CN115062528A - Prediction method for industrial process time sequence data - Google Patents

Prediction method for industrial process time sequence data Download PDF

Info

Publication number
CN115062528A
CN115062528A CN202210295183.9A CN202210295183A CN115062528A CN 115062528 A CN115062528 A CN 115062528A CN 202210295183 A CN202210295183 A CN 202210295183A CN 115062528 A CN115062528 A CN 115062528A
Authority
CN
China
Prior art keywords
data
lstm
prediction
deep
training
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
CN202210295183.9A
Other languages
Chinese (zh)
Inventor
沈继红
关昊夫
彭立章
谭思超
曾占魁
王宇晴
张康慧
戴运桃
王淑娟
廉春波
赵富龙
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.)
Harbin Engineering University
Original Assignee
Harbin Engineering University
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 Harbin Engineering University filed Critical Harbin Engineering University
Priority to CN202210295183.9A priority Critical patent/CN115062528A/en
Publication of CN115062528A publication Critical patent/CN115062528A/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]
    • G06F30/20—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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Abstract

The invention discloses a prediction method for industrial process time sequence data, and belongs to the field of industrial process time sequence data prediction. The invention includes: preparing a training data set; constructing a Deep long-short term memory network (Deep-LSTM) model, and increasing the number of LSTM units and the number of network layers; training a Deep-LSTM network model; inputting the test data set into Deep-LSTM after training, predicting lambda unknown data through the network, calculating the error between the predicted value and the true value, and verifying the prediction effect and accuracy of the network. Aiming at the problems of insufficient fault data samples and prediction accuracy in time sequence data prediction, the invention utilizes an improved GRU-BEGAN generation countermeasure network model to generate artificial samples to expand an original data set, and utilizes a Deep-LSTM model to predict the time sequence data of the industrial process.

Description

Prediction method for industrial process time sequence data
Technical Field
The invention belongs to the field of industrial process time sequence data prediction, and particularly provides a prediction method for industrial process time sequence data.
Background
With the continuous progress of scientific technology, the importance of data is more and more prominent. The goal of data processing is to obtain hidden knowledge from a large amount of data. The time sequence is a group of data with sequence of fixed point sampling, and is organized into a sequence according to the time characteristic and the space characteristic of the related variable. The time series prediction is that a time series prediction model of corresponding variables is established through time sequence relation among data, data characteristics, generation mechanism of the data and other factors, and prediction information obtained according to the time series prediction model provides assistance or guidance for production and life. In the field of industrial process faults, a time series prediction method is used for predicting observation variables in advance, and judging whether faults occur or not by combining a process monitoring method so as to achieve the aim of finding the faults in advance, perform multi-step prediction on the basis of single-step prediction, early warn the faults in advance and avoid larger loss caused by fault diffusion.
The traditional time series prediction model comprises the following steps: an autoregressive model, a moving average model, an autoregressive moving average model, etc., which are established based on stationary time series. In practice, however, many process variables are non-stationary in time series and tend to be somewhat trending, seasonal or periodic. Aiming at the limitation of the traditional time series prediction method, various artificial intelligence algorithm prediction models based on a neural network are rapidly developed in recent years, for example, the research of a machine learning time series prediction model represented by a Bayesian network and a support vector machine; on the basis of deep learning-based prediction model research, a long-short-term memory neural network (LSTM) provided on the basis of a Recurrent Neural Network (RNN) provides more applications and innovations for prediction in various fields, and the problem of long-order memory of data is solved by independent output of LSTM memory information and neuron operation results. For the prediction problem in the field of industrial processes, due to the fact that the sample data size required by model prediction is large and multi-step prediction capable of early warning needs to be achieved, the LSTM model is directly applied to fault data, and the problems that the time required by model training is long, a network is unstable, the time sequence characteristics of sampled data cannot be effectively extracted and the like exist.
The application of the deep neural network model can predict faults in the process industry, reduce unsafe accidents caused by system faults and play an important role in promoting the development of fault prediction technology. However, in many fields such as medical treatment, national defense, aerospace and the like, many factors such as environment, time, cost and the like do not provide enough data. In general, for an algorithm or a technology, along with the increase of the size of a data set, the relationship between each feature and a sample is more clearly described. In addition, a large amount of sample data is needed for training the deep learning model, and the sample data (especially fault samples) are very scarce in industrial practice. Therefore, the method has high application value on how to solve the problem of predicting the time sequence data of the industrial process with insufficient training samples.
Disclosure of Invention
The invention aims to provide a prediction method for time series data of an industrial process.
The purpose of the invention is realized by the following technical scheme:
the method comprises the following steps: preparing a training data set
Taking off-line data in the historical operation process of the industrial system as original data X of time sequence data prediction, constructing an improved generation countermeasure network model GRU-BEGAN, and generating a following X score through the improved generation countermeasure network model GRU-BEGANDistributing similar number sets X ', combining X and X' to obtain a data set for training a prediction network, namely X new ＝{X，X′}；
Step two: constructing a Deep long-short term memory network (Deep-LSTM) model, and increasing the number of LSTM units and the number of network layers;
step three: training Deep-LSTM network model
3.1 initialization parameters
Inputting batch Batchsize of training data, training iteration times epoch, time sequence length sl, predicted step number lambda (sl is larger than or equal to lambda), and learning rate alpha, and when verifying that a data set loss value val _ loss is not reduced in continuous theta iterations, changing the learning rate into alpha multiplied by beta, wherein the minimum value is not lower than k, theta, beta and k are hyper-parameters, and beta is more than 0 and less than 1;
3.2 setting the loss function
Modifying the loss function W-MAE by adjusting the weight coefficient of the prediction step length, and improving the prediction precision;
3.3 training Deep-LSTM
Processing training data into a plurality of subsets with the time sequence length of sl, randomly disordering the normalized subsets with the length of sl, inputting a data set into a built Deep-LSTM model for training, and updating weight parameters of Deep-LSTM according to an optimization algorithm;
step four: inputting the test data set into Deep-LSTM after training, predicting lambda unknown data through the network, calculating the error between the predicted value and the true value, and verifying the prediction effect and accuracy of the network.
Further, the LSTM state update in step two satisfies the following formula:
f t ＝σ(W f [h t-1 ，z t ]+b f )，
i t ＝σ(W i [h t-1 ，z t ]+b i )，
o t ＝σ(W o [h t-1 ，z t ]+b o )，
l t ＝tanh(W l [h t-1 ，z t ]+b l )，
c t ＝f t ·c t-1 +i t ·l t ，
h t ＝o t ·tanh(c t )
wherein, f t 、i t 、o t Respectively representing the output values of the forgetting gate, the input gate and the output gate,/, respectively t Indicating the state of the input cell, c t Representing hidden layer neuron states, [ h ] t-1 ，z t ]A concatenated vector, W, representing the output at time t-1 and the output at time t f 、W i 、W o 、W l Representing the respective weight matrix, b f 、b i 、b o 、b l Represents each bias term, σ () is an activation function, and tanh () is a hyperbolic tangent activation function.
Further, the modified loss function W-MAE in step 3.2 is given by the following formula:
Figure BDA0003561586450000031
wherein, y i Representing the true value of the sample, f (x) i ) Representing the predicted value of the sample, ω i Representing the weight coefficients.
Further, the formula in step 3.3 for updating the weight parameter of Deep-LSTM according to the optimization algorithm is as follows:
Figure BDA0003561586450000032
m t ＝β 1 m t-1 +(1-β 1 )g t
Figure BDA0003561586450000033
Figure BDA0003561586450000034
wherein, theta t-1 Representing the parameter to be updated, g t The gradient of the random objective function is represented,
Figure BDA0003561586450000035
denotes performing a gradient operation with respect to θ, f t () Representing the objective function, m t Representing an estimate of the first moment, v t Representing the partial second moment estimate, beta 1 、β 2 Exponential decay rate representing moment estimation, alpha represents learning rate, and initial value m 0 ＝0，v 0 0, epsilon > 0 and close to 0.
The invention has the beneficial effects that:
aiming at the problems of insufficient fault data samples and prediction accuracy in time sequence data prediction, the invention utilizes an improved GRU-BEGAN generation countermeasure network model to generate artificial samples to expand an original data set, and utilizes a Deep-LSTM model to predict the time sequence data of the industrial process. The method has the advantages that: (1) an LSTM network based on a recurrent neural network is used, the number of LSTM units is increased, and the prediction performance of the whole network model is improved; (2) aiming at the problems of insufficient training data sample size and poor model fitting effect in time series data prediction, an improved GRU-BEGAN generation countermeasure model is combined with a Deep-LSTM model to generate an artificial sample size required by prediction network training, and the model learning capability is improved to improve the parameter prediction precision; (3) the method uses the proposed W-MAE as a loss function of the network, realizes the multi-step prediction effect of the time sequence data, and further improves the accuracy of network prediction.
Drawings
FIG. 1 is a schematic diagram of the architecture of Deep-LSTM long and short term memory neural network neurons of the present invention;
FIG. 2 is a diagram of a GRU-BEGAN generation countermeasure network framework according to the present invention;
FIGS. 3(a) -3 (d) are graphs comparing the results of five-step predictions using Deep-LSTM + GRU-BEGAN (W-MAE) in accordance with the present invention;
FIGS. 4(a) -4 (b) are graphs comparing predicted results with real data according to the present invention;
FIGS. 5(a) -5 (b) are graphs of the relative error between the predicted result and the actual data according to the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
The invention provides an improved long-short term memory network algorithm Deep-LSTM based on industrial process data expansion, which is based on a long-short term memory network LSTM algorithm and fully considers model training deficiency caused by less time sequence data samples of an industrial process. And constructing an improved boundary balance generation confrontation network model GRU-BEGAN on the basis of the boundary balance generation confrontation network model, and obtaining generated artificial data to enhance the original industrial data set. Aiming at the defect that the LSTM model carries out long-term memory and transmission on time series data characteristic information, the number of layers of the network and the number of LSTM units are increased, on the basis of ensuring that the time series characteristic and the space characteristic of sampling data are extracted to the maximum degree, the long-term dependence of the network structure is further enhanced, and the industrial fault time series is effectively processed. On the loss function, an improved loss function W-MAE based on the mean absolute error loss is constructed. The expanded data set is trained using a modified Deep-LSTM network algorithm to achieve multi-step prediction of industrial process data.
The invention provides a prediction method of industrial process time series data based on an improved long-short term memory neural network (Deep-LSTM), which comprises the following steps:
the method comprises the following steps: the method comprises the steps that offline data in the historical operation process of an industrial system are used as original data X of fault prediction, when the data volume of X is m, an improved generation confrontation network model GRU-BEGAN is constructed, the generation G comprises a discriminator D and a generator G, the generator G is composed of a gated cyclic neural network GRU model, and the discriminator D is a self-encoder model. Generating a number set X 'similar to X distribution through an improved generation countermeasure network model GRU-BEGAN, combining X and X' to obtain a data set for training a prediction network, namely X new ＝{X，X′}；
Step two: constructing a Deep long-short term memory network (Deep-LSTM) model, increasing the number of LSTM units to enhance the capability of feature extraction and information memory, and deepening the layer number of the network, wherein the LSTM state update meets the following formula:
f t ＝σ(W f [h t-1 ，z t ]+b f )，
i t ＝σ(W i [h t-1 ，z t ]+b i )，
o t ＝σ(W o [h t-1 ，z t ]+b o )，
l t ＝tanh(W l [h t-1 ，z t ]+b l )，
c t ＝f t ·c t-1 +i t ·l t ，
h t ＝o t ·tanh(c t )
wherein f is t 、i t 、o t Respectively representing the outputs of the forgetting gate, the input gate and the output gate, | t Indicating the state of the input cell, c t Representing hidden layer neuron states, [ h ] t-1 ，z t ]A concatenated vector, W, representing the output at time t-1 and the output at time t f 、W i 、W o 、W l Representing the respective weight matrix, b f 、b i 、b o 、b l Representing each offset term, wherein sigma () is a sigmoid activation function, and tanh () is a hyperbolic tangent activation function;
step three: Deep-LSTM network model constructed by training
Based on Deep-LSTM long-short term memory artificial neural network, processing training data into a plurality of subsets with the time sequence length of sl, randomly disordering the normalized subsets with the length of sl, using 80% of the data sets as training data sets, and using the rest 20% of the data sets as verification data sets. And inputting the training data set and the verification data set into the built Deep-LSTM model for training, and reversely calculating the error term value of each neuron. Finally, the gradient of each weight is calculated by the corresponding error term. Updating the weight parameters of Deep-LSTM according to Adam optimization algorithm:
