CN113094822A - Method and system for predicting residual life of mechanical equipment - Google Patents

Method and system for predicting residual life of mechanical equipment Download PDF

Info

Publication number
CN113094822A
CN113094822A CN202110268518.3A CN202110268518A CN113094822A CN 113094822 A CN113094822 A CN 113094822A CN 202110268518 A CN202110268518 A CN 202110268518A CN 113094822 A CN113094822 A CN 113094822A
Authority
CN
China
Prior art keywords
neural network
data
predicting
training
mechanical equipment
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.)
Pending
Application number
CN202110268518.3A
Other languages
Chinese (zh)
Inventor
袁烨
黄虹
黎家骐
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.)
Huazhong University of Science and Technology
Original Assignee
Huazhong University of Science and Technology
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 Huazhong University of Science and Technology filed Critical Huazhong University of Science and Technology
Priority to CN202110268518.3A priority Critical patent/CN113094822A/en
Publication of CN113094822A publication Critical patent/CN113094822A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/17—Mechanical parametric or variational design
    • 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
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/04—Ageing analysis or optimisation against ageing

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Geometry (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Hardware Design (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a method and a system for predicting the residual life of mechanical equipment, and belongs to the field of state monitoring and residual life prediction. According to the invention, a convolutional neural network and a bidirectional gating cyclic unit are combined to form a hybrid neural network, so that time and space characteristics are effectively extracted, and the residual life prediction precision is improved.

Description

Method and system for predicting residual life of mechanical equipment
Technical Field
The invention belongs to the field of state monitoring and residual life prediction, and particularly relates to a residual life prediction method and a residual life prediction system for mechanical equipment.
Background
In recent years, with the rapid development of smart manufacturing, condition monitoring technology has been widely adopted in various engineering systems, such as wind energy conversion systems, battery systems, rolling bearing systems, and the like. Monitoring techniques are of great significance for prognosis and health management. Through analysis of the monitored data, one can make a remaining life prediction to estimate the available life of the device before failure occurs. Accurate residual life prediction for industrial equipment such as turbofan engines, bearings, etc. can timely prevent machine failure and effectively reduce maintenance costs. Based on recent research, the technologies for predicting the residual life of mechanical equipment are mainly classified into four categories: physical model-based methods, statistical model-based methods, machine learning-based methods, and fusion methods.
In recent years, deep learning algorithms are often used to automatically extract deeper abstract features from large-scale data, which is widely used in the field of residual life prediction. Common methods include convolutional neural networks, cyclic neural networks, long-and short-term memory artificial neural networks, gated cyclic units, and the like. In most conventional methods, data points of different time steps are usually assigned the same weight. In practice, however, these data points with different time steps do not provide as much information, and therefore, the prediction accuracy is not ideal.
Disclosure of Invention
In view of the above drawbacks or needs for improvement in the prior art, the present invention provides a method and a system for predicting remaining useful life of a mechanical device, which aim to assign different weights to data points of different time steps to highlight data points containing more degradation information, thereby solving the technical problem of low accuracy in predicting remaining useful life in the conventional method.
In order to achieve the above object, the present invention provides a method for predicting a remaining life of a mechanical device, including:
s1, acquiring and preprocessing mechanical equipment operation data to serve as a training set and a verification set;
s2, constructing a hybrid neural network; the hybrid neural network comprises a convolutional neural network and a bidirectional gated cyclic unit; the convolutional neural network is used for extracting data information under the input data space dimension; the input data is obtained by weighting data points of different time steps by corresponding attention weights; the bidirectional gating circulation unit is used for capturing the time correlation in the output data of the convolutional neural network;
s3, transmitting the initialized attention weight to a hybrid neural network, training the network by using a training set and feeding back loss to a genetic algorithm, continuously learning the genetic algorithm to obtain an optimal attention weight set, transmitting the optimal attention weight set to the hybrid neural network so as to achieve collaborative training and find an optimal solution, and returning to a trained prediction model when the loss on a verification set is minimum;
and S4, inputting the running state data of the mechanical equipment to be predicted into the trained prediction model to obtain a residual life prediction result.
Further, the preprocessing in step S1 includes:
data were normalized as follows:
Figure BDA0002973089150000021
wherein x isi,jIs the raw sensor data that is to be sensed,
Figure BDA0002973089150000022
is the true value of the normalized data, i represents the ith sensor, j represents the jth data point,
Figure BDA0002973089150000023
is the maximum and minimum values of the ith sensor data;
all data is segmented using a sliding window.
Further, the number of channels of the convolutional neural network is set to a time step.
Further, the calculation of the bi-directional gated loop unit is as follows:
Figure BDA0002973089150000031
wherein Xt,ht-1,ht,rt,zt,
Figure BDA0002973089150000032
ytRespectively representing an input vector, a state storage variable at the previous moment, a state storage variable at the current moment, a state of a reset gate, a state of an update gate, a state of a current candidate set and a state of an output vector at the current moment; wr,Wz,
Figure BDA0002973089150000033
WORespectively representing reset gate, update gate, candidate set, output vector sum by xtAnd ht-1Weight parameters of the formed connection matrix; i represents an identity matrix; []Representing a vector join; represents the matrix dot product; x represents the matrix product; σ represents a sigmoid activation function; tanh represents the tanh activation function; br、bz、bhAnd represents the offset parameter learned in the training process.
Further, the genetic algorithm specifically comprises the following steps:
population initialization: binary coding attention weights on chromosomes of a genetic space, wherein each attention weight corresponds to a string of binary codes;
weight transfer: converting binary codes into decimal fractions as attention weight values, transferring the attention weight values into a hybrid neural network, starting training and learning by the hybrid neural network, and returning corresponding loss values generated by prediction errors;
loss ordering: randomly selecting each group of binary codes, dividing the binary codes into a plurality of groups, sorting according to returned loss, and selecting the weight recoding code with the minimum error value in each group;
cross recombination: randomly combining the minimum error set selected in the last step two by two to obtain a plurality of sets I and II, randomly selecting part of gene points in the sets I, providing the rest unselected gene points by the sets II, and recombining the gene points in the two sets into a new gene set to obtain a recombined gene fragment;
mutation: carrying out original gene inversion operation on the genotype subjected to cross recombination according to the mutation probability;
generating a new population: the population is reconstructed from the crossovers and variations and a new training is started.
In general, the above technical solutions contemplated by the present invention can achieve the following advantageous effects compared to the prior art.
The invention combines a convolutional neural network and a bidirectional gating circulation unit to form a hybrid neural network so as to effectively extract time and space characteristics and improve the residual life prediction precision, on the basis, attention weight is introduced into the hybrid neural network, a collaborative training mechanism of the hybrid neural network and an optimized genetic algorithm is adopted, the hybrid neural network and the optimized genetic algorithm share, transmit parameters and loss feedback, the genetic algorithm searches for an optimal attention weight parameter through continuous transmission of parameters and feedback loss, and the attention weight distribution of different time step lengths in the residual life prediction is optimized, so that the importance of different time step lengths is accurately embodied The prediction scores are much lower, and the problem of low prediction precision of the existing residual life is effectively solved.
The invention also improves the traditional genetic algorithm, prevents the algorithm from falling into a local optimal state, and introduces the loss feedback of the hybrid neural network into the improved genetic algorithm, so that the process can obviously improve the performance by controlling the direction of random search.
Drawings
FIG. 1 is a schematic diagram of a method for predicting the remaining life of a turbofan engine based on a genetic optimization hybrid neural network according to an embodiment of the present invention;
FIG. 2 is the measurements of 14 selected sensors from all engines of the FD001 subset;
FIG. 3 is a graph of the true remaining life and predicted remaining life derived by the proposed method on two data sets;
fig. 4 is an attention profile of different experimental data sets.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
The method for predicting the residual life of mechanical equipment, as shown in fig. 1, includes: preprocessing the operation data of the mechanical equipment; model establishment and training: firstly, transferring initialized attention weight to a convolution-bidirectional gate control unit hybrid neural network, then training a model and feeding loss back to an improved genetic algorithm, continuously learning the genetic algorithm to obtain an optimal attention weight set, then transferring the optimal attention weight set to the hybrid neural network so as to achieve collaborative training and find an optimal solution, and returning to the model when the loss on a verification set is minimum; and predicting the returned model on the test set, and returning a residual life prediction result. The model established by the invention is a framework formed by a convolutional neural network and a bidirectional gating circulation unit, wherein the convolutional neural network extracts the spatial characteristics of data points, the bidirectional gating circulation unit acquires the time correlation of data, and the spatial characteristics and the time characteristics can be extracted by combining the two units, so that the complete characteristics in the data are learned.
As will be described in detail below with reference to the embodiment, for example, measurements of sensors of all engines of the FD001 subset are obtained, and 14 sensor data having a suitable variance and sufficient degradation information are selected for subsequent use, as shown in fig. 2, where the abscissa represents time and the unit is a period. Normalizing the data to eliminate the influence among different sensor characteristics, and segmenting the data by using a sliding window for the data with longer time, wherein the specific implementation mode is as follows:
(11) the normalization operation is performed by the following formula:
Figure BDA0002973089150000051
wherein x isi,jIs the raw sensor data that is to be sensed,
Figure BDA0002973089150000052
is the true value of the normalized data, i represents the ith sensor, j represents the jth data point,
Figure BDA0002973089150000053
is the maximum and minimum values of the ith sensor data.
(12) The sliding window is segmented, and in order to capture the time correlation between the data, the sliding window is used for packaging the data of adjacent time points, and the residual life prediction of the last data point in the time window is used as the residual life prediction of the time window. By analysis, the data sliding window time length is set to 30 and the step size is set to 1.
Preferably, the mixed neural network with attention weight introduced for training is composed of a convolutional neural network and a bidirectional gated cyclic unit, and the whole is as follows:
Y＝fH(x)
wherein the content of the first and second substances,
Figure BDA0002973089150000061
is a vector containing m time steps, n features, Y is the network output representing the prediction of remaining life, fH(. is) a neural network consisting of a nonlinear function that maps inputs to outputs.
The specific way of training is as follows:
(21) first, the original input data is selected through a sliding window, and then attention weights for different time steps are defined as follows:
Figure BDA0002973089150000062
i denotes the ith weight set, l denotes the ith time step,
Figure BDA0002973089150000063
representing the weight of the ith time step in the ith weight set;
by weighted calculation of attention weights for different time steps, the following input data can be obtained:
Figure BDA0002973089150000064
Figure BDA0002973089150000065
a value representing the ith data point in the tth sample;
thus, the data value X ═ X to which the attention weight is assigned1,X2,...,XT]Transmission to a convolutional-bi-directional cyclic gated hybrid neural network, XTThe data point representing the T-th sample is multiplied by the attention weight.
(22) The convolutional neural network is used to obtain the spatial features of the attention weighted input data, and the number of channels is set as a time step for the time correlation of uncompressed data. Thus, the convolutional neural network does not interfere with the temporal relationships in the data, but only compresses the spatial information. The output data is then input to a bi-directional gated recurrent neural network.
(23) The gating cycle unit learns the spatial characteristics extracted by the convolutional neural network and learns the time dependency relationship, and the unit is calculated as follows:
Figure BDA0002973089150000071
wherein Xt,ht-1,ht,zt,rt,
Figure BDA0002973089150000072
tRespectively representing the input vector, the state storage variable of the previous moment, the state storage variable of the current moment, the state of an update gate, the state of a reset gate, the state of the current candidate set and the state of the output vector of the current moment. Wz,Wr,
Figure BDA0002973089150000073
oRespectively representing the sum of the update gate, reset gate, candidate set, output vector and the sum of xtAnd ht-1And forming a weight parameter of the connection matrix. I represents an identity matrix; []Representing a vector join; represents the matrix dot product; x represents the matrix product; σ represents a sigmoid activation function; tanh represents the tanh activation function. br、bz、bhIs the offset learned during the training process;
the genetic algorithm sends the learned attention weights back to the hybrid neural network. At the same time, the hybrid network returns training losses to the genetic algorithm to guide its further learning and training. The invention can adopt the traditional genetic algorithm and the mixed neural network for cooperative training, and preferably, the invention also improves the traditional genetic algorithm as follows, and the main steps are as follows:
population initialization: binary coding attention weights on chromosomes of a genetic space, wherein each weight corresponds to a string of binary gene codes;
weight transfer: converting the binary code into decimal, namely an initial attention weight value, transferring the attention weight into a previous mixed neural network, starting training and learning by the neural network, and returning a corresponding loss value generated by a prediction error of the neural network;
loss ordering: randomly selecting each group of binary codes, dividing the binary codes into a plurality of groups, and selecting the weight with the minimum error value in each group for recoding according to returned loss sequencing (prediction error);
cross recombination: randomly combining the minimum error set selected in the last step two by two to obtain a plurality of sets I and II, randomly selecting part of gene points in the sets I, providing the rest unselected gene points by the sets II, and recombining the gene points in the two sets into a new gene set to obtain a recombined gene fragment; the gene points of the selected set are not fixed, but 50% are selected in this example; the common genetic algorithm is to select gene points randomly for interchange, but the method is not to select gene points in a set I randomly for recombination, and the gene points in the selected set I and the gene points in a set II which are not selected are recombined to obtain a new genotype. For example: if two gene segments with six positions are selected randomly at the three positions of 135 in the first set (246 is not selected, 246 is provided by the second set), recombining the gene points of the first set 135 and the gene points of the second set 246 into a new 6-position gene segment;
mutation: carrying out original gene inversion operation on the genotype subjected to cross recombination according to the mutation probability;
generating a new population: the population is reconstructed from the crossovers and variations and a new training is started.
The improved genetic algorithm of the invention is distinguished from the common genetic algorithm by comprising the following steps:
the selection strategy in the common genetic algorithm generally calculates the probability of each individual appearing in the filial generation to randomly select the individuals to form a population, and the invention randomly combines the selected minimum error set two by two to obtain a new set to carry out the next step of cross variation, so that the cross variation process has more possibilities.
The crossover operator in the ordinary genetic algorithm is mainly single-point or multi-point crossover, which carries out gene exchange by randomly selecting crossover points, and the finally obtained new gene segment still comprises part of the original genotype.
Compared with the common genetic algorithm, the improved genetic algorithm mainly adds a random mechanism in a selection strategy and a crossover operator to reconstruct the final population space, so that the algorithm has stronger global search capability.
In addition, the present invention introduces the loss feedback of previous hybrid neural networks into an improved genetic algorithm, which allows the process to significantly improve performance by controlling the direction of random searches.
Fig. 3 shows predicted remaining life and actual remaining life prediction curves obtained by the above method on two data sets FD001 and FD004, and it can be found that the actual value and the predicted value are very close, i.e. the method has high prediction accuracy.
Fig. 4 shows attention distribution plots of different experimental data sets to facilitate a better understanding of importance sampling for different time steps in a time series.
Table 1 compares the effect of the convolutional neural network plus several different recurrent neural networks, and also compares the common genetic algorithm and the improved genetic algorithm, and compares the root mean square error and the score of each method by controlling the type of the hybrid neural network and whether to introduce the attention weight, and it can be found that the effect of the method of combining the convolutional neural network and the bidirectional gated cyclic unit plus the improved genetic algorithm has significant superiority. Table 2 compares the method of the present invention with other methods commonly used in the prediction of remaining life, and the results are the same, and the effect of the method of the present invention is superior to that of the other methods in both evaluation indexes. The two model evaluation indexes are respectively Root Mean Square Error (RMSE) and score (score), the score is a model evaluation method commonly used in the field of residual life prediction, and the two evaluation indexes are calculated in the following mode:
Figure BDA0002973089150000091
Figure BDA0002973089150000092
wherein n is the number of data, yiIs the true value of the remaining life span,
Figure BDA0002973089150000093
the predicted value of the remaining life is,
Figure BDA0002973089150000094
a1＝10，a213, wherein a1、a2Are commonly specified values.
Table 1: comparative performance of different remaining life prediction methods
Figure BDA0002973089150000095
Figure BDA0002973089150000101
Table 2: results of experiments on different data sets for FD001 and FD004
Figure BDA0002973089150000102
In summary, the above technical solution conceived by the present invention enables a higher prediction accuracy with a relatively limited technical complexity compared to the prior art.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (7)

1. A method for predicting the residual life of mechanical equipment is characterized by comprising the following steps:
s1, acquiring and preprocessing mechanical equipment operation data to serve as a training set and a verification set;
s2, constructing a hybrid neural network; the hybrid neural network comprises a convolutional neural network and a bidirectional gated cyclic unit; the convolutional neural network is used for extracting data information under the input data space dimension; the input data is obtained by weighting data points of different time steps by corresponding attention weights; the bidirectional gating circulation unit is used for capturing the time correlation in the output data of the convolutional neural network;
s3, transmitting the initialized attention weight to a hybrid neural network, training the network by using a training set and feeding back loss to a genetic algorithm, continuously learning the genetic algorithm to obtain an optimal attention weight set, transmitting the optimal attention weight set to the hybrid neural network so as to achieve collaborative training and find an optimal solution, and returning to a trained prediction model when the loss on a verification set is minimum;
and S4, inputting the running state data of the mechanical equipment to be predicted into the trained prediction model to obtain a residual life prediction result.
2. The method for predicting the remaining life of mechanical equipment according to claim 1, wherein the preprocessing in step S1 includes:
data were normalized as follows:
Figure FDA0002973089140000011
wherein x isi,jIs the raw sensor data that is to be sensed,
Figure FDA0002973089140000012
is the true value of the normalized data, i represents the ith sensor, j represents the jth data point,
Figure FDA0002973089140000013
is the maximum and minimum values of the ith sensor data;
all data is segmented using a sliding window.
3. The method for predicting the residual life of mechanical equipment according to claim 1 or 2, wherein the number of channels of the convolutional neural network is set as a time step.
4. The method for predicting the residual life of mechanical equipment according to claim 1 or 2, wherein the calculation of the bidirectional gating cyclic unit is as follows:
Figure FDA0002973089140000021
wherein Xt,ht-1,ht,rt,zt,
Figure FDA0002973089140000022
ytRespectively representing an input vector, a state storage variable at the previous moment, a state storage variable at the current moment, a state of a reset gate, a state of an update gate, a state of a current candidate set and a state of an output vector at the current moment; wr,Wz,
Figure FDA0002973089140000023
WORespectively representing reset gate, update gate, candidate set, output vector sum by xtAnd ht-1Weight parameters of the formed connection matrix; i represents an identity matrix; []Representing a vector join; represents the matrix dot product; x represents the matrix product; σ represents a sigmoid activation function; tanh represents the tanh activation function; br、bz、bhAnd represents the offset parameter learned in the training process.
5. The method for predicting the residual life of mechanical equipment according to any one of claims 1 to 4, wherein the genetic algorithm comprises the following steps:
population initialization: binary coding attention weights on chromosomes of a genetic space, wherein each attention weight corresponds to a string of binary codes;
weight transfer: converting binary codes into decimal fractions as attention weight values, transferring the attention weight values into a hybrid neural network, starting training and learning by the hybrid neural network, and returning corresponding loss values generated by prediction errors;
loss ordering: randomly selecting each group of binary codes, dividing the binary codes into a plurality of groups, sorting according to returned loss, and selecting the weight recoding code with the minimum error value in each group;
cross recombination: randomly combining the minimum error set selected in the last step two by two to obtain a plurality of sets I and II, randomly selecting part of gene points in the sets I, providing the rest unselected gene points by the sets II, and recombining the gene points in the two sets into a new gene set to obtain a recombined gene fragment;
mutation: carrying out original gene inversion operation on the genotype subjected to cross recombination according to the mutation probability;
generating a new population: the population is reconstructed from the crossovers and variations and a new training is started.
6. A system for predicting remaining life of a mechanical device, comprising a processor and a machine-readable storage medium storing machine-executable instructions executable by the processor, the processor executing the machine-executable instructions to implement a method for predicting remaining life of a mechanical device according to any one of claims 1 to 5.
7. A machine-readable storage medium having stored thereon machine-executable instructions which, when invoked and executed by a processor, cause the processor to implement a method of predicting remaining life of a mechanical device as claimed in any one of claims 1 to 5.
CN202110268518.3A 2021-03-12 2021-03-12 Method and system for predicting residual life of mechanical equipment Pending CN113094822A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110268518.3A CN113094822A (en) 2021-03-12 2021-03-12 Method and system for predicting residual life of mechanical equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110268518.3A CN113094822A (en) 2021-03-12 2021-03-12 Method and system for predicting residual life of mechanical equipment

Publications (1)

Publication Number Publication Date
CN113094822A true CN113094822A (en) 2021-07-09

Family

ID=76667971

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110268518.3A Pending CN113094822A (en) 2021-03-12 2021-03-12 Method and system for predicting residual life of mechanical equipment

Country Status (1)

Country Link
CN (1) CN113094822A (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113722833A (en) * 2021-09-09 2021-11-30 湖南工业大学 Turbofan engine residual service life prediction method based on dual-channel long-short time memory network
CN114510870A (en) * 2022-01-07 2022-05-17 华东交通大学 Method and device for predicting residual life of underground structure of urban rail transit
CN115047350A (en) * 2022-06-24 2022-09-13 哈尔滨工业大学 Digital-analog linkage based lithium ion battery remaining service life prediction method
CN115481788A (en) * 2022-08-31 2022-12-16 北京建筑大学 Load prediction method and system for phase change energy storage system
WO2023174064A1 (en) * 2022-03-14 2023-09-21 华为技术有限公司 Automatic search method, automatic-search performance prediction model training method and apparatus
CN116956751A (en) * 2023-09-19 2023-10-27 北京航空航天大学 Binary quantization-based life prediction method and system for aero-engine

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113722833A (en) * 2021-09-09 2021-11-30 湖南工业大学 Turbofan engine residual service life prediction method based on dual-channel long-short time memory network
CN113722833B (en) * 2021-09-09 2023-06-06 湖南工业大学 Turbofan engine residual service life prediction method based on double-channel long-short-term memory network
CN114510870A (en) * 2022-01-07 2022-05-17 华东交通大学 Method and device for predicting residual life of underground structure of urban rail transit
CN114510870B (en) * 2022-01-07 2024-04-16 华东交通大学 Method and device for predicting residual life of underground structure of urban rail transit
WO2023174064A1 (en) * 2022-03-14 2023-09-21 华为技术有限公司 Automatic search method, automatic-search performance prediction model training method and apparatus
CN115047350A (en) * 2022-06-24 2022-09-13 哈尔滨工业大学 Digital-analog linkage based lithium ion battery remaining service life prediction method
CN115047350B (en) * 2022-06-24 2023-04-18 哈尔滨工业大学 Digital-analog linkage based lithium ion battery remaining service life prediction method
CN115481788A (en) * 2022-08-31 2022-12-16 北京建筑大学 Load prediction method and system for phase change energy storage system
CN115481788B (en) * 2022-08-31 2023-08-25 北京建筑大学 Phase change energy storage system load prediction method and system
CN116956751A (en) * 2023-09-19 2023-10-27 北京航空航天大学 Binary quantization-based life prediction method and system for aero-engine
CN116956751B (en) * 2023-09-19 2023-12-12 北京航空航天大学 Binary quantization-based life prediction method and system for aero-engine

Similar Documents

Publication Publication Date Title
CN113094822A (en) Method and system for predicting residual life of mechanical equipment
CN111710375B (en) Molecular property prediction method and system
Xu et al. Fedcorr: Multi-stage federated learning for label noise correction
CN111967343A (en) Detection method based on simple neural network and extreme gradient lifting model fusion
CN108876044B (en) Online content popularity prediction method based on knowledge-enhanced neural network
US20210334658A1 (en) Method for performing clustering on power system operation modes based on sparse autoencoder
CN111612125B (en) Novel HTM time pool method and system for online learning
CN113536670A (en) Federal transfer learning method for predicting remaining life of group products
CN115099296A (en) Sea wave height prediction method based on deep learning algorithm
CN114841072A (en) Differential fusion Transformer-based time sequence prediction method
CN115345222A (en) Fault classification method based on TimeGAN model
CN116244484B (en) Federal cross-modal retrieval method and system for unbalanced data
CN112465184A (en) Cloud energy storage system control method of small-sample generation type counterstudy network
CN116796617A (en) Rolling bearing equipment residual life prediction method and system based on data identification
Wang et al. Elongation prediction of steel-strips in annealing furnace with deep learning via improved incremental extreme learning machine
CN113762591B (en) Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning
CN116561302A (en) Fault diagnosis method, device and storage medium based on mixed knowledge graph reasoning
Zha et al. Recognizing plans by learning embeddings from observed action distributions
CN115293249A (en) Power system typical scene probability prediction method based on dynamic time sequence prediction
CN114943277A (en) Method for identifying operation condition and predicting operation performance of comprehensive energy system
CN114841063A (en) Aero-engine residual life prediction method based on deep learning
CN110909254B (en) Method and system for predicting question popularity of question-answering community based on deep learning model
CN113420492A (en) Modeling method for frequency response model of wind-solar-fire coupling system based on GAN and GRU neural network
CN113469013A (en) Motor fault prediction method and system based on transfer learning and time sequence
CN112348275A (en) Regional ecological environment change prediction method based on online incremental learning

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