * 2020-12-31 2021-04-30 平安科技（深圳）有限公司 Data processing method and device, electronic equipment and storage medium
CN112734195B (en) * 2020-12-31 2023-07-07 平安科技（深圳）有限公司 Data processing method, device, electronic equipment and storage medium
CN112712129A (en) * 2021-01-11 2021-04-27 深圳力维智联技术有限公司 Multi-sensor fusion method, device, equipment and storage medium
CN112712129B (en) * 2021-01-11 2024-04-19 深圳力维智联技术有限公司 Multi-sensor fusion method, device, equipment and storage medium
CN113159566A (en) * 2021-04-19 2021-07-23 华南理工大学 Adaptive collection method, system, apparatus and medium for equipment health stage detection
CN113393488A (en) * 2021-06-08 2021-09-14 南京师范大学 Behavior track sequence multi-feature simulation method based on quantum migration
CN113280908A (en) * 2021-06-16 2021-08-20 深圳防灾减灾技术研究院 Building structure real-time diagnosis method and device based on fusion index
CN113449473A (en) * 2021-06-29 2021-09-28 南京工业大学 Wind power bearing degradation trend prediction method, system and device based on DFCAE-BilSTM
CN113449473B (en) * 2021-06-29 2023-09-15 南京工业大学 Wind power bearing degradation trend prediction method, system and device based on DFCAE-BiLSTM
CN113487084A (en) * 2021-07-06 2021-10-08 新智数字科技有限公司 Method and device for predicting service life of equipment, computer equipment and computer-readable storage medium
CN113486586A (en) * 2021-07-06 2021-10-08 新智数字科技有限公司 Equipment health state evaluation method and device, computer equipment and storage medium
CN113486586B (en) * 2021-07-06 2023-09-05 新奥新智科技有限公司 Device health state evaluation method and device, computer device and storage medium
CN113357138A (en) * 2021-07-21 2021-09-07 中国人民解放军陆军工程大学 Method and device for predicting remaining service life of hydraulic pump and terminal equipment
CN113705887A (en) * 2021-08-27 2021-11-26 浙江工业大学 Data-driven photovoltaic power generation power prediction method and system
CN113744087A (en) * 2021-09-10 2021-12-03 江苏省电子信息产品质量监督检验研究院(江苏省信息安全测评中心) Method and device for intelligently detecting running state of transformer for electronic appliance
CN113779882A (en) * 2021-09-10 2021-12-10 中国石油大学(北京) Method, device, equipment and storage medium for predicting residual service life of equipment
CN113779884A (en) * 2021-09-14 2021-12-10 慧镕电子系统工程股份有限公司 Detection method for service life of recovered chip
CN114298443B (en) * 2022-03-03 2022-06-14 北京航空航天大学杭州创新研究院 Industrial equipment predictive maintenance method and device based on health state index and electronic equipment
CN114298443A (en) * 2022-03-03 2022-04-08 北京航空航天大学杭州创新研究院 Industrial equipment predictive maintenance method and device based on health state index and electronic equipment
CN114429316A (en) * 2022-04-06 2022-05-03 北京全路通信信号研究设计院集团有限公司 Equipment health state prediction method and system in centralized monitoring system
CN117609716A (en) * 2023-10-30 2024-02-27 广州发展南沙电力有限公司 Power equipment residual life remote prediction method and system based on state health information

Also Published As

Publication number Publication date
CN111222290B (en) 2024-04-09

Similar Documents

Publication Publication Date Title
CN111222290A (en) Large-scale equipment residual service life prediction method based on multi-parameter feature fusion
CN110263846B (en) Fault diagnosis method based on fault data deep mining and learning
CN107992976B (en) Hot topic early development trend prediction system and prediction method
Alhoniemi et al. Process monitoring and modeling using the self-organizing map
US8140301B2 (en) Method and system for causal modeling and outlier detection
CN116757534B (en) Intelligent refrigerator reliability analysis method based on neural training network
CN110757510B (en) Method and system for predicting remaining life of robot
CN109409425B (en) Fault type identification method based on neighbor component analysis
CN109917777B (en) Fault detection method based on mixed multi-sampling rate probability principal component analysis model
CN110571792A (en) Analysis and evaluation method and system for operation state of power grid regulation and control system
CN113762329A (en) Method and system for constructing state prediction model of large rolling mill
CN112085252B (en) Anti-fact prediction method for set type decision effect
CN114861788A (en) Load abnormity detection method and system based on DBSCAN clustering
CN109240276B (en) Multi-block PCA fault monitoring method based on fault sensitive principal component selection
CN111639304B (en) CSTR fault positioning method based on Xgboost regression model
CN114118219A (en) Data-driven real-time abnormal detection method for health state of long-term power-on equipment
CN114048546B (en) Method for predicting residual service life of aeroengine based on graph convolution network and unsupervised domain self-adaption
CN110084301B (en) Hidden Markov model-based multi-working-condition process working condition identification method
CN116930042B (en) Building waterproof material performance detection equipment and method
CN109389313B (en) Fault classification diagnosis method based on weighted neighbor decision
CN115935814A (en) Transformer fault prediction method based on ARIMA-SVM model
Burmeister et al. Exploration of production data for predictive maintenance of industrial equipment: A case study
Foldager et al. On the role of model uncertainties in Bayesian optimisation
Febriansyah et al. Outlier detection and decision tree for wireless sensor network fault diagnosis
CN117544021B (en) Rotating speed control method and system of brushless direct current motor

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