CN108197743A - A kind of prediction model flexible measurement method based on deep learning - Google Patents
A kind of prediction model flexible measurement method based on deep learning Download PDFInfo
- Publication number
- CN108197743A CN108197743A CN201711498840.5A CN201711498840A CN108197743A CN 108197743 A CN108197743 A CN 108197743A CN 201711498840 A CN201711498840 A CN 201711498840A CN 108197743 A CN108197743 A CN 108197743A
- Authority
- CN
- China
- Prior art keywords
- data
- historical data
- variable
- deep learning
- model
- 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
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/04—Manufacturing
-
- 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 kind of prediction model flexible measurement method based on deep learning, including：Obtain historical data；Historical data is carried out according to time window regular；Using Stationary Wavelet Transform extract it is regular after historical data multi-scale information；Multi-scale information current observable variable data corresponding with each time point are combined, to form sample data set；Training set and test set are formed according to sample data set；The depth model with attention mechanism is trained and tested using training set and test set, to form complete model；Predicted value is obtained according to current considerable measured data and complete model.Feature measured directly is difficult to for some significant variables among chemical process, prediction model flexible measurement method provided by the invention based on deep learning realizes the accurate prediction to the Unobservable variable among chemical process, reference index is provided for subsequent Energy Efficiency Analysis, so as to improve production capacity and reduce energy consumption.
Description
Technical field
The present invention relates to deep learning technology field more particularly to a kind of prediction model hard measurement sides based on deep learning
Method.
Background technology
P-phthalic acid (Pure Terephthalic Acid, PTA) is as one of important chemical industry Organic Ingredients, quilt
It is widely used in the various aspects such as chemical fibre, light industry, is the important embodiment of national economy.Correspondingly, PTA process units
Energy consumption directly affects national economy, how to control the energy consumption in PTA production processes into the critical issue of Modernized Chemical Enterprises.
PTA solvent systems are a parts crucial in PTA productions, and the consumption of wherein acetic acid is the important of measurement PTA process units
Index is the primary evaluation means of assessment technique quality.Therefore, how to reduce the consumption of the acetic acid of tower top in solvent system into
Optimize the main target of PTA production processes.
The prior art proposes many methods for analyzing efficiency problem, for example, by DEA Method (Data
Envelopment Analysis, DEA) it be combined with each other with analytic hierarchy process (AHP) (Analytic Hierarchy Process, AHP)
For analyzing the efficiency of the energy, energy consumption is reduced to improve energy utilization rate to complete decision.However, the above method can not all be realized
Accurate prediction to acetic acid content among PTA production processes.
Invention content
To solve the above problems, the present invention provides a kind of prediction model flexible measurement method based on deep learning, at least portion
Decompose above-mentioned technical problem of determining.
For this purpose, the present invention provides a kind of prediction model flexible measurement method based on deep learning, including：
Obtain historical data；
It is regular to historical data progress according to time window, so that the historical data and current observable variable data
Form correspondence；
Using Stationary Wavelet Transform extract it is regular after historical data multi-scale information；
The multi-scale information current observable variable data corresponding with each time point are combined, to form sample
Notebook data collection；
Training set and test set are formed according to the sample data set；
The depth model with attention mechanism is trained and tested using the training set and the test set, with
Form complete model；
Predicted value is obtained according to current considerable measured data and the complete model.
Optionally, it is described using Stationary Wavelet Transform extract it is regular after historical data multi-scale information the step of wrap
It includes：
The data of each time point are decomposed using Stationary Wavelet Transform, are to obtain the approximate of multigroup different scale
Number and detail coefficients：
Wherein,WithFor kth group variable historical data i-stage decompose as a result, and
Optionally, it further includes：
The complete model is evaluated using root-mean-square error and average relative error；
The calculation formula of the root-mean-square error is：
The calculation formula of the average relative error is：
Wherein, yiFor the true output of i-th group of sample, prediThe predicted value of depth model for i-th group of sample,For yi
It is after renormalization as a result,For prediResult after renormalization.
Optionally, include after the step of acquisition historical data：
The historical data is normalized：
Wherein, historical dataxtRepresent all variable states of t moment Chemical Manufacture；X=
[x1..., xk], xkRepresent the state of kth variable；Represent that kth variable is all history numbers among this all segment datas
According to minimum value；Represent maximum value of the kth variable for all historical datas among this all segment datas.
The present invention has following advantageous effects：
Prediction model flexible measurement method provided by the invention based on deep learning includes：Obtain historical data；According to when
Between window the historical data is carried out it is regular so that the historical data and current observable variable data form correspondence；
Using Stationary Wavelet Transform extract it is regular after historical data multi-scale information；By the multi-scale information and each time
The corresponding current observable variable data of point are combined, to form sample data set；It is formed and instructed according to the sample data set
Practice collection and test set；The depth model with attention mechanism is trained and surveyed using the training set and the test set
Examination, to form complete model；Predicted value is obtained according to current considerable measured data and the complete model.For chemical process
Among some significant variables be difficult to feature measured directly, the prediction model hard measurement provided by the invention based on deep learning
Method realizes the accurate prediction to the Unobservable variable among chemical process, and providing reference for subsequent Energy Efficiency Analysis refers to
Mark, so as to improve production capacity and reduce energy consumption.
Description of the drawings
Fig. 1 is the flow chart of the prediction model flexible measurement method based on deep learning that the embodiment of the present invention one provides；
Fig. 2 is the network model schematic diagram that the embodiment of the present invention one provides；
Fig. 3 is the training result schematic diagram that the embodiment of the present invention one provides；
Fig. 4 is the prediction difference schematic diagram that the embodiment of the present invention one provides.
Specific embodiment
For those skilled in the art is made to more fully understand technical scheme of the present invention, the present invention is carried below in conjunction with the accompanying drawings
The prediction model flexible measurement method based on deep learning supplied is described in detail.
Embodiment one
Due to the promotion of calculated performance and the rise of computational intelligence, deep learning has obtained the more and more extensive concern of people
And research, and achieve extraordinary result in some fields.Based on convolutional neural networks (Convolutional Neural
Networks, CNN) various models all achieve good achievement in the various aspects of visual field.Exclusively for sequence data
The Recognition with Recurrent Neural Network (Recurrent Neural Networks, RNNs) of design sequence data prediction as speech analysis,
Natural language processing etc. all achieves good achievement.Many fields are all gone to analyze and be solved in the method for trial deep learning
Problem.Different sampling intervals and the method for sampling in time series, it will lead to time series data complicated and changeable in itself.Cause
This, there has been proposed the methods of multiscale analysis, obtain feature of the time series data under different scale and resolution ratio, quite
In the signal for time series data being resolved into different frequency, the most common multiscale analysis method used in practical operation for
Wavelet decomposition and empirical mode decomposition.
Industrial data measures to obtain to equipment in different time points, therefore industrial data is very typical
Sequence data.According to this feature, the present embodiment proposes a kind of time series predicting model based on deep learning, for complete
The accurate prediction of pairs of industrial data key variables.This implementation example first by industrial data carry out regular history of forming data with it is pre-
Then the corresponding form of measured data carries out multi-resolution decomposition with Stationary Wavelet Transform to historical data, historical data and currently
Observable variable combines to form entire sample, above-mentioned sample be used for Unobservable variable or " difficulty " observational variable into
Row prediction.In order to obtain complete model, all samples are carried out tissue by the present embodiment, form training set and test set to depth
Model is trained, finally new data is carried out with same method it is regular, decompose, most at last it is regular, decompose after new number
Prediction result is obtained according to depth model is put into.These prediction results provide reference index for subsequent Energy Efficiency Analysis, so as to improve
Production capacity and reduction energy consumption.
Fig. 1 is the flow chart of the prediction model flexible measurement method based on deep learning that the embodiment of the present invention one provides.Such as
Shown in Fig. 1, the prediction model flexible measurement method provided in this embodiment based on deep learning includes：Obtain historical data；According to
Time window is regular to historical data progress, so that the historical data forms corresponding pass with current observable variable data
System；Using Stationary Wavelet Transform extract it is regular after historical data multi-scale information；By the multi-scale information and each
Time point, corresponding current observable variable data were combined, to form sample data set；According to the sample data set shape
Into training set and test set；The depth model with attention mechanism is trained using the training set and the test set
And test, to form complete model；Predicted value is obtained according to current considerable measured data and the complete model.
The present embodiment obtains the industrial data of one section of continuous time firstxtRepresent that t moment chemical industry is given birth to
Produce all variable states of equipment；X=[x1..., xk], wherein xkRepresent the state of kth variable.In order to train depth model, this
Industrial data D is normalized in embodiment：
Wherein,Represent minimum value of the kth variable for all historical datas among this all segment datas,Table
Show maximum value of the kth variable for all historical datas among this all segment datas.
The present embodiment predicts current unobservable data with the industrial data of the past period, it is therefore desirable to
These time series datas are carried out with regular, the regular form for forming needs, that is, the form used for neural network.It is false
If the time span is n, the present embodiment uses the additional current observable variable data of industrial data at n group time points in the past, right
Current unobservable dataIt is predicted, wherein p is the mark of key variables for needing to predict.In order to facilitate nerve net below
The training of network needs data to be arranged, and is formed and current considerable measured data and current key variable are added per n groups industrial data
The corresponding form of data, the form are as follows：
Wherein,Represent t moment L observable variables.
The complexity of chemical industry equipment causes chemical engineering data to be also extremely complex, in order to reduce the complexity of data, needs
These data are carried out to simplify decomposition.The present embodiment using Stationary Wavelet Transform (Stationary Wavelet Transform,
SWT multi-resolution decomposition) is carried out to the historical information of industrial data, the data at each n time points are decomposed to obtain it is multigroup not
With the approximation coefficient and detail coefficients of scale：
Wherein,WithFor kth group variable historical data i-stage decompose as a result, and
Therefore, each group of variable can obtain p+1 row multi-scale informations, these multi-scale informations represent industrial data when different
Between tendency information under scale different frequency.
The present embodiment has set up a deep learning model dedicated for sequence information, and RNNs is used exclusively for processing sequence
The neural network model of column information, the current output of the neural network model is related with the output at front all time points, however
The information of whole prior time points is not needed to sometimes, but only that a part of information for paying special attention to prior time can
With.Therefore, the deep learning model that the present embodiment uses is that have the length of attention mechanism memory network (Attention in short-term
Long Short-Term Memory, Attention LSTM) model, it allows for that the chemical engineering data predicted is needed more to close in this way
Lie in the information of interest of front.Long memory network in short-term increases one on traditional RNNs forgets door will recall to control
Memory length, it is as follows in the computational methods of t moment：
ft=σ (Wf·[ht-1, xt]+bf)
it=σ (Wi·[ht-1, xt]+bi)
Ot=σ (Wo·[ht-1, xt]+bo)
ht=ot*tanh(Ct)
Wherein, W is weights, and b is biasing, and σ is activation primitive htOutput for t moment hidden layer.Attention mechanism is phase
Output has been contacted with correlated inputs when in increasing a weight to the input space for different inputs, being also achieved that
Come.
Wherein,α is a weight matrix, it represent which information it is emphasized that：
eij=a (si-1, hj)
Wherein, s is the output of the full articulamentum of a saving sequence information.
The present embodiment divides training set and test set is trained and tests to deep learning model, here using mean square error
Difference learns as error function for the error back propagation of neural network, and the mean square error is：
After training model, the present embodiment is according to the method tissue data of front, before then being carried out using depth model
Feedback calculates, by the result of acquisition by renormalization, the key variables information as predicted：
The present embodiment evaluates the complete model using root-mean-square error and average relative error, so as to model
As a result quality is evaluated.
The calculation formula of the root-mean-square error is：
The calculation formula of the average relative error is：
Wherein, yiFor the true output of i-th group of sample, prediThe predicted value of depth model for i-th group of sample,For yi
It is after renormalization as a result,For prediResult after renormalization.
The present embodiment using a PTA data instance illustrate technical solution provided in this embodiment specific implementation details and
Mode.The present embodiment has chosen 17 factors that can influence tower top acetic acid consumption, and the factor is as described in Table 1, then this reality
It is exactly to predict the consumption of acetic acid using this 17 variables to apply example.
Table 1 influences the factor of acetic acid consumption
This group of data one provided in this embodiment share the data at 173 time points, and each time point has corresponded to above
The consumption data of the measured value of 17 influence factors and the time point acetic acid.First, the present embodiment using formula (1) to data into
Row normalized so that the value of each variable of entire data is in [0,1].Then, the present embodiment utilizes above-mentioned side
Method is regular to the progress of these data, chooses n=10, that is, the historical data at 10 time points before utilizing is along with current 17
A observable data predict current significant variable acetic acid consumption to realize.It is important for historical information
The consumption of variable acetic acid is known, therefore historical information here is complete 18 groups of data.The acetic acid at current time disappears
Consumption is unknown, this needs is predicted, therefore there was only 17 variables for current time.
In order to obtain the data that can be used by neural network model, the present embodiment is by the data and length at each time point
The historical information spent for 10 is stacked so that each moment corresponds to the historical information at 10 time points, allows for so originally
Dimension size is the data that the data of [18] become that dimension size is [10,18].When the present embodiment uses formula (2) each
Between put 10 historical informations of upper every group of variable and carry out stationary wavelet decomposition, Stationary Wavelet Transform used herein is dmey small echos
Function, Decomposition order 1, that is to say, that each group of variable has all been broken down into the factor data that two groups of sizes are all 10, in total
There are 18 groups of variables.Present data dimension size becomes [10,36], and the observable variable dimension size at current time is [17,1],
These data are the reference data to give a forecast at the moment among the model.
By obtaining 163 groups of data after regular, in order to train neural network, entire data set be divided into training set and
Test set.Due to being sequence data, because without random manner is taken to divide data set, but simply 70 groups of numbers below
According to as test set, 93 groups of data of front are as training set.
Fig. 2 is the network model schematic diagram that the embodiment of the present invention one provides.As shown in Fig. 2, the present embodiment uses depth
Frame Keras is practised to build deep learning model, for current time, the historical information conduct at all 10 time points
Then 17 observable variables at above-mentioned input and current time are stitched together, then by the input of Attention LSTM
The last one full articulamentum is put into, the output of model is final output, also as needs the data predicted.Specific network ginseng
Number is as follows：
● input dimension：[None, 10,36]
● time stride：10
● FC1 node sizes：50
● LSTM node sizes：50
● the LSTM numbers of plies：1
● FC2 node sizes：100
● each layer activation primitive：tanh
● error function：Referring to the MSE of formula (3)
Finally, the training result and test result that the present embodiment obtains are as shown in table 2：
Table 2PTA network training results
Fig. 3 is the training result schematic diagram that the embodiment of the present invention one provides.As shown in figure 3, the present embodiment delineates in detail
Thin prediction result curve.Fig. 4 is the prediction difference schematic diagram that the embodiment of the present invention one provides.As shown in figure 4, the present embodiment is painted
The prediction difference of depth model is made.The present embodiment uses the model and extreme learning machine to verify the validity of model
(Extreme Learning Machine, ELM) network and multilayer perceptron (Multi-Layer Perceptron, MLP) net
Network carries out Comparative result, and judgment criteria is ARGE (%), and comparing result is as shown in table 3：
3 model comparing result of table
As shown in Table 3, the performance of complete model provided in this embodiment will be substantially better than MLP and ELM.Therefore, by this
Model can obtain a fairly accurate predicted value of acetic acid consumption among PTA techniques, which can be used as efficiency etc.
The estimation and differentiation of grade.For example, the acetic acid consumption of the 138th group of data display equipment is obtained according to Fig. 4 will be significantly greater than model
Anticipated output, show that energy utilization rate is relatively low at this time, need to be correspondingly improved measure to improve energy utilization rate, with reduce
Energy consumption level.
Prediction model flexible measurement method provided in this embodiment based on deep learning includes：Obtain historical data；According to
Time window is regular to historical data progress, so that the historical data forms corresponding pass with current observable variable data
System；Using Stationary Wavelet Transform extract it is regular after historical data multi-scale information；By the multi-scale information and each
Time point, corresponding current observable variable data were combined, to form sample data set；According to the sample data set shape
Into training set and test set；The depth model with attention mechanism is trained using the training set and the test set
And test, to form complete model；Predicted value is obtained according to current considerable measured data and the complete model.For Chemical Manufacture
Some significant variables among process are difficult to feature measured directly, the prediction model provided in this embodiment based on deep learning
Flexible measurement method realizes the accurate prediction to the Unobservable variable among chemical process, is provided for subsequent Energy Efficiency Analysis
Reference index, so as to improve production capacity and reduce energy consumption.
It is understood that the principle that embodiment of above is intended to be merely illustrative of the present and the exemplary implementation that uses
Mode, however the present invention is not limited thereto.For those skilled in the art, in the essence for not departing from the present invention
In the case of refreshing and essence, various changes and modifications can be made therein, these variations and modifications are also considered as protection scope of the present invention.
Claims (4)
1. a kind of prediction model flexible measurement method based on deep learning, which is characterized in that including：
Obtain historical data；
It is regular to historical data progress according to time window, so that the historical data is formed with current observable variable data
Correspondence；
Using Stationary Wavelet Transform extract it is regular after historical data multi-scale information；
The multi-scale information current observable variable data corresponding with each time point are combined, to form sample number
According to collection；
Training set and test set are formed according to the sample data set；
The depth model with attention mechanism is trained and tested using the training set and the test set, to be formed
Complete model；
Predicted value is obtained according to current considerable measured data and the complete model.
2. the prediction model flexible measurement method according to claim 1 based on deep learning, which is characterized in that the use
The step of multi-scale information of historical data after Stationary Wavelet Transform extraction is regular, includes：
The data of each time point are decomposed using Stationary Wavelet Transform, with obtain the approximation coefficient of multigroup different scale and
Detail coefficients：
Wherein,The historical data of kth group variable i-stage decompose as a result, and
3. the prediction model flexible measurement method according to claim 1 based on deep learning, which is characterized in that further include：
The complete model is evaluated using root-mean-square error and average relative error；
The calculation formula of the root-mean-square error is：
The calculation formula of the average relative error is：
Wherein, yiFor the true output of i-th group of sample, prediThe predicted value of depth model for i-th group of sampleFor yiAnti- normalizing
Result after changeFor prediResult after renormalization.
4. the prediction model flexible measurement method according to claim 1 based on deep learning, which is characterized in that the acquisition
Include after the step of historical data：
The historical data is normalized：
Wherein, historical dataxtRepresent all variable states of t moment Chemical Manufacture；X=[x1...,
xk], xkRepresent the state of kth variable；Represent among this all segment datas kth variable for all historical datas most
Small value；Represent maximum value of the kth variable for all historical datas among this all segment datas.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711498840.5A CN108197743A (en) | 2017-12-31 | 2017-12-31 | A kind of prediction model flexible measurement method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711498840.5A CN108197743A (en) | 2017-12-31 | 2017-12-31 | A kind of prediction model flexible measurement method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108197743A true CN108197743A (en) | 2018-06-22 |
Family
ID=62587815
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711498840.5A Pending CN108197743A (en) | 2017-12-31 | 2017-12-31 | A kind of prediction model flexible measurement method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108197743A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109034497A (en) * | 2018-08-31 | 2018-12-18 | 广东工业大学 | Prediction technique, system, medium and the equipment of polycrystalline reduction process energy consumption value |
| CN109142976A (en) * | 2018-09-10 | 2019-01-04 | 国网辽宁省电力有限公司电力科学研究院 | Cable fault examination method and device |
| CN109242140A (en) * | 2018-07-24 | 2019-01-18 | 浙江工业大学 | A kind of traffic flow forecasting method based on LSTM_Attention network |
| CN109409567A (en) * | 2018-09-17 | 2019-03-01 | 西安交通大学 | Complex device method for predicting residual useful life based on the double-deck shot and long term memory network |
| CN110033126A (en) * | 2019-03-14 | 2019-07-19 | 贵州大学 | Shot and long term memory network prediction technique based on attention mechanism and logistic regression |
| CN110378044A (en) * | 2019-07-23 | 2019-10-25 | 燕山大学 | Multiple Time Scales convolutional neural networks flexible measurement method based on attention mechanism |
| CN111428906A (en) * | 2020-02-17 | 2020-07-17 | 浙江大学 | Industrial boiler steam quantity prediction method based on image transformation |
| CN111814101A (en) * | 2020-07-10 | 2020-10-23 | 北京无线电测量研究所 | Flight path prediction method and system and electronic equipment |
| CN112001115A (en) * | 2020-07-17 | 2020-11-27 | 西安理工大学 | Soft measurement modeling method of semi-supervised dynamic soft measurement network |
| CN113485261A (en) * | 2021-06-29 | 2021-10-08 | 西北师范大学 | CAEs-ACNN-based soft measurement modeling method |
| CN113960925A (en) * | 2021-08-30 | 2022-01-21 | 中科苏州微电子产业技术研究院 | Building energy consumption control method and device based on artificial intelligence |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101118593A (en) * | 2007-09-04 | 2008-02-06 | 西安电子科技大学 | Texture image classification method based on SWBCT |
| CN102385706A (en) * | 2011-10-14 | 2012-03-21 | 浙江大学 | Soft sensing method based on model learning and used in pure terephthalic acid (PTA) production |
| CN103729687A (en) * | 2013-12-18 | 2014-04-16 | 国网山西省电力公司晋中供电公司 | Electricity price forecasting method based on wavelet transform and neural network |
| CN106447103A (en) * | 2016-09-26 | 2017-02-22 | 河海大学 | Deep learning based QoS prediction method of Web service |
| US20170091615A1 (en) * | 2015-09-28 | 2017-03-30 | Siemens Aktiengesellschaft | System and method for predicting power plant operational parameters utilizing artificial neural network deep learning methodologies |
| CN107292383A (en) * | 2017-07-06 | 2017-10-24 | 郑保宁 | The variation water quality interval prediction method being combined based on deep learning algorithm with MILP |
| CN107341462A (en) * | 2017-06-28 | 2017-11-10 | 电子科技大学 | A kind of video classification methods based on notice mechanism |
-
2017
- 2017-12-31 CN CN201711498840.5A patent/CN108197743A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101118593A (en) * | 2007-09-04 | 2008-02-06 | 西安电子科技大学 | Texture image classification method based on SWBCT |
| CN102385706A (en) * | 2011-10-14 | 2012-03-21 | 浙江大学 | Soft sensing method based on model learning and used in pure terephthalic acid (PTA) production |
| CN103729687A (en) * | 2013-12-18 | 2014-04-16 | 国网山西省电力公司晋中供电公司 | Electricity price forecasting method based on wavelet transform and neural network |
| US20170091615A1 (en) * | 2015-09-28 | 2017-03-30 | Siemens Aktiengesellschaft | System and method for predicting power plant operational parameters utilizing artificial neural network deep learning methodologies |
| CN106447103A (en) * | 2016-09-26 | 2017-02-22 | 河海大学 | Deep learning based QoS prediction method of Web service |
| CN107341462A (en) * | 2017-06-28 | 2017-11-10 | 电子科技大学 | A kind of video classification methods based on notice mechanism |
| CN107292383A (en) * | 2017-07-06 | 2017-10-24 | 郑保宁 | The variation water quality interval prediction method being combined based on deep learning algorithm with MILP |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109242140A (en) * | 2018-07-24 | 2019-01-18 | 浙江工业大学 | A kind of traffic flow forecasting method based on LSTM_Attention network |
| CN109034497A (en) * | 2018-08-31 | 2018-12-18 | 广东工业大学 | Prediction technique, system, medium and the equipment of polycrystalline reduction process energy consumption value |
| CN109142976A (en) * | 2018-09-10 | 2019-01-04 | 国网辽宁省电力有限公司电力科学研究院 | Cable fault examination method and device |
| CN109409567B (en) * | 2018-09-17 | 2022-03-08 | 西安交通大学 | Complex equipment residual life prediction method based on double-layer long-short term memory network |
| CN109409567A (en) * | 2018-09-17 | 2019-03-01 | 西安交通大学 | Complex device method for predicting residual useful life based on the double-deck shot and long term memory network |
| CN110033126A (en) * | 2019-03-14 | 2019-07-19 | 贵州大学 | Shot and long term memory network prediction technique based on attention mechanism and logistic regression |
| CN110378044B (en) * | 2019-07-23 | 2021-06-11 | 燕山大学 | Multi-time scale convolution neural network soft measurement method based on attention mechanism |
| CN110378044A (en) * | 2019-07-23 | 2019-10-25 | 燕山大学 | Multiple Time Scales convolutional neural networks flexible measurement method based on attention mechanism |
| CN111428906A (en) * | 2020-02-17 | 2020-07-17 | 浙江大学 | Industrial boiler steam quantity prediction method based on image transformation |
| CN111428906B (en) * | 2020-02-17 | 2023-05-09 | 浙江大学 | Industrial boiler steam volume prediction method based on image transformation |
| CN111814101A (en) * | 2020-07-10 | 2020-10-23 | 北京无线电测量研究所 | Flight path prediction method and system and electronic equipment |
| CN112001115A (en) * | 2020-07-17 | 2020-11-27 | 西安理工大学 | Soft measurement modeling method of semi-supervised dynamic soft measurement network |
| CN112001115B (en) * | 2020-07-17 | 2024-04-02 | 西安理工大学 | Soft measurement modeling method of semi-supervised dynamic soft measurement network |
| CN113485261A (en) * | 2021-06-29 | 2021-10-08 | 西北师范大学 | CAEs-ACNN-based soft measurement modeling method |
| CN113485261B (en) * | 2021-06-29 | 2022-06-28 | 西北师范大学 | CAEs-ACNN-based soft measurement modeling method |
| CN113960925A (en) * | 2021-08-30 | 2022-01-21 | 中科苏州微电子产业技术研究院 | Building energy consumption control method and device based on artificial intelligence |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108197743A (en) | A kind of prediction model flexible measurement method based on deep learning | |
| CN108510741B (en) | Conv1D-LSTM neural network structure-based traffic flow prediction method | |
| CN108197648A (en) | A kind of Fault Diagnosis Method of Hydro-generating Unit and system based on LSTM deep learning models | |
| CN108647643B (en) | Packed tower flooding state online identification method based on deep learning | |
| Bernal et al. | Financial market time series prediction with recurrent neural networks | |
| CN105243398A (en) | Method of improving performance of convolutional neural network based on linear discriminant analysis criterion | |
| CN109635245A (en) | A kind of robust width learning system | |
| CN105740984A (en) | Product concept performance evaluation method based on performance prediction | |
| CN111340110B (en) | Fault early warning method based on industrial process running state trend analysis | |
| CN111768000A (en) | Industrial process data modeling method for online adaptive fine-tuning deep learning | |
| Zhao et al. | On-line least squares support vector machine algorithm in gas prediction | |
| CN108628164A (en) | A kind of semi-supervised flexible measurement method of industrial process based on Recognition with Recurrent Neural Network model | |
| CN111122162A (en) | Industrial system fault detection method based on Euclidean distance multi-scale fuzzy sample entropy | |
| CN111222798B (en) | Complex industrial process key index soft measurement method | |
| CN114239397A (en) | Soft measurement modeling method based on dynamic feature extraction and local weighted deep learning | |
| CN114596726A (en) | Parking position prediction method based on interpretable space-time attention mechanism | |
| CN109033524A (en) | A kind of chemical process concentration variable On-line Estimation method based on robust mixed model | |
| Yuan et al. | A novel hybrid approach to mooring tension prediction for semi-submersible offshore platforms | |
| CN116842358A (en) | Soft measurement modeling method based on multi-scale convolution and self-adaptive feature fusion | |
| CN116662925A (en) | Industrial process soft measurement method based on weighted sparse neural network | |
| Wang | A new variable selection method for soft sensor based on deep learning | |
| CN116451118B (en) | Deep learning-based radar photoelectric outlier detection method | |
| CN114298183B (en) | Intelligent recognition method for flight actions | |
| Zhan et al. | Long Short-term Memory modeling method with monotonicity analysis as constraints base on Spearman coefficient | |
| CN114580798B (en) | Device point location prediction method and system based on transformer |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180622 |
|
| RJ01 | Rejection of invention patent application after publication |