CN106570592A - Artificial-neural-network-based intelligent numerical value forecasting correction system - Google Patents
Artificial-neural-network-based intelligent numerical value forecasting correction system Download PDFInfo
- Publication number
- CN106570592A CN106570592A CN201610980927.5A CN201610980927A CN106570592A CN 106570592 A CN106570592 A CN 106570592A CN 201610980927 A CN201610980927 A CN 201610980927A CN 106570592 A CN106570592 A CN 106570592A
- Authority
- CN
- China
- Prior art keywords
- aqi
- correcting
- node
- layer
- formula
- 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
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Computational Linguistics (AREA)
- Development Economics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Game Theory and Decision Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an artificial-neural-network-based intelligent numerical value forecasting correction system. The system is erected based on CUDA architecture, a serial program is processed by a CPU, and a parallel program is processed by a GPU. The system comprises a time module, a space module and a sharp drop module. The time module carries out correction to obtain a time correction value, the space module carries out correction to obtain a space correction value, and the sharp drop module carries out correction to obtain an experience correction value; and weighted average processing is carried out on the time correction value, the space correction value, and the experience correction value to obtain correction results of all grids in a forecasting area. Using the system disclosed by the invention, the software execution efficiency can be improved substantially and the operation time and fund cost can be reduced; after correction of the time module, the space module, and the sharp drop module, defects that the correction efficiency is low, all meteorological elements are independent of each other in correction and changing trend differences of different areas are neglected according to the traditional method can be overcome. Therefore, the correction efficiency can be improved; and the high-resolution correction service with deep analysis can be provided for refined grid point forecasting in future.
Description
Technical field
It is the present invention relates to numerical weather forecast result Correction Technology field more particularly to a kind of based on artificial neural network
Intelligent numerical forecast corrects system.
Background technology
At present, the statistics correction method for using in a large number in weather forecasts services, is entered by the historical data in middle or short term
A certain item meteorological element variation tendency at no distant date is found as basis is corrected, after capable simply linear regression in conjunction with forecaster
Experience means are corrected to the routine that forecast result carries out experiential modification.However, the revision means of routine mainly have four shortcomings:
First, the average data of forecast area is processed in statistical disposition, each key element of air rule over time has only been caught,
And have ignored spatial diversity;Second, each meteorological element is independently considered, is corrected respectively, and have ignored tight between some key elements
Close contact, such as precipitation and aerosol concentration, wind speed and air quality index, relative humidity and cloud amount etc.;3rd, it is excessive to rely on
Artificial experience, the error in judgement of forecaster will greatly improve rate of false alarm;4th, process is corrected for larger section, it is impossible to suitable
Answer the resolution of following superfinishing refinement lattice point forecast.
The content of the invention
For weak point present in above-mentioned technology, the present invention provides a kind of high-precision based on artificial neural network
Intelligent numerical forecast corrects system.
In order to achieve the above object, the present invention corrects for a kind of intelligent numerical forecast based on artificial neural network and is
System, is built using CUDA frameworks, and serial program transfers to CPU process, concurrent program to transfer to GPU process, and including following three big moulds
Block:
Time module:The linear relationship of each meteorological element and target component is found to forecast district by the method for statistical analysiss
Domain carries out correcting for integration;
Space module:Using the hidden of the method for artificial neural network deep learning, each meteorological element of analysis and target component
Sexual intercourse, and carry out the differentiation of pointwise and correct;
Steep drop module:Based on the experience of local forecast person, to specific meteorological element given threshold, once reach threshold value some
Specifically correcting rule will be activated, and artificially be adjusted to correcting result;
The time module obtains the time amount of correcting after correcting, the space module must be to the space amount of correcting, institute after correcting
State after steep drop module is corrected and obtain the experience amount of correcting, and the time amount of correcting, the space amount of correcting and the experience amount of correcting three carry out adding
Weight average, obtains correcting result for interior each grid of forecast area.
Wherein, the time module corrects detailed process and is:
The first historical data forecast in first three day is chosen as foundation is corrected, different weather key element and air quality are referred to
Number AQI carries out correlation analysiss, the sample for reference that the Weather Elements checked by correlation analysiss are retained as correcting;Pass through
Method of least square is fitted respectively to the linear relationship between different weather key element and AQI, such as formula (1):
In formula (1), x and y is respectively two groups of variables being fitted;
A certain meteorological element and AQI, coefficient R can be tried to achieve by formula (2), and formula (2) is:
The meteorological field at forecast moment is obtained by numerical simulation, is obtained and each reference further according to the linear relationship set up before
The predictive value AQI of the corresponding AQI of key elementi, by AQIiMake difference with the AQI of observation and obtain Δ AQIi, further according to each reference feature with
The coefficient R of AQI as weight, to Δ AQIiTime module being obtained after being weighted averagely, parameter, Δ is corrected to AQI
AQI。
Wherein, the space module includes input layer, output layer, the first hidden layer and the second hidden layer, each Weather Elements
Pass order is followed successively by input layer, the first hidden layer, the second hidden layer and output layer, each node of last layer and next layer
There is a corresponding weighting function w between each node, it is neural that data follow sigmoid in every one-level going down
First excitation function, formula (3), (4) are as follows:
Ij=∑iwijxi+θj (4)
Wherein, xiFor last layer i-node, xjFor the data of next layer of j node;wijFor last layer i-node to next layer of j
Weight when node is transmitted, θjFor last layer i-node to next layer of j node transmit when bias amount.
Wherein, the space module needs the second historical data to be trained, and selected second historical data be 1 year and
Data set above, for each Weather Elements is normalized respectively in the second historical data, such as formula (5):
Y in formula1For the result after normalized, x1For the data before normalized, by the training after normalized
In data input neutral net, the initial value of w and θ is set to the random number between [- 1,1].
Wherein, the number of nodes of second hidden layer is arranged according to formula (6):
In formula, m is the second node in hidden layer;N is the first node in hidden layer;α is a certain constant of 1-10.
Wherein, the AQI of prediction for the first time is obtained using untrained artificial neural networkp1With actual AQIt1Between mistake
Difference Erro1, such as formula (7):
Erro1=AQIP1(1-AQIP1)(AQIt1-AQIP1) (7)
After trying to achieve output layer error, the error in data Err of each node of the second hidden layer is calculatedjo1, such as formula (8):
Errjo1=AQIP1(1-AQIP1)∑kErro1wjo (8)
Wherein wjoFor weight from the second hidden layer j nodes to output node;According to the error in data meter of second layer hidden layer
Calculate the error in data of the first hidden layer.
Wherein, obtaining between the first hidden layer, the second hidden layer and output layer after the error in data of each node, adjustment is each
Weight w between nodeijWith bias amount θj, can be calculated according to formula (9):
Δwij=(l) Errjoi (9)
Wherein l is learning rate;ErrjFor the error in data of j nodes in next layer;oiFor the data of i-node in last layer.
Bias variation delta θjThen can be calculated according to formula (10):
Δθj=(l) Errj (10)
Weight and bias amount after adjustment is all that the numerical value of its script is added with its variable quantity and obtains.
Wherein, after the weight of all nodes and bias amