CN108197809A - Weights based on dimension optimum translation share the real-time scheduling method of depth network - Google Patents
Weights based on dimension optimum translation share the real-time scheduling method of depth network Download PDFInfo
- Publication number
- CN108197809A CN108197809A CN201711497688.9A CN201711497688A CN108197809A CN 108197809 A CN108197809 A CN 108197809A CN 201711497688 A CN201711497688 A CN 201711497688A CN 108197809 A CN108197809 A CN 108197809A
- Authority
- CN
- China
- Prior art keywords
- data
- network
- matrix
- layer
- dimension
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
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
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
- G06Q10/06312—Adjustment or analysis of established resource schedule, e.g. resource or task levelling, or dynamic rescheduling
-
- 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
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Economics (AREA)
- Software Systems (AREA)
- Development Economics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Educational Administration (AREA)
- Life Sciences & Earth Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Health & Medical Sciences (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Image Analysis (AREA)
Abstract
Weights based on dimension optimum translation share the real-time scheduling method of depth network, including：Step 1. obtains the real time data of acquisition actual schedule occasion and dispatches data as training data；Step 2. handles the real time data obtained in step 1, is treated as the multilayer two-dimension matrix form for meeting depth network inputs；Step 3., respectively as the input and output of depth network, is trained depth network using the scheduling data obtained in the multilayer two-dimension matrix in step 2 and step 1；Step 4. is by convolutional neural networks trained in step 3, using in practical dispatch environment；Carry out practical network scheduling.
Description
The present invention relates to the real-time scheduling methods that a kind of weights share depth network.
Background technology
For production scheduling problems, current main stream approach combines optimization heuritic approach using mathematical model and is asked
Solution, can obtain higher solution accuracy.But under the background of big data, the explosive increasing of manufacturing parameter in production environment
Long and stringent scheduling time mark sense dispatching method proposes further requirement.Apparently traditional dispatching algorithm is difficult at present
Meet industry requirement in terms of in processing magnanimity scheduling data and completing quick response two to scheduling problem.
Invention content
The present invention will overcome the disadvantages mentioned above of the prior art, propose a kind of real-time big data tune based on depth convolutional network
Spend technical method.
The present invention combines deep neural network algorithm and weights technology of sharing, it is proposed that " the power based on dimension optimum translation
The real-time scheduling method of the shared depth network of value " trains depth net by a large amount of schedule history data that big data system provides
Network understands the tacit knowledge in scheduling scenario.And the real-time scheduling at scheduling scene is completed using trained depth network
Response.
The present invention proposes a kind of scheduling neural network structure of quick response based on big data and a set of depth network tune
Degree method completes complicated industrial control and the production scheduling problems in the case of big data.
Weights based on dimension optimum translation share the real-time scheduling method of depth network, include the following steps：
Step 1. obtains the real time data of acquisition actual schedule occasion and scheduling data (correspond to flow chart as training data
Middle data acquiring portion)
Step 2. handles the real time data obtained in step 1, is treated as the multilayer for meeting depth network inputs
Two-dimensional matrix form.(data processing section in corresponding flow chart)." real time data is handled " specifically includes：
The pretreatment of 2.1 real time datas, whether the data of verificating sensor acquisition are reasonable, and 0 rank can be used to keep
Principle illegal data are substituted.
2.2 share to carry out weights in depth network training, and convolution group is carried out to dynamic history samples data
It closes.Convolutionization combination is as follows.
A1. set in step 1 that there are k sensors, respectively S1,S2,…,Sk, what i-th of sensor acquired under the j times
Data are dij。
A2., sampled signal input time window t is set in processing procedurew, rule of thumb parameter tw=10.In time window
Under limitation, data input matrix D is obtaineds.There is equation below：
D in formulai jI-th of sensor is all represented in j moment locality numerical value, what each one scheduling sensor of behavior transmitted
Sampling parameter, in twThe lower j=t of thresholding effectw=10.
A3. following mapping relations are established, multilayer two-dimension matrix form is generated by following two modes.
Method one：The signal that any two sensor acquires is converted into multilayer two-dimension matrix using cartesian product operation
M, mathematical description are as follows.
In formula, c be two-dimensional matrix the number of plies, Sp, Sq be serial number p and q sensor acquisition data picture vector, due to by
T is arrivedwInfluence, only taken 10 and immediate ten values of present moment.
If not to McThe number of plies abandoned, then the maximum number of plies of c meets formula combinations number formula
The result of the step operation is multilayer two-dimension matrix Mc
The maximum number c of 2.3 multilayer two-dimension matrixes is generated by number of combinations, therefore can equally generate showing for multiple shot array
As.In order to contain multiple shot array, using optimal method, established with reference to the correlativity of the practical significance of sensor gathered data
Optimal combination chain.In the case of determining chain length, the maximum correlation of chain is found.
B1. mathematical description is carried out to optimizing equation, the descriptive equation of optimization is as follows：
s.t.：0 ＜ i ＜ tw
E represents entire overall relevancy in a row in formula, sums with being described as R (l), and R (l) is each two plane earth in arrangement
Correlation, i are relatively indexing parameter.
B2. the multilayer two-dimension matrix M that the optimization problem optimized is solved using genetic algorithmc', Mc' matrix dimension
Degree is [n*10*10*c], wherein the number of n bands just training data.
Step 3. is using the scheduling data obtained in the multilayer two-dimension matrix in step 2 and step 1 respectively as depth net
The input and output of network are trained depth network.(network training part in corresponding flow chart)
3.1 use the optimization multilayer two-dimension matrix M handled well in step 2c' as input matrix, using in step 1)
The control matrix B of acquisition is brought convolutional neural networks into as label matrix and is trained.Trained process can be in nerve of increasing income
Network platform karas is carried out.
The training flow of convolutional neural networks is described below：
Convolutional neural networks (Convolutional Neural Network, CNN) are a kind of feedforward neural networks, it
Artificial neuron can respond the surrounding cells in a part of coverage area, have outstanding performance for large-scale image procossing.It is wrapped
Include convolutional layer (convolutional layer) and pond layer (pooling layer).
Usually, the basic structure of CNN includes two layers, and one is characterized extract layer, the input of each neuron with it is previous
The local acceptance region of layer is connected, and extract the feature of the part.After the local feature is extracted, it is between other feature
Position relationship is also decided therewith；The second is Feature Mapping layer, each computation layer of network is made of multiple Feature Mappings, often
A Feature Mapping is a plane, and the weights of all neurons are equal in plane.Feature Mapping structure is small using influence function core
Activation primitive of the sigmoid functions as convolutional network so that Feature Mapping has shift invariant.Further, since one
Neuron on mapping face shares weights, thus reduces the number of network freedom parameter.Each in convolutional neural networks
Convolutional layer all followed by one is used for asking the computation layer of local average and second extraction, this distinctive feature extraction structure twice
Reduce feature resolution.
The training step of 3.2CNN is as follows：
Input training set
For each sample M in training setc', the corresponding activation value a of setting input layer Input layer1：
3.2.1 the propagated forward of input data, it is as follows that propagated forward meets formula：
zl=wlMc'+bl,al=σ (zl) (3)
In formula, zlFor the information of downward Primary Transmit, wlFor the weights of neural network, alFor the biasing at network, σ (zl) be
Nonlinear neuron processing.Convolutional network carries out the setting of weights by sharing the convolution kernel of weights, does not do in the present invention
It is discussed in detail.
3.2.2 the error that output layer generates is calculated, error meets equation below：
δL=▽aC⊙σ'(zL) (4)
δLFor the final output of network and the wrong error of label, ▽aFor gradient operator, C ⊙ σ ' (zL) represent in output layer
It obtains target output and exports to obtain gap with model calculating.
3.2.3 each layer of reverse propagated error is calculated, the error of backpropagation meets equation below：
δl=((wl+1)Tδl+1)⊙σ'(zl) (5)
δlTo use δLThe every layer of error to front transfer calculated, l is the number of plies.
3.2.4 it is trained using gradient decline, the method that training uses meets equation below：
η is convergence step-length in formula, and m is to obtain number, δ using datax,l(ax,l-1)TGradient direction is obtained obtaining for each iteration,
The formula describes the mode of the change of weights.
Step 4. is by convolutional neural networks trained in step 3, using in practical dispatch environment.Carry out reality
Network scheduling.(Real-Time Scheduling part in corresponding flow chart)
The solution of traditional algorithm could only start dependent on specific scheduling data after data all obtain
The solution of scheduling is optimized, solution generally requires a large amount of time, so real-time is poor.In addition for big data in the case of
The problem of extensive ultra-large, solves, even if using didactic optimizing algorithm solve when being also required to largely calculate
Between.
It is an advantage of the invention that：The depth network algorithm that the present invention uses can will determine deep in prior training process
Network structure is spent, finds the implicit knowledge of scheduling, so as to be obtained under true application environment by very a small amount of operation
To real-time scheduling result, accomplish quick corresponding.Response speed is also very quick in the case of big data inputs.
Description of the drawings
Fig. 1 is the flow chart of the method for the present invention.
Fig. 2 is deep neural network pattern and input and output.
Fig. 3 is small data set Comparative result schematic diagram.
Fig. 4 is intermediate data collection result contrast schematic diagram.
Fig. 5 is large data collection result contrast schematic diagram.
Specific embodiment
The technical solution that 1-5 is further illustrated the present invention below in conjunction with the accompanying drawings.
Fig. 1 shows the flow chart of the method for the present invention.
Embodiment is summarized
Setting Shop-floor Scheduling, n platform machines, p workpiece, workpiece have q processing flow.Purpose is to each work
Each flow arrangement processing machine of part.N is changed, the value of p, q can change the scale of problem, setting small scale problem n=3, q=
3, q=3.Intermediate scale problem n=30, q=30, q=30.Large-scale problem n=300, q=300, q=300.Specific data mode
It is as shown in table 1 below：
Table 1
| Workpiece flow | Expend the time (input) | Serial number (output) |
| 1 flow 1 of workpiece | 5 | 5 |
| 1 flow 2 of workpiece | 12 | 6 |
| 1 flow 3 of workpiece | 3 | 2 |
| … | … | … |
| 2 flow 1 of workpiece | 4 | 1 |
| 2 flow 1 of workpiece | 8 | 4 |
| 2 flow 1 of workpiece | 6 | 8 |
| … | … | … |
| 3 flow 1 of workpiece | 9 | 3 |
| 3 flow 2 of workpiece | 41 | 9 |
| 3 flow 3 of workpiece | 4 | 7 |
| … | … | … |
The problem is the classical NP problems of comparison, is difficult to solve in the case where data volume is big.But this problem is very
It is ancient, it has been carried out more adequately studying.And the problem is the scheduling problem under a kind of more line, it is impossible to solve burst
Problem on line.Therefore it is all relatively advanced in electronic hardware and big data system, it is necessary to study the real-time streams on line
Scheduling problem between waterwheel.
The input of real-time Flow Shop Scheduling does not expend the time in the processing of only part, is actually processing
Preceding is difficult really to obtain the process time of part, therefore it is a kind of model it is assumed that so in the real-time of design to expend the time
Scheduling system in the processing of part is predicted.
Embodiment inputs and output
The data of input model are：
1) estimated value of each workpiece time.
2) machining state of current each machine.
3) discreet value of the remaining process time of the current workpiece of each machine.
Output data is：
Some workpiece is sent to the control signal that some machine completes some manufacturing procedure, as shown in Figure 2 by current state.
Embodiment exports result：Response speed compares
Small data set, medium-sized data set and large data collection are solved using two methods respectively, is rung in obtained result
The comparison for answering speed is as shown in Figure 3.
The result of traditional optimal method represents with light Nogata, the dark Nogata of methods and results of deep neural network
Figure represents.In the modelling phase, different network performances can be obtained by carrying out modeling using different network structure and network depth,
In Fig. 3, the time for listing the wherein network structure randomly generated spends.It is it can be seen that optimal in small data set
The response time of change method is generally more quicker than neural network method.It is such the result is that it is anticipated that because optimal
Change method is more suitable for solving the problems, such as small data set.
The results are shown in Figure 4 for intermediate data collection calculating method, with the increase of data volume, the time loss of optimal method into
It is linearly increasing, but the increase of neural network method response time is not obvious.In the case of intermediate data collection, the response of two methods
Time is in the same order of magnitude, and depth network method is slightly won.
With the increase again of data dimension, problem becomes complicated scheduling problem, in this case, optimal method
Response time becomes very huge, can not meet the needs of Real-Time Scheduling.But the dispatching method of depth network due to the use of
The good neural network structure of precondition in the time delay very little that scheduling is live, can complete the task of spot dispatch, such as Fig. 5 institutes
Show.