Figure BDA0003561586450000051
m t ＝β 1 m t-1 +(1-β 1 )g t
Figure BDA0003561586450000052
Figure BDA0003561586450000053
wherein, theta t-1 Representing the parameter to be updated, g t The gradient of the random objective function is represented,
Figure BDA0003561586450000054
denotes performing a gradient operation with respect to θ, f t () Representing the objective function, m t Representing an estimate of the first moment, v t Representing the partial second moment estimate, beta 1 、β 2 Exponential decay rate representing moment estimation, alpha represents learning rate, and initial value m 0 ＝0，v 0 0, epsilon > 0 and close to 0;
the specific training comprises the following steps:
(1) initializing parameters: inputting batch Batchsize of training data, training iteration times epoch, time sequence length sl, predicted step size number lambda (sl is more than or equal to lambda), and learning rate alpha, wherein when val _ loss continues for theta iterations without reduction, the learning rate is changed to alpha multiplied by beta, but the minimum is not lower than k, wherein theta, beta and k are hyper-parameters, beta is more than 0 and less than 1, and k is 0.000001 in the embodiment;
(2) training Deep-LSTM long and short term memory neural network: and taking the processed data set as the input of the Deep-LSTM network, and calculating a loss function W-MAE of the network:
Figure BDA0003561586450000061
wherein, y i Representing the true value of the sample, f (x) i ) Watch (CN)Predicted value of sample, ω i Representing a weight coefficient;
(3) training Deep-LSTM according to the mode until the W-MAE loss function value is not reduced any more and the network model is converged;
step four: and (3) predicting parameters according to the trained Deep-LSTM + GRU-BEGAN (W-MAE) long-short term memory neural network: inputting the test data set into the trained model, predicting lambda unknown data through the network, calculating the average percentage error, relative error, mean square error, average absolute error and root mean square error of the predicted value and the true value, and verifying the prediction effect and accuracy of the network.
In combination with a specific parameter embodiment, the embodiment adopts a public data set TennesseeEastman (TE) chemical process as an experimental object, and the TennesseeEastman process is a typical chemical process composed of a reactor, a condenser, a vapor-liquid separator, a circulating compressor, a stripping tower and other devices and proposed by eastman chemical company of the united states. The TE chemical process can simulate 21 industrial production process faults and is mainly divided into 6 types including step, random variable, slow drift, sticking, constant position, 5 unknown faults and the like. The process has 53 observed variables, including 41 (XMEAS (1) -XMEAS (41)) measured variables and 12 (XMV (1) -XMV (12)) manipulated variables.
Analysis of experimental results of Tennessee Eastman (TE) failure prediction:
the experimental data set comprises a training set (d01-d21) and a testing set (d01_ te-d21_ te) of 22 working conditions, wherein the working condition of a d01 step fault is taken as the training set of the experiment, d01_ te is taken as the testing set of the experiment, and XMEAS (3) and XMEAS (4) in the working condition are taken as predicted target parameters. According to the experimental steps, firstly, a selected fault data set is expanded by using an improved generation countermeasure network model GRU-BEGAN, 480 samples in an original data set are expanded to 960 for model training so as to strengthen the learning capacity of a prediction network, and then training is carried out according to a built Deep-LSTM network model and a training mode, wherein the weight coefficient of an improved loss function W-MAE provided by the invention is set as: w1 to W5 are 0.1, 0.3, and 0.4.
Inputting a fault data time sequence with the length of 25 into a trained model, and realizing 5-step prediction through a network, namely predicting 5 unknown data points, wherein in the graphs shown in (a) to (d) in fig. 3, the results of predicting 5 sample points by using Deep-LSTM + GRU-BEGAN (W-MAE) show a comparison graph, wherein a square represents real data, a round point represents predicted data, and the 5 predicted data points are close to the data values of the original 5 data points, the ascending and descending trends are close, but errors of a certain size exist through comparison of a broken line trend and corresponding data points. By processing the entire d01_ te dataset into several 25-long time series and inputting into the prediction network, a complete prediction curve similar to the original data set can be obtained, and fig. 4(a) -4 (b) are comparison graphs of the prediction result and the real data, wherein the dotted line is a time series data curve of the raw data, the solid line is a time series data curve of the raw data using Deep-LSTM + GRU-BEGAN (W-MAE), FIGS. 5(a) -5 (b) are graphs showing the relative error between the predicted result and the actual data, the comparison of the whole prediction curve shows that the proposed prediction method can achieve good prediction on the parameter trend when the industrial fault occurs, the curve trend change is close to the real time sequence data curve change, however, the fluctuation caused by some small noises can not be accurately identified, and the obtained prediction result is approximate to the numerical value of the original data through the result of relative error.
Figure BDA0003561586450000071
TABLE 1
Table 1 shows a failure prediction result list of the embodiment of the present invention, under the same sampling time series data condition and the same training parameters, the parameter XMEAS (4) of the d01 failure condition is predicted by respectively applying the full-connection network, GRU, LSTM, Deep-LSTM + GRU-BEGAN (W-MAE) models, and the model evaluation index mean percentage error (MAPE), Mean Square Error (MSE), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE) are used as 4 indexes for measuring the quality of the prediction data, and when the 4 evaluation indexes are lower, the error between the prediction result and the real data is smaller, so the prediction precision of 4 analysis Deep-LSTM + GRU-BEGAN (W-MAE) is better than that of the other five prediction models applied in the present invention.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (4)

1. A prediction method for industrial process time series data is characterized in that:
the method comprises the following steps: preparing a training data set
Taking offline data in the historical operation process of the industrial system as original data X of time sequence data prediction, constructing an improved generation confrontation network model GRU-BEGAN, generating a number set X 'similar to X distribution through the improved generation confrontation network model GRU-BEGAN, merging X and X', and obtaining a data set for training a prediction network, namely X new ＝{X,X′}；
Step two: constructing a Deep long-short term memory network (Deep-LSTM) model, and increasing the number of LSTM units and the number of network layers;
step three: training Deep-LSTM network model
3.1 initialization parameters
Inputting batch Batchsize of training data, training iteration times epoch, time sequence length sl, predicted step number lambda (sl is larger than or equal to lambda), and learning rate alpha, and when verifying that a data set loss value val _ loss is not reduced in continuous theta iterations, changing the learning rate into alpha multiplied by beta, wherein the minimum value is not lower than k, theta, beta and k are hyper-parameters, and beta is more than 0 and less than 1;
3.2 setting the loss function
Modifying the loss function W-MAE by adjusting the weight coefficient of the prediction step length, and improving the prediction precision;
3.3 training Deep-LSTM
Processing training data into a plurality of subsets with the time sequence length of sl, randomly disordering the normalized subsets with the length of sl, inputting a data set into a built Deep-LSTM model for training, and updating weight parameters of Deep-LSTM according to an optimization algorithm;
step four: inputting the test data set into Deep-LSTM after training, predicting lambda unknown data through the network, calculating the error between the predicted value and the true value, and verifying the prediction effect and accuracy of the network.
2. The method of claim 1, wherein the LSTM state update in step two satisfies the following equation:
f t ＝σ(W f [h t-1 ,z t ]+b f )，
i t ＝σ(W i [h t-1 ,z t ]+b i )，
o t ＝σ(W o [h t-1 ,z t ]+b o )，
l t ＝tanh(W l [h t-1 ,z t ]+b l )，
c t ＝f t ·c t-1 +i t ·l t ，
h t ＝o t ·tanh(c t )
wherein f is t 、i t 、o t Respectively representing the output values of the forgetting gate, the input gate and the output gate,/, respectively t Indicating the state of the input cell, c t Representing hidden layer neuron states, [ h ] t-1 ,z t ]A concatenated vector, W, representing the output at time t-1 and the output at time t f 、W i 、W o 、W l Representing the respective weight matrix, b f 、b i 、b o 、b l Represents each bias term, σ () is an activation function, and tanh () is a hyperbolic tangent activation function.
3. A prediction method for industrial process timing data according to claim 1, wherein the modified loss function W-MAE formula in step 3.2 is as follows:
Figure FDA0003561586440000021
wherein, y i Representing the true value of the sample, f (x) i ) Representing the predicted value of the sample, ω i Representing the weight coefficients.
4. A prediction method for industrial process timing data according to claim 1, wherein the formula for updating the weight parameter of Deep-LSTM according to the optimization algorithm in step 3.3 is as follows:
Figure FDA0003561586440000022
m t ＝β 1 m t-1 +(I-β 1 )g t
Figure FDA0003561586440000023
Figure FDA0003561586440000024
wherein, theta t-1 Representing the parameter to be updated, g t The gradient of the random objective function is represented,
Figure FDA0003561586440000025
denotes performing a gradient operation with respect to θ, f t () Representing the objective function, m t Representing an estimate of the first moment, v t Representing the partial second moment estimate, beta 1 、β 2 Exponential decay rate representing moment estimation, alpha represents learning rate, and initial value m 0 ＝0，v 0 0, epsilon > 0 and close to 0.
CN202210295183.9A 2022-03-23 2022-03-23 Prediction method for industrial process time sequence data Pending CN115062528A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210295183.9A CN115062528A (en) 2022-03-23 2022-03-23 Prediction method for industrial process time sequence data

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210295183.9A CN115062528A (en) 2022-03-23 2022-03-23 Prediction method for industrial process time sequence data

Publications (1)

Publication Number Publication Date
CN115062528A true CN115062528A (en) 2022-09-16

Family

ID=83197203

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210295183.9A Pending CN115062528A (en) 2022-03-23 2022-03-23 Prediction method for industrial process time sequence data

Country Status (1)

Country Link
CN (1) CN115062528A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116663338A (en) * 2023-08-02 2023-08-29 中国电子信息产业集团有限公司第六研究所 Simulation analysis method, device, equipment and medium based on similar calculation example
CN116700168A (en) * 2023-06-02 2023-09-05 中国五洲工程设计集团有限公司 Virtual-real synchronization method and system for production line

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116700168A (en) * 2023-06-02 2023-09-05 中国五洲工程设计集团有限公司 Virtual-real synchronization method and system for production line
CN116663338A (en) * 2023-08-02 2023-08-29 中国电子信息产业集团有限公司第六研究所 Simulation analysis method, device, equipment and medium based on similar calculation example
CN116663338B (en) * 2023-08-02 2023-10-20 中国电子信息产业集团有限公司第六研究所 Simulation analysis method, device, equipment and medium based on similar calculation example

Similar Documents

Publication Publication Date Title
CN112241608B (en) Lithium battery life prediction method based on LSTM network and transfer learning
CN112036084B (en) Similar product life migration screening method and system
Fan et al. A novel machine learning method based approach for Li-ion battery prognostic and health management
CN111860982A (en) Wind power plant short-term wind power prediction method based on VMD-FCM-GRU
Yang et al. A novel self-constructing radial basis function neural-fuzzy system
CN115062528A (en) Prediction method for industrial process time sequence data
CN113095550B (en) Air quality prediction method based on variational recursive network and self-attention mechanism
CN110059867B (en) Wind speed prediction method combining SWLSTM and GPR
CN109766745B (en) Reinforced learning tri-state combined long-time and short-time memory neural network system and training and predicting method
CN113205233B (en) Lithium battery life prediction method based on wolf algorithm and multi-core support vector regression
CN112734002B (en) Service life prediction method based on data layer and model layer joint transfer learning
Ayodeji et al. Causal augmented ConvNet: A temporal memory dilated convolution model for long-sequence time series prediction
Suryo et al. Improved time series prediction using LSTM neural network for smart agriculture application
CN113344288A (en) Method and device for predicting water level of cascade hydropower station group and computer readable storage medium
Wang et al. A novel restricted Boltzmann machine training algorithm with fast Gibbs sampling policy
CN115796327A (en) Wind power interval prediction method based on VMD (vertical vector decomposition) and IWOA-F-GRU (empirical mode decomposition) -based models
CN116304587A (en) Rolling bearing degradation trend prediction method based on CAE and AGRU
CN114692507A (en) Counting data soft measurement modeling method based on stacking Poisson self-encoder network
Bi et al. Remaining life prediction for aircraft turbine engines based on LSTM-RNN-HMM–APPROACH
CN112183814A (en) Short-term wind speed prediction method
CN114282614B (en) Medium-long runoff prediction method for optimizing CNN-GRU based on random forest and IFDA
CN113537539B (en) Multi-time-step heat and gas consumption prediction model based on attention mechanism
CN116303786B (en) Block chain financial big data management system based on multidimensional data fusion algorithm
Ma et al. Multi-Step Prediction Method of Chemical Process Variables Based on KPCA and GRU Neural Networks
Yu et al. Prediction Method for Rolling Bearing Remaining Useful Life Based on the RFHD Model and ALSTM

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