CN110428053A - A kind of steam produces the dynamic prediction method of consumption - Google Patents
A kind of steam produces the dynamic prediction method of consumption Download PDFInfo
- Publication number
- CN110428053A CN110428053A CN201910731785.2A CN201910731785A CN110428053A CN 110428053 A CN110428053 A CN 110428053A CN 201910731785 A CN201910731785 A CN 201910731785A CN 110428053 A CN110428053 A CN 110428053A
- Authority
- CN
- China
- Prior art keywords
- steam
- data
- consumption
- prediction
- node
- 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
- 238000000034 method Methods 0.000 title claims abstract description 44
- 238000004519 manufacturing process Methods 0.000 claims abstract description 89
- 238000010606 normalization Methods 0.000 claims abstract description 32
- 238000012549 training Methods 0.000 claims abstract description 25
- 238000013528 artificial neural network Methods 0.000 claims abstract description 21
- 230000007787 long-term memory Effects 0.000 claims abstract description 19
- 238000005457 optimization Methods 0.000 claims abstract description 14
- 238000012360 testing method Methods 0.000 claims abstract description 10
- 238000004364 calculation method Methods 0.000 claims description 43
- 239000011159 matrix material Substances 0.000 claims description 33
- 230000006870 function Effects 0.000 claims description 20
- 238000012545 processing Methods 0.000 claims description 10
- 230000000644 propagated effect Effects 0.000 claims description 8
- 238000012423 maintenance Methods 0.000 claims description 7
- 238000012546 transfer Methods 0.000 claims description 6
- 230000007774 longterm Effects 0.000 claims description 4
- 210000004218 nerve net Anatomy 0.000 claims description 2
- 238000006467 substitution reaction Methods 0.000 claims description 2
- 230000015654 memory Effects 0.000 claims 1
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical compound [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 description 20
- 229910000831 Steel Inorganic materials 0.000 description 12
- 239000000571 coke Substances 0.000 description 12
- 239000010959 steel Substances 0.000 description 12
- 238000010791 quenching Methods 0.000 description 11
- 230000000171 quenching effect Effects 0.000 description 11
- 229910052742 iron Inorganic materials 0.000 description 10
- 230000008569 process Effects 0.000 description 6
- 230000008859 change Effects 0.000 description 4
- 230000001186 cumulative effect Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 239000002918 waste heat Substances 0.000 description 4
- 239000007789 gas Substances 0.000 description 3
- 241000208340 Araliaceae Species 0.000 description 2
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 2
- 235000003140 Panax quinquefolius Nutrition 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 235000008434 ginseng Nutrition 0.000 description 2
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 1
- 239000003034 coal gas Substances 0.000 description 1
- 238000005034 decoration Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000013468 resource allocation Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000005245 sintering Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/903—Querying
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/906—Clustering; Classification
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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"
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Data Mining & Analysis (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Game Theory and Decision Science (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Development Economics (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides a kind of dynamic prediction method of steam production consumption comprising: normalization set is divided into training dataset and test data set；It is trained training dataset input shot and long term Memory Neural Networks prediction model to form training result data；The parameter in prediction model is updated according to training result data；By test data set for being predicted to obtain prediction result in prediction model, the prediction model of the corresponding optimization of the smallest prediction result of Select Error is as optimum prediction model；Consumption data are produced using the steam of optimum prediction model prediction time interval to be predicted.Influence of the equipment working condition difference to steam generation and consumption is comprehensively considered, it establishes based on the steam production of equipment multi-state operation and consumption dynamic prediction method and solves the steam forecasting problem under the complex working conditions that the present invention is previously mentioned, more accurately predict the steam consumption of the steam production of each equipment and each production user in the future scheduling period.
Description
Technical field
The invention belongs to steam predictions and the dynamic prediction method that consumption is produced using field more particularly to a kind of steam.
Background technique
In China, steel and iron industry is the basic pillar industry of national economy, is resource, energy intensive industry, and
The emphasis of energy-saving and emission-reduction work.The steam of iron and steel enterprise is the secondary energy sources of enterprise-essential.Steam prediction is the weight of vapour system
Component part is wanted, the yield and consumption of steam are predicted, to carry out subsequent steam resource allocation and optimization tune
Degree provides foundation for the safety and steady operation, reduction release, raising gas utilization rate of coal gas system.Therefore, steam prediction becomes
One of the important link of iron and steel enterprise's energy-saving and emission-reduction.
Iron and steel enterprise's steam is to be generated in process of production by gas boiler, waste heat boiler, and be widely used in steel
Iron produces each process, and steam can also be used to generate electric power.The yield of steam refers to gas boiler, what waste heat boiler generated
Quantity of steam；Steam consumption refers to the steam flow of the main production process consumption of iron and steel enterprise.In actual production, since steel is raw
Dynamic change is presented in the supply and demand of the fluctuation of production, steam, and consumption is certain under normal production status, and it is larger to encounter steam production
The case where steam cannot balance, cause steam releasing.For rational management steam, steam pipe network is adjusted, it would be desirable in advance
Steam production and consumption are predicted.If can look-ahead steam state between supply and demand, steam will be increased substantially
Service efficiency reduces unnecessary release, saves the energy.
Steam dynamic prediction method is to be directed to dynamic prediction of a certain particular device under the conditions of specific operation mostly at present,
It such as can use successive Regression and neural network building steam prediction model predict the steam production of sintering process；It can be with
Iron and steel enterprise's steam production is predicted based on wavelet transformation and least square method supporting vector machine；Above method is all specific
Steam prediction under working condition, but since iron and steel enterprise's actual production process is by the human factors such as production plan and operating condition
Etc. uncertain factors influence, equipment working condition change frequently, above method equipment working condition change when be difficult to Accurate Prediction, cause
Steam prediction data accuracy is not high.
Summary of the invention
(1) technical problems to be solved
It is an object of the invention to propose that a kind of steam produces the dynamic prediction method of consumption, it is intended to solve existing to steel enterprise
Each equipment of vapour system is not comprehensively considered in industry steam production and the dynamic prediction of consumption by the shadow of different operating conditions
It rings, makes the problem of dynamic prediction result inaccuracy.
(2) technical solution
In order to achieve the above object, the present invention provides a kind of dynamic prediction method of steam production consumption, the main skill of use
Art scheme includes:
S1, the production for obtaining each equipment steam consume historical data, and production consumption historical data is normalized,
Normalization set is obtained, the production consumption historical data includes steam production historical data and steam consumption historical data；
S2, the multiple groups setup parameter for obtaining shot and long term Memory Neural Networks prediction model；
S3, normalization set is divided into training dataset and test data set；
S4, the shot and long term Memory Neural Networks that training dataset input multiple groups have been arranged to the setup parameter
Prediction model is trained by propagated forward algorithm and back-propagation algorithm, forms training result data；
S5, according to the training result data, using adaptability moments estimation algorithm, remember mind to the shot and long term is set to
Node matrix equation and bias vector through setup parameter described in the multiple groups in Network Prediction Model are updated, and obtain multiple groups optimization
Prediction model；
S6, it will be predicted to obtain prediction result in the prediction model of optimization described in test data set substitution multiple groups,
The error between the prediction result and actual value, the smallest prediction result of Select Error are calculated by error calculation formula
The prediction model of the corresponding optimization is as optimum prediction model；
S7, consumption data are produced using the steam of the optimum prediction model prediction time interval to be predicted, the steam produces consumption
Data include steam production data and steam consumption data.
Preferably, the S1 includes:
S11, the production for obtaining each equipment steam consume historical data, and the corresponding production for producing consumption historical data
Plan and maintenance plan, the production status of each equipment is determined according to the production plan and the maintenance plan；
S12, the operating point that each equipment is obtained according to the production status, the production consumption history according to the pair of operation points
Data are classified to obtain classification set, and are normalized to obtain the normalization to the data in the classification set
Set.
Preferably, the production consumption historical data according to the pair of operation points, which is classified to obtain to classify to gather, includes:
Production consumption historical data under first of operating condition of k-th of equipment is assembled into the classification set Sk,l, Sk,lTable
It is shown as:
Sk,l={ St k,l| t=1,2 ..., T }
Wherein, T indicates that all number of nodes, t indicate timing node, and the difference between t and t+1 is setting time interval.
Preferably, the data in the classification set, which are normalized to obtain to normalize to gather, includes:
It is mapped to all data in the classification set linearly between 0-1 using min-max method, to obtain institute
State normalization set X, the normalization set expression are as follows: X={ Xt| t=1,2 ..., T }, transfer function indicates are as follows:
In formula, XtIndicate the data in the normalization set, min indicates the minimum value in the classification set, max table
Show the maximum value in the classification set.
Preferably, the setup parameter includes input layer number, hidden layer number, the number of hidden nodes and output node layer
Number.
Preferably, the propagation algorithm forward includes:
Normalization set is inputted in the shot and long term Memory Neural Networks prediction model, is exported by iterative calculation
Prediction sets Y, the prediction sets indicate are as follows: Y={ Yt| t=1,2 ..., T }, wherein
Yt=(Wy·Nt,h)+by(1)
Nt,g=tanh [(Wg·Xt,c)+bg](2)
Nt,i=σ [(Wi·Xt,c)+bi](3)
Nt,f=σ [(Wf·Xt,c)+bf](4)
Nt,o=σ [(Wo·Xt,c)+bo](5)
Nt,s=Nt,g*Nt,i+Nt-1,s*Nt,f(6)
Nt,h=tanh (Nt,s)*Nt,o(7)
YtIndicate the data in the prediction sets, formula (2)-(7) are the calculation formula of cell hidden layer, wherein formula
(2) input door state, input gate, forgetting door, out gate, cell state and the node that-(7) respectively indicate t-th of node are defeated
Vector calculation formula out, Nt,g、Nt,i、Nt,f、Nt,o、Nt,s、Nt,hRespectively indicate the input door state of t-th of node, input gate,
Forget door, out gate, cell state and the vector calculated value of node output, WyIndicate the output matrix of node, byIndicate node
Bias vector, Wg、Wi、Wf、WoRespectively indicate input door state, input gate, the weight matrix for forgeing door, out gate, bg、bi、
bf、boRespectively indicate input door state, input gate, the bias vector for forgeing door, out gate, Xt,cWhat is indicated is t-th of node
Input data, Xt,c=[Xt,ht-1], ht-1Indicate the output data of the t-1 node, Nt-1,sIndicate the cell of the t-1 node
The vector calculated value of state, σ indicate Logistic function, and expression formula indicates are as follows:
Tanh indicates hyperbolic tangent function, and expression formula indicates are as follows:
Wherein, what x was indicated is the data in each vector.
Preferably, the back-propagation algorithm includes:
Since the last one node, principle is declined by gradient, is completed with error function is gradually minimized to the length
The training of short-term memory neural network prediction model.
Preferably, described error between the prediction result and actual value is calculated by error calculation formula to include:
Error between the prediction result and the actual value, RMSE calculation formula are calculated using root-mean-square error RMSE
It indicates are as follows:
YtIndicate the prediction result of t-th of node,Indicate the actual value of t-th of node.
Preferably, the S7 includes:
S71: the production consumption historical data in time interval corresponding with the time interval to be predicted is returned
One change processing；
S72: the production consumption historical data after normalized is inputted into the optimum prediction model, passes through propagated forward
It calculates, obtains output data；
S73: the output data is subjected to anti-normalization processing, obtains prediction data.
Preferably, the S73 includes:
The transfer function that the anti-normalization processing uses are as follows:
Y=y*×(max-min)+min
In formula, y indicates the prediction data, y*Indicate the output data, min is indicated and the time interval to be predicted
The minimum value in production consumption historical data in corresponding time interval, max are indicated and the time interval phase to be predicted
The maximum value in production consumption historical data in corresponding time interval.
(3) beneficial effect
The beneficial effects of the present invention are: having comprehensively considered influence of the equipment working condition difference to steam generation and consumption, establish
The complexity that the present invention is previously mentioned is solved based on the steam production of equipment multi-state operation and consumption dynamic prediction method
Steam forecasting problem under working condition more accurately predicts the steam production and Ge Sheng of each equipment in the future scheduling period
Produce the steam consumption of user.
Detailed description of the invention
Fig. 1 is the flow chart for the dynamic prediction method that steam of the invention produces consumption；
Fig. 2 is the operating point classification schematic diagram for the dynamic prediction method that steam of the invention produces consumption；
Fig. 3 is the operational flow diagram of shot and long term Memory Neural Networks prediction model；
Fig. 4 is the predicted value and actual comparison figure that steam of the invention produces quantity of steam in the dynamic prediction method of consumption.
Specific embodiment
In order to preferably explain the present invention, in order to understand, with reference to the accompanying drawing, by specific embodiment, to this hair
It is bright to be described in detail.
The present invention provides a kind of dynamic prediction method of steam production consumption, and Fig. 1 is the dynamic that steam of the invention produces consumption
The flow chart of prediction technique, Fig. 3 are the operational flow diagram of shot and long term Memory Neural Networks prediction model, and steam produces the dynamic of consumption
Prediction technique includes the following steps:
S1, the production for obtaining each equipment steam consume historical data, and are normalized to consumption historical data is produced, and obtain
Normalization set, producing consumption historical data includes steam production historical data and steam consumption historical data；
S2, the multiple groups setup parameter for obtaining shot and long term Memory Neural Networks prediction model；
S3, normalization set is divided into training dataset and test data set；
S4, by the shot and long term Memory Neural Networks prediction model of training dataset input multiple groups setup parameter, by preceding
It is trained to propagation algorithm and back-propagation algorithm, forms training result data；
S5, according to training result data, using adaptability moments estimation algorithm, to being set to, shot and long term Memory Neural Networks are pre-
The node matrix equation and bias vector for the multiple groups setup parameter surveyed in model are updated, and obtain the prediction model of multiple groups optimization；
S6, test data set is substituted into the prediction model that multiple groups optimize and is predicted to obtain prediction result, pass through error
Calculation formula calculates the error between prediction result and actual value, the prediction of the corresponding optimization of the smallest prediction result of Select Error
Model is as optimum prediction model；
S7, consumption data are produced using the steam of optimum prediction model prediction time interval to be predicted, steam produces consumption data and includes
Steam production data and steam consumption data.
Preferably, step S1 specifically includes the following steps:
S11, the production for obtaining each equipment steam consume historical data, and the corresponding production plan and inspection for producing consumption historical data
The plan of repairing determines the production status of each equipment according to production plan and maintenance plan；
S12, the operating point that each equipment is obtained according to production status produce consumption historical data according to pair of operation points and classify
Classification set is obtained, and the data in classification set are normalized to obtain normalization set.
Specifically, classified to obtain to classify to gather according to pair of operation points production consumption historical data and include:
Production consumption historical data under first of operating condition of k-th of equipment is assembled into category set and closes Sk,l, Sk,lIt indicates are as follows:
Sk,l={ St k,l| t=1,2 ..., T }
Wherein, T indicates that all number of nodes, t indicate timing node, and the difference between t and t+1 is setting time interval.
Data in classification set are normalized to obtain to normalize to gather and include:
It is mapped to all data in classification set linearly between 0-1 using min-max method, to be normalized
Set X normalizes set expression are as follows: X={ Xt| t=1,2 ..., T }, transfer function indicates are as follows:
In formula, XtIndicate the data in normalization set, the minimum value in min presentation class set, max presentation class collection
Maximum value in conjunction.
Preferably, setup parameter includes input layer number, hidden layer number, the number of hidden nodes and output layer number of nodes.
Propagation algorithm includes: to gather normalization in input shot and long term Memory Neural Networks prediction model forward, by repeatedly
In generation, calculates output prediction sets Y, and prediction sets indicate are as follows: Y={ Yt| t=1,2 ..., T }, wherein
Yt=(Wy·Nt,h)+by(1)
Nt,g=tanh [(Wg·Xt,c)+bg](2)
Nt,i=σ [(Wi·Xt,c)+bi](3)
Nt,f=σ [(Wf·Xt,c)+bf](4)
Nt,o=σ [(Wo·Xt,c)+bo](5)
Nt,s=Nt,g*Nt,i+Nt-1,s*Nt,f(6)
Nt,h=tanh (Nt,s)*Nt,o(7)
YtIndicate the data in prediction sets, formula (2)-(7) are the calculation formula of cell hidden layer, wherein formula (2)-
(7) it respectively indicates the input door state of t-th of node, input gate, forget door, out gate, cell state and node output
Vector calculation formula, Nt,g、Nt,i、Nt,f、Nt,o、Nt,s、Nt,hRespectively indicate input door state, the input gate, forgetting of t-th of node
Door, out gate, cell state and the vector calculated value of node output, WyIndicate the output matrix of node, byIndicate the inclined of node
Difference vector, Wg、Wi、Wf、WoRespectively indicate input door state, input gate, the weight matrix for forgeing door, out gate, bg、bi、bf、bo
Respectively indicate input door state, input gate, the bias vector for forgeing door, out gate, Xt,cWhat is indicated is the input of t-th of node
Data, Xt,c=[Xt,ht-1], ht-1Indicate the output data of the t-1 node, Nt-1,sIndicate the cell state of the t-1 node
Vector calculated value, σ indicate Logistic function, expression formula indicate are as follows:
Tanh indicates hyperbolic tangent function, and expression formula indicates are as follows:
Wherein, what x was indicated is the data in each vector.
Back-propagation algorithm includes: to decline principle by gradient, with gradually minimizing error since the last one node
Function completes the training to shot and long term Memory Neural Networks prediction model.
Calculating the error between prediction result and actual value by error calculation formula includes: using root-mean-square error RMSE
The error between prediction result and actual value is calculated, RMSE calculation formula indicates are as follows:
YtIndicate the prediction result of t-th of node,Indicate the actual value of t-th of node.
Preferably, step S7 specifically includes the following steps:
S71: the production consumption historical data in time interval corresponding with time interval to be predicted is normalized；
S72: the production consumption historical data after normalized is inputted into optimum prediction model, is calculated, is obtained by propagated forward
To output data；
S73: output data is subjected to anti-normalization processing, obtains prediction data.The conversion letter that anti-normalization processing uses
Number are as follows:
Y=y*×(max-min)+min
In formula, y indicates prediction data, y*Indicate output data, min indicates the time corresponding with time interval to be predicted
The minimum value in production consumption historical data in interval, max indicate the production in time interval corresponding with time interval to be predicted
Consume the maximum value in historical data.
Technical solution in order to further illustrate the present invention is said below by the embodiment of dry coke quenching residual heat boiler
It is bright.
By taking the steam of iron and steel enterprise produces consumption as an example, consumption specifically is produced to the steam of the dry coke quenching residual heat boiler in iron and steel enterprise
Dynamic prediction is measured, as shown in FIG. 1, FIG. 1 is the flow chart that steam of the invention produces the dynamic prediction method of consumption, the dynamic predictions
Specifically comprise the following steps:
Step 1: choosing the production consumption historical data and right of the steam in dry coke quenching residual heat boiler the past period interval
Should section time interval production plan and maintenance plan；Production status is obtained by production maintenance plan；Wherein, more than dry coke quenching
The production of heat boiler steam consumes historical data, can be by obtaining from enterprise database.The production consumption historical data of acquisition is dry puts out
Steam production and steam consumption of the coke waste heat boiler in this section of time interval, when time interval can be according to required prediction
Between be spaced and formulated.The steam flow data at dry coke quenching residual heat boiler setting time interval are obtained from database.It is following certain
The steam flow data of node refer to the steam flow data from this timing node behind setting time interval, setting time
It is preferably spaced 5 minutes.
Operating point division is carried out according to the production status of dry coke quenching residual heat boiler, is had to dry coke quenching residual heat boiler production status
Body subdivision, Fig. 2 are the operating point classification schematic diagram for the dynamic prediction method that steam of the invention produces consumption；According to dry coke quenching waste heat
The operating condition of boiler classifies to quantity of steam historical data, and the production under first of operating condition of dry coke quenching residual heat boiler is consumed history number
According to assemble classification set Sl, SlIt indicates are as follows:
Sl={ st l| t=1,2 ..., T }
Wherein, T indicates that all number of nodes, t indicate timing node, and the difference between t and t+1 is setting time interval.
Operating point is the label for capableing of the production status of unique identification dry coke quenching residual heat boiler in a certain period of time, operating condition
Point includes normal production status and various improper production status.
Step 2: the production consumption historical data of dry coke quenching residual heat boiler being normalized, the side min-max is mainly used
Method is mapped to all data in classification set linearly between 0-1, to obtain normalization set X, normalizes set expression
Are as follows: X={ Xt| t=1,2 ..., T }, transfer function indicates are as follows:
In formula, XtIndicate the data in normalization set, the minimum value in min presentation class set, max presentation class collection
Maximum value in conjunction.
Step 3: obtaining the multiple groups setup parameter of shot and long term Memory Neural Networks prediction model, setup parameter includes input layer
Number of nodes, hidden layer number, the number of hidden nodes, output layer number of nodes.
Step 4: the data in normalization set are divided into training dataset and test data set.
Step 5: training dataset input has been arranged to the shot and long term Memory Neural Networks prediction model of multiple groups setup parameter,
It is trained by propagated forward algorithm and back-propagation algorithm, forms training result data；
Specifically, the detailed process of propagated forward algorithm are as follows:
Normalization is gathered in input shot and long term Memory Neural Networks prediction model, prediction sets are exported by iterative calculation
Y, prediction sets indicate are as follows: Y={ Yt| t=1,2 ..., T }, wherein
Yt=(Wy·Nt,h)+by(1)
Nt,g=tanh [(Wg·Xt,c)+bg](2)
Nt,i=σ [(Wi·Xt,c)+bi](3)
Nt,f=σ [(Wf·Xt,c)+bf](4)
Nt,o=σ [(Wo·Xt,c)+bo](5)
Nt,s=Nt,g*Nt,i+Nt-1,s*Nt,f(6)
Nt,h=tanh (Nt,s)*Nt,o(7)
YtIndicate the data in prediction sets, formula (2)-(7) are the calculation formula of cell hidden layer, wherein formula (2)-
(7) it respectively indicates the input door state of t-th of node, input gate, forget door, out gate, cell state and node output
Vector calculation formula, Nt,g、Nt,i、Nt,f、Nt,o、Nt,s、Nt,hRespectively indicate input door state, the input gate, forgetting of t-th of node
Door, out gate, cell state and the vector calculated value of node output, WyIndicate the output matrix of node, byIndicate the inclined of node
Difference vector, Wg、Wi、Wf、WoRespectively indicate input door state, input gate, the weight matrix for forgeing door, out gate, bg、bi、bf、bo
Respectively indicate input door state, input gate, the bias vector for forgeing door, out gate, Xt,cWhat is indicated is the input of t-th of node
Data, Xt,c=[Xt,ht-1], ht-1Indicate the output data of the t-1 node, Nt-1,sIndicate the cell state of the t-1 node
Vector calculated value, σ indicate Logistic function, expression formula indicate are as follows:
Tanh indicates hyperbolic tangent function, and expression formula indicates are as follows:
Wherein, what x was indicated is the data in each vector.
On the other hand, specifically, error backpropagation algorithm is since the last one node, principle is declined by gradient,
The training to shot and long term Memory Neural Networks prediction model, detailed process are completed with gradually minimum error function are as follows:
What formula (8) indicated is the methods for computing error function erf x of prediction model, and that E is indicated is cumulative errors, YtIndicate t
The prediction data of a node,Indicate the actual value of t-th of node.
That formula (9) indicates is error local derviation of the node output result relative to actual value, Dy,tWhat is indicated is t-th of section
The error local derviation of point prediction data and actual value.
Dt,h=Dy,t·Wy(10)
That formula (10) indicates is the error local derviation calculation method of node output, Dt,hWhat is indicated is error relative to current t
The local derviation of node output, WyIndicate the output matrix of node.
Dt,s=tanh (Nt,s)*Ni,o*Dtop,h+Dtop,s(11)
That formula (11) indicates is the error local derviation calculation method of node state, Dt,sIndicate be cumulative errors relative to
The local derviation of current t node state, Dtop,sIndicate be current t node and later the error of all nodes relative to current t node
The local derviation of state, Dtop,hWhat is indicated is local derviation of the error of all nodes after current t node relative to present node state, he
Calculating be respectively:
That formula (12) indicates is the cumulative errors local derviation calculation method of node output, Dtop,hWhat is indicated is t-th of node
The accumulative node output error local derviation of all nodes later.
That formula (13) indicates is the cumulative errors local derviation calculation method of node output, Dtop,sWhat is indicated is t-th of node
The accumulative node output error local derviation of all nodes later, when the last one node calculates, the local derviation is 0.
Dt,o=tanh (Nt,s)*Dtop,h(14)
That formula (14) indicates is the error local derviation calculation method of t node input gate, Dt,oWhat is indicated is error relative to t
The local derviation of node input gate.
Dt,i=Nt,g*Dt,s(15)
That formula (15) indicates is the error local derviation calculation method of t node input gate, Dt,iWhat is indicated is error relative to t
The local derviation of node input gate.
Dt,g=Nt,i*Dt,s(16)
That formula (16) indicates is the error local derviation calculation method of t node input door state, Dt,gWhat is indicated is that error is opposite
In the local derviation of t node input door state.
Dt,f=Nt-1,s*Dt,s(17)
That formula (17) indicates is the error local derviation calculation method that t node forgets door, Dt,fWhat is indicated is error relative to t
The local derviation of node forgetting door.Ni-1,sWhat is indicated is the state of a upper node.
Dt,bi=Nt,s*(1-Nt,s)*Dt,i(18)
That formula (18) indicates is the error local derviation calculation method of t node input gate matrix biasing, Dt,biWhat is indicated is to miss
Local derviation of the difference relative to the input gate matrix biasing of t node, Nt,sWhat is indicated is the state of t node.
Dt,bf=Nt,f*(1-Nt,f)*Dt,f(19)
That formula (19) indicates is the error local derviation calculation method that t node forgets gate matrix biasing, Dt,bfWhat is indicated is to miss
Difference forgets the local derviation of gate matrix biasing, N relative to t nodet,fWhat is indicated is the forgetting door of t node.
Di,bo=Ni,o*(1-Ni,o)*Di,o(20)
That formula (20) indicates is the error local derviation calculation method of t node output gate matrix biasing, Di,boWhat is indicated is to miss
Local derviation of the difference relative to the output gate matrix biasing of t node, Ni,oWhat is indicated is the out gate of t node.
Dt,bg=Nt,g*(1-Nt,g)*Dt,g(21)
That formula (21) indicates is the error local derviation calculation method of t node input gate state matrix biasing, Dt,bgIt indicates
It is the local derviation that error is biased relative to t node input gate state matrix, Nt,gWhat is indicated is the input door state of t node.
Dt,wi=Dt,bi·Xt,c(22)
That formula (22) indicates is the error local derviation calculation method that t node forgets gate matrix, Dt,wiWhat is indicated is error phase
The local derviation of gate matrix, D are forgotten for t nodet,biWhat is indicated is the local derviation that error forgets gate matrix biasing relative to t node.
Dt,wf=Dt,bf·Xt,c(23)
That formula (23) indicates is the error local derviation calculation method that t node forgets gate matrix, Dt,wfWhat is indicated is error phase
The local derviation of gate matrix, D are forgotten for t nodet,bfWhat is indicated is the local derviation that error forgets gate matrix biasing relative to t node.
Dt,wo=Dt,bo·Xt,c(24)
That formula (24) indicates is the error local derviation calculation method of t node output gate matrix, Dt,woWhat is indicated is error phase
For the local derviation of t node output gate matrix, Dt,boWhat is indicated is local derviation of the error relative to the output gate matrix biasing of t node.
Dt,wg=Dt,bg·Xt,c(25)
That formula (25) indicates is the error local derviation calculation method of node input gate state matrix, Dt,wgWhat is indicated is error
Relative to the local derviation of node input gate matrix, Dt,bgIndicate to be that error is biased relative to node input gate state matrix inclined
It leads.
Step 6: according to training result data, using adaptability moments estimation algorithm, remembering nerve net to shot and long term is set to
The node matrix equation and bias vector of multiple groups setup parameter in network prediction model are updated, and obtain the prediction mould of multiple groups optimization
Type；Detailed process is as follows:
Mθ, n=β1*Mθ,n-1+(1-β1)*Dθ,n(26)
That formula (26) indicates is the single order moments estimation calculation formula of parameter θ, Mθ,nWhen expression parameter θ nth iteration updates
Single order moments estimation, β1That indicate is the exponential decay rate of single order moments estimation, Mθ,n-1What is indicated is ginseng when (n-1)th iteration updates
The single order moments estimation of number θ, Dθ,nIndicate local derviation of the error relative to parameter θ when nth iteration updates.
Vθ,n=β2*Vθ,n-1+(1-β2)*(Dθ,t)2(27)
That formula (27) indicates is the second order moments estimation calculation formula of parameter θ, Vθ,nWhen expression parameter θ nth iteration updates
Second order moments estimation, β2That indicate is the exponential decay rate of single order moments estimation, Vθ,n-1What is indicated is ginseng when (n-1)th iteration updates
The second order moments estimation of number θ, Dθ,nIndicate local derviation of the error relative to parameter θ when nth iteration updates.
What formula (28) indicated is the single order moments estimation calculation formula of parameter θ drift correction,Expression parameter θ deviation is repaired
Positive single order moments estimation.
What formula (29) indicated is the second order moments estimation calculation formula of parameter θ drift correction,Expression parameter θ deviation is repaired
Positive single order moments estimation.
What formula (30) indicated is the calculation formula that parameter θ updates, and what lr was indicated is learning rate.
Step 7: test data set being substituted into the prediction model of multiple groups optimization and predicted to obtain prediction result, pass through mistake
Poor calculation formula calculates the error between prediction result and actual value, the corresponding optimization of the smallest prediction result of Select Error it is pre-
Model is surveyed as optimum prediction model, the comparison between prediction result and actual value is as shown in figure 4, Fig. 4 is steam of the invention
Produce the predicted value and actual comparison figure of quantity of steam in the dynamic prediction method of consumption.Using root-mean-square error RMSE (Root
Mean Square Error, RMSE) error between prediction result and actual value is calculated, RMSE calculation formula indicates are as follows:
YtIndicate the prediction result of t-th of node,Indicate the actual value of t-th of node.
Step 8: place is normalized in the production consumption historical data in time interval corresponding with time interval to be predicted
Reason；
Step 9: the production consumption historical data after normalized is inputted into optimum prediction model, is calculated by propagated forward,
Obtain output data；
Step 10: output data being subjected to anti-normalization processing, obtains prediction data.The conversion that anti-normalization processing uses
Function are as follows:
Y=y*×(max-min)+min
In formula, y indicates prediction data, y*Indicate output data, min indicates the time corresponding with time interval to be predicted
The minimum value in production consumption historical data in interval, max indicate the production in time interval corresponding with time interval to be predicted
Consume the maximum value in historical data.
Step 11: prediction result being uploaded on the steam dispatch server of enterprise, steam dispatches system according to prediction knot
Fruit is scheduled, and enterprise energy administrative staff are scheduled according to steam prediction result with steam scheduling scheme.
The present invention has comprehensively considered steam production and consumption of each equipment under different operating conditions, is based on distinct device
The production of different operating conditions consumes historical data, and the parameter of training optimization shot and long term memory network prediction model obtains optimum prediction model,
It predicts that the steam of time interval to be predicted produces consumption data by optimum prediction model, the future scheduling period is more accurately predicted
The steam generating amount of interior steam generation device and it is each production user steam consumption, for enterprise's schedule provide reliably according to
According to.
It is to be appreciated that describing the skill simply to illustrate that of the invention to what specific embodiments of the present invention carried out above
Art route and feature, its object is to allow those skilled in the art to can understand the content of the present invention and implement it accordingly, but
The present invention is not limited to above-mentioned particular implementations.All various changes made within the scope of the claims are repaired
Decorations, should be covered by the scope of protection of the present invention.
Claims (10)
1. the dynamic prediction method that a kind of steam produces consumption, which is characterized in that the described method includes:
S1, the production for obtaining each equipment steam consume historical data, and production consumption historical data is normalized, and obtain
Normalization set, the production consumption historical data includes steam production historical data and steam consumption historical data；
S2, the multiple groups setup parameter for obtaining shot and long term Memory Neural Networks prediction model；
S3, normalization set is divided into training dataset and test data set；
S4, the shot and long term Memory Neural Networks that the setup parameter has been arranged in training dataset input multiple groups are predicted
Model is trained by propagated forward algorithm and back-propagation algorithm, forms training result data；
S5, according to the training result data, using adaptability moments estimation algorithm, remember nerve net to the shot and long term is set to
The node matrix equation and bias vector of setup parameter described in multiple groups in network prediction model are updated, and obtain the prediction of multiple groups optimization
Model；
S6, it will be predicted to obtain prediction result in the prediction model of optimization described in test data set substitution multiple groups, be passed through
Error calculation formula calculates the error between the prediction result and actual value, and the smallest prediction result of Select Error is corresponding
The optimization prediction model as optimum prediction model；
S7, consumption data are produced using the steam of the optimum prediction model prediction time interval to be predicted, the steam produces consumption data
Including steam production data and steam consumption data.
2. steam as described in claim 1 produces the dynamic prediction method of consumption, which is characterized in that the S1 includes:
S11, the production for obtaining each equipment steam consume historical data, and the corresponding production plan for producing consumption historical data
And maintenance plan, the production status of each equipment is determined according to the production plan and the maintenance plan；
S12, the operating point that each equipment is obtained according to the production status, the production consumption historical data according to the pair of operation points
Classified to obtain classification set, and the data in the classification set are normalized to obtain the normalization collection
It closes.
3. the dynamic prediction method that steam as claimed in claim 2 produces consumption, which is characterized in that described according to the operating point
Production consumption historical data is classified to obtain to classify to gather and includes:
Production consumption historical data under first of operating condition of k-th of equipment is assembled into the classification set Sk,l, Sk,lIt indicates
Are as follows:
Sk,l={ St k,l| t=1,2 ..., T }
Wherein, T indicates that all number of nodes, t indicate timing node, and the difference between t and t+1 is setting time interval.
4. the dynamic prediction method that steam as claimed in claim 3 produces consumption, which is characterized in that described to gather the classification
In data be normalized to obtain normalization set and include:
It is mapped to all data in the classification set linearly between 0-1 using min-max method, to return described in obtaining
One changes set X, the normalization set expression are as follows: X={ Xt| t=1,2 ..., T }, transfer function indicates are as follows:
In formula, XtIndicate the data in the normalization set, min indicates the minimum value in the classification set, and max indicates institute
State the maximum value in classification set.
5. the dynamic prediction method that steam as claimed in claim 4 produces consumption, which is characterized in that the setup parameter includes defeated
Enter node layer number, hidden layer number, the number of hidden nodes and output layer number of nodes.
6. the dynamic prediction method that steam as claimed in claim 5 produces consumption, which is characterized in that the propagation algorithm packet forward
It includes:
Normalization set is inputted in the shot and long term Memory Neural Networks prediction model, iterative calculation output prediction is passed through
Set Y, the prediction sets indicate are as follows: Y={ Yt| t=1,2 ..., T }, wherein
Yt=(Wy·Nt,h)+by (1)
Nt,g=tanh [(Wg·Xt,c)+bg] (2)
Nt,i=σ [(Wi·Xt,c)+bi] (3)
Nt,f=σ [(Wf·Xt,c)+bf] (4)
Nt,o=σ [(Wo·Xt,c)+bo] (5)
Nt,s=Nt,g*Nt,i+Nt-1,s*Nt,f (6)
Nt,h=tanh (Nt,s)*Nt,o (7)
YtIndicate the data in the prediction sets, formula (2)-(7) are the calculation formula of cell hidden layer, wherein formula (2)-
(7) it respectively indicates the input door state of t-th of node, input gate, forget door, out gate, cell state and node output
Vector calculation formula, Nt,g、Nt,i、Nt,f、Nt,o、Nt,s、Nt,hRespectively indicate input door state, the input gate, forgetting of t-th of node
Door, out gate, cell state and the vector calculated value of node output, WyIndicate the output matrix of node, byIndicate the inclined of node
Difference vector, Wg、Wi、Wf、WoRespectively indicate input door state, input gate, the weight matrix for forgeing door, out gate, bg、bi、bf、bo
Respectively indicate input door state, input gate, the bias vector for forgeing door, out gate, Xt,cWhat is indicated is the input of t-th of node
Data, Xt,c=[Xt,ht-1], ht-1Indicate the output data of the t-1 node, Nt-1,sIndicate the cell state of the t-1 node
Vector calculated value, σ indicate Logistic function, expression formula indicate are as follows:
Tanh indicates hyperbolic tangent function, and expression formula indicates are as follows:
Wherein, what x was indicated is the data in each vector.
7. the dynamic prediction method that steam as described in claim 1 produces consumption, which is characterized in that the back-propagation algorithm packet
It includes:
Since the last one node, principle is declined by gradient, is completed with error function is gradually minimized to the shot and long term
The training of Memory Neural Networks prediction model.
8. the dynamic prediction method that steam as described in claim 1 produces consumption, which is characterized in that described to pass through error calculation public affairs
The error that formula calculates between the prediction result and actual value includes:
Error between the prediction result and the actual value is calculated using root-mean-square error RMSE, RMSE calculation formula indicates
Are as follows:
YtIndicate the prediction result of t-th of node,Indicate the actual value of t-th of node.
9. the dynamic prediction method that steam as described in claim 1 produces consumption, which is characterized in that the S7 includes:
S71: the production consumption historical data in time interval corresponding with the time interval to be predicted is normalized
Processing；
S72: the production consumption historical data after normalized is inputted into the optimum prediction model, passes through propagated forward meter
It calculates, obtains output data；
S73: the output data is subjected to anti-normalization processing, obtains prediction data.
10. the dynamic prediction method that steam as claimed in claim 9 produces consumption, which is characterized in that the S73 includes:
The transfer function that the anti-normalization processing uses are as follows:
Y=y*×(max-min)+min
In formula, y indicates the prediction data, y*Indicate the output data, min indicates opposite with the time interval to be predicted
The minimum value in production consumption historical data in the time interval answered, max indicate corresponding with the time interval to be predicted
Time interval in the production consumption historical data in maximum value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910731785.2A CN110428053A (en) | 2019-08-08 | 2019-08-08 | A kind of steam produces the dynamic prediction method of consumption |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910731785.2A CN110428053A (en) | 2019-08-08 | 2019-08-08 | A kind of steam produces the dynamic prediction method of consumption |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110428053A true CN110428053A (en) | 2019-11-08 |
Family
ID=68415165
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910731785.2A Pending CN110428053A (en) | 2019-08-08 | 2019-08-08 | A kind of steam produces the dynamic prediction method of consumption |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110428053A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112434873A (en) * | 2020-12-03 | 2021-03-02 | 上海应用技术大学 | Novel method for predicting dry quenching circulating air quantity by using element calculation LSTM |
| CN112906307A (en) * | 2021-03-24 | 2021-06-04 | 青岛洪锦智慧能源技术有限公司 | Steam yield prediction method of steam boiler based on data mining |
| CN113242226A (en) * | 2021-05-05 | 2021-08-10 | 航天云网云制造科技(浙江)有限公司 | Big data-based intelligent network security situation prediction method |
| CN113705849A (en) * | 2020-05-21 | 2021-11-26 | 富士通株式会社 | Information processing apparatus, information processing method, and computer program |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106485353A (en) * | 2016-09-30 | 2017-03-08 | 中国科学院遥感与数字地球研究所 | Air pollutant concentration forecasting procedure and system |
| CN107239859A (en) * | 2017-06-05 | 2017-10-10 | 国网山东省电力公司电力科学研究院 | The heating load forecasting method of Recognition with Recurrent Neural Network is remembered based on series connection shot and long term |
| CN107918368A (en) * | 2017-11-28 | 2018-04-17 | 东北大学 | The dynamic prediction method and equipment of iron and steel enterprise's coal gas yield and consumption |
-
2019
- 2019-08-08 CN CN201910731785.2A patent/CN110428053A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106485353A (en) * | 2016-09-30 | 2017-03-08 | 中国科学院遥感与数字地球研究所 | Air pollutant concentration forecasting procedure and system |
| CN107239859A (en) * | 2017-06-05 | 2017-10-10 | 国网山东省电力公司电力科学研究院 | The heating load forecasting method of Recognition with Recurrent Neural Network is remembered based on series connection shot and long term |
| CN107918368A (en) * | 2017-11-28 | 2018-04-17 | 东北大学 | The dynamic prediction method and equipment of iron and steel enterprise's coal gas yield and consumption |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113705849A (en) * | 2020-05-21 | 2021-11-26 | 富士通株式会社 | Information processing apparatus, information processing method, and computer program |
| CN113705849B (en) * | 2020-05-21 | 2024-05-17 | 富士通株式会社 | Information processing apparatus and information processing method |
| CN112434873A (en) * | 2020-12-03 | 2021-03-02 | 上海应用技术大学 | Novel method for predicting dry quenching circulating air quantity by using element calculation LSTM |
| CN112906307A (en) * | 2021-03-24 | 2021-06-04 | 青岛洪锦智慧能源技术有限公司 | Steam yield prediction method of steam boiler based on data mining |
| CN113242226A (en) * | 2021-05-05 | 2021-08-10 | 航天云网云制造科技(浙江)有限公司 | Big data-based intelligent network security situation prediction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110428053A (en) | A kind of steam produces the dynamic prediction method of consumption | |
| CN109711620B (en) | Short-term power load prediction method based on GRU neural network and transfer learning | |
| CN111353656B (en) | Steel enterprise oxygen load prediction method based on production plan | |
| CN109325624A (en) | A kind of monthly electric power demand forecasting method based on deep learning | |
| CN111880044A (en) | Online fault positioning method for power distribution network with distributed power supply | |
| Zheng et al. | Short-term load forecasting of power system based on neural network intelligent algorithm | |
| CN109255726A (en) | A kind of ultra-short term wind power prediction method of Hybrid Intelligent Technology | |
| CN113111577B (en) | Cement mill operation index decision method based on multi-target cuckoo search | |
| CN115758246A (en) | Non-invasive load identification method based on EMD and AlexNet | |
| CN111340305A (en) | Building operation energy consumption prediction method | |
| CN114021848A (en) | Generating capacity demand prediction method based on LSTM deep learning | |
| CN106980910A (en) | Long Electric Power Load calculating system and method | |
| Qiu et al. | Fuzzy time series forecasting model based on automatic clustering techniques and generalized fuzzy logical relationship | |
| Akpinar et al. | Forecasting natural gas consumption with hybrid neural networks—Artificial bee colony | |
| CN114066069A (en) | Combined weight byproduct gas generation amount prediction method | |
| CN115828768A (en) | Data-driven power distribution network reactive power prediction method | |
| Lao et al. | The optimized multivariate grey prediction model based on dynamic background value and its application | |
| CN108304976A (en) | A kind of electric system fining load prediction and analysis method | |
| CN114091776A (en) | K-means-based multi-branch AGCNN short-term power load prediction method | |
| Zhang | Short‐Term Power Load Forecasting Based on SAPSO‐CNN‐LSTM Model considering Autocorrelated Errors | |
| Cao et al. | Combined electricity load-forecasting system based on weighted fuzzy time series and deep neural networks | |
| CN110659775A (en) | LSTM-based improved electric power short-time load prediction algorithm | |
| Yang et al. | A novel exhaust gas temperature prediction method of hot blast stove | |
| CN116128211A (en) | Wind-light-water combined short-term optimization scheduling method based on wind-light uncertainty prediction scene | |
| Jun-Ma et al. | Photovoltaic power generation prediction based on MEA-BP neural network |
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: 20191108 |
|
| RJ01 | Rejection of invention patent application after publication |