Content described in this specification embodiment is only enumerating to the way of realization of inventive concept, protection of the invention
Range is not construed as being only limitted to the concrete form that embodiment is stated, protection scope of the present invention is also and in art technology
Personnel according to present inventive concept it is conceivable that equivalent technologies mean.
Claims (1)
1. the weights based on dimension optimum translation share the real-time scheduling method of depth network, include the following steps：
Step 1. obtains the real time data of acquisition actual schedule occasion and dispatches data as training data；
Step 2. handles the real time data obtained in step 1, is treated as the multilayer two-dimension for meeting depth network inputs
Matrix form, the real time data carry out processing and specifically include：
The pretreatment of 2.1 real time datas, the data principle pair that is whether reasonable, and being kept using 0 rank of verificating sensor acquisition
Illegal data are substituted；
2.2 share to carry out weights in depth network training, and convolution combination is carried out to dynamic history samples data；Volume
Productization combination is as follows；
A1. set in step 1 that there are k sensors, respectively S1,S2,…,Sk, data that i-th of sensor acquires under the j times
For dij；
A2., sampled signal input time window t is set in processing procedurew, rule of thumb parameter tw=10；In the limitation of time window
Under, obtain data input matrix Ds；There is equation below：
D in formulai jI-th of sensor is all represented in j moment locality numerical value, the sampling that each one scheduling sensor of behavior transmits
Parameter, in twThe lower j=t of thresholding effectw=10；
A3. following mapping relations are established, multilayer two-dimension matrix form is generated by following two modes；
Method one：The signal that any two sensor acquires is converted into multilayer two-dimension matrix M using cartesian product operation, number
Be described as follows；
In formula, c is the number of plies of two-dimensional matrix, and Sp, Sq are the data picture vector of serial number p and q sensor acquisition, due to receiving
twInfluence, only taken 10 and immediate ten values of present moment；
If not to McThe number of plies abandoned, then the maximum number of plies of c meets formula combinations number formula
Method two：It using Lie groupoid method, is converted, step is as follows；
T1. blank two-dimensional matrix, dimension t are generatedw*tw
T2. successively willIt depicts to obtain matrix Mc；
The numbers of plies of the c for two-dimensional matrix, S in formulap,SqData picture for the acquisition of serial number p and q sensor is vectorial, due to receiving tw
Influence, only taken 10 and immediate ten values of present moment；If not to McThe number of plies abandoned, then the maximum layer of c
Number meets formula combinations number formula
The result of the step operation is multilayer two-dimension matrix Mc；
The maximum number c of 2.3 multilayer two-dimension matrixes is generated by number of combinations, therefore the phenomenon that can equally generate multiple shot array；
In order to contain multiple shot array, using optimal method, established most with reference to the correlativity of the practical significance of sensor gathered data
Optimum organization chain；In the case of determining chain length, the maximum correlation of chain is found；
B1. mathematical description is carried out to optimizing equation, the descriptive equation of optimization is as follows：
E represents entire overall relevancy in a row in formula, sums with being described as R (l), and R (l) is related for each two plane earth in arrangement
Property, i is relatively indexing parameter；
B2. the multilayer two-dimension matrix M that the optimization problem optimized is solved using genetic algorithmc', Mc' dimension of matrix is
The number of [n*10*10*c], wherein n bands just training data；
Step 3. is using the scheduling data obtained in the multilayer two-dimension matrix in step 2 and step 1 respectively as depth network
Input and output are trained depth network；
3.1 use the optimization multilayer two-dimension matrix M handled well in step 2c' as input matrix, use what is acquired in step 1
Control matrix B is brought convolutional neural networks into as label matrix and is trained；Trained process can be put down in neural network of increasing income
Platform karas is carried out；
The training flow of convolutional neural networks is as follows：
Convolutional neural networks CNN is a kind of feedforward neural network, its artificial neuron can be responded in a part of coverage area
Surrounding cells, have outstanding performance for large-scale image procossing；It includes convolutional layer and pond layer；
Usually, the basic structure of CNN includes two layers, and one is characterized extract layer, the input of each neuron and preceding layer
Local acceptance region is connected, and extracts the feature of the part；After the local feature is extracted, its position between other feature
Relationship is also decided therewith；The second is Feature Mapping layer, each computation layer of network is made of multiple Feature Mappings, Mei Gete
Sign mapping is a plane, and the weights of all neurons are equal in plane；Feature Mapping structure is small using influence function core
Activation primitive of the sigmoid functions as convolutional network so that Feature Mapping has shift invariant；Further, since one reflects
The neuron penetrated on face shares weights, thus reduces the number of network freedom parameter；Each volume in convolutional neural networks
Lamination all followed by one is used for asking the computation layer of local average and second extraction, and this distinctive structure of feature extraction twice subtracts
Small feature resolution；
The training step of 3.2 CNN is as follows：
Input training set
For each sample M in training setc', the corresponding activation value a of setting input layer Input layer1：
3.2.1 the propagated forward of input data, it is as follows that propagated forward meets formula：
zl=wlMc'+bl,al=σ (zl) (3)
In formula, zlFor the information of downward Primary Transmit, wlFor the weights of neural network, alFor the biasing at network, σ (zl) it is non-linear
Neuron processing；Convolutional network carries out the setting of weights by sharing the convolution kernel of weights, is not detailed Jie in the present invention
It continues；
3.2.2 the error that output layer generates is calculated, error meets equation below：
δLFor the final output of network and the wrong error of label,For gradient operator, C ⊙ σ ' (zL) represent and in output layer obtain mesh
Mark output exports to obtain gap with model calculating；
3.2.3 each layer of reverse propagated error is calculated, the error of backpropagation meets equation below：
δl=((wl+1)Tδl+1)⊙σ'(zl) (5)
δlTo use δLThe every layer of error to front transfer calculated, l is the number of plies；
3.2.4 it is trained using gradient decline, the method that training uses meets equation below：
η is convergence step-length in formula, and m is to obtain number, δ using datax,l(ax,l-1)TGradient direction is obtained obtaining for each iteration, the public affairs
Formula describes the mode of the change of weights；
Step 4. is by convolutional neural networks trained in step 3, using in practical dispatch environment；Carry out practical network
Scheduling.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711497688.9A CN108197809B (en) | 2017-12-28 | 2017-12-28 | Real-time scheduling method of weight sharing deep network based on dimension optimal conversion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711497688.9A CN108197809B (en) | 2017-12-28 | 2017-12-28 | Real-time scheduling method of weight sharing deep network based on dimension optimal conversion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108197809A true CN108197809A (en) | 2018-06-22 |
| CN108197809B CN108197809B (en) | 2021-06-08 |
Family
ID=62587596
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711497688.9A Active CN108197809B (en) | 2017-12-28 | 2017-12-28 | Real-time scheduling method of weight sharing deep network based on dimension optimal conversion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108197809B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110599496A (en) * | 2019-07-30 | 2019-12-20 | 浙江工业大学 | Sun shadow displacement positioning method based on deep learning |
| CN112184056A (en) * | 2020-10-19 | 2021-01-05 | 中国工商银行股份有限公司 | Data feature extraction method and system based on convolutional neural network |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2005013537A2 (en) * | 2003-07-28 | 2005-02-10 | Cetacean Networks, Inc. | Systems and methods for the synchronization of a real-time scheduled packet network using relative timing |
| CN102075014A (en) * | 2011-01-06 | 2011-05-25 | 清华大学 | Large grid real-time scheduling method for accepting access of wind power |
| US8762190B1 (en) * | 2012-12-21 | 2014-06-24 | PagerDuty, Inc. | Realtime schedule management interface |
| CN105844350A (en) * | 2016-03-21 | 2016-08-10 | 广西电网有限责任公司电力科学研究院 | Short period wind power prediction system based on covariance preferable combination model |
| CN106210727A (en) * | 2016-08-16 | 2016-12-07 | 广东中星电子有限公司 | Video spatial scalable code stream coded method based on neural network processor array and framework |
| CN106650982A (en) * | 2016-08-30 | 2017-05-10 | 华北电力大学 | Depth learning power prediction method based on multi-point NWP |
| CN106940790A (en) * | 2017-03-13 | 2017-07-11 | 重庆文理学院 | A kind of flow congestion's Forecasting Methodology and system |
-
2017
- 2017-12-28 CN CN201711497688.9A patent/CN108197809B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2005013537A2 (en) * | 2003-07-28 | 2005-02-10 | Cetacean Networks, Inc. | Systems and methods for the synchronization of a real-time scheduled packet network using relative timing |
| CN102075014A (en) * | 2011-01-06 | 2011-05-25 | 清华大学 | Large grid real-time scheduling method for accepting access of wind power |
| US8762190B1 (en) * | 2012-12-21 | 2014-06-24 | PagerDuty, Inc. | Realtime schedule management interface |
| CN105844350A (en) * | 2016-03-21 | 2016-08-10 | 广西电网有限责任公司电力科学研究院 | Short period wind power prediction system based on covariance preferable combination model |
| CN106210727A (en) * | 2016-08-16 | 2016-12-07 | 广东中星电子有限公司 | Video spatial scalable code stream coded method based on neural network processor array and framework |
| CN106650982A (en) * | 2016-08-30 | 2017-05-10 | 华北电力大学 | Depth learning power prediction method based on multi-point NWP |
| CN106940790A (en) * | 2017-03-13 | 2017-07-11 | 重庆文理学院 | A kind of flow congestion's Forecasting Methodology and system |
Non-Patent Citations (1)
| Title |
|---|
| 赵霞，张荣荣，赵瑞锋，颜伟，余娟: "CPS标准下AGC机组动态优化调度的", 《电工技术学报》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110599496A (en) * | 2019-07-30 | 2019-12-20 | 浙江工业大学 | Sun shadow displacement positioning method based on deep learning |
| CN112184056A (en) * | 2020-10-19 | 2021-01-05 | 中国工商银行股份有限公司 | Data feature extraction method and system based on convolutional neural network |
| CN112184056B (en) * | 2020-10-19 | 2024-02-09 | 中国工商银行股份有限公司 | Data feature extraction method and system based on convolutional neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108197809B (en) | 2021-06-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Fernando et al. | Runoff forecasting using RBF networks with OLS algorithm | |
| CN111192270A (en) | Point cloud semantic segmentation method based on point global context reasoning | |
| CN115018021B (en) | Machine room abnormity detection method and device based on graph structure and abnormity attention mechanism | |
| CN110222760B (en) | Quick image processing method based on winograd algorithm | |
| CN107220734A (en) | CNC Lathe Turning process Energy Consumption Prediction System based on decision tree | |
| CN110245709A (en) | Based on deep learning and from the 3D point cloud data semantic dividing method of attention | |
| CN107705806A (en) | A kind of method for carrying out speech emotion recognition using spectrogram and deep convolutional neural networks | |
| CN110196928B (en) | Fully parallelized end-to-end multi-turn dialogue system with domain expansibility and method | |
| Wang et al. | Process cost modelling using neural networks | |
| CN110264079A (en) | Hot-rolled product qualitative forecasting method based on CNN algorithm and Lasso regression model | |
| CN108197809A (en) | Weights based on dimension optimum translation share the real-time scheduling method of depth network | |
| CN107578822B (en) | Pretreatment and feature extraction method for medical multi-modal big data | |
| CN105844334A (en) | Radial basis function neural network-based temperature interpolation algorithm | |
| CN110289987B (en) | Multi-agent system network anti-attack capability assessment method based on characterization learning | |
| CN113534678B (en) | Migration method from simulation of operation question-answering task to physical system | |
| CN110532545A (en) | A kind of data information abstracting method based on complex neural network modeling | |
| Zuo et al. | Domain selection of transfer learning in fuzzy prediction models | |
| CN116030537B (en) | Three-dimensional human body posture estimation method based on multi-branch attention-seeking convolution | |
| CN110691319B (en) | Method for realizing high-precision indoor positioning of heterogeneous equipment in self-adaption mode in use field | |
| CN109816103A (en) | A kind of PSO-BFGS neural network BP training algorithm | |
| CN111880489A (en) | Regression scheduling method for complex manufacturing system | |
| CN116403054A (en) | Image optimization classification method based on brain-like network model | |
| Zhiyuan et al. | Research on the evaluation of enterprise competitiveness based on the wavelet neural network forecasting system | |
| CN112990618A (en) | Prediction method based on machine learning method in industrial Internet of things | |
| CN110322037A (en) | Method for predicting and device based on inference pattern |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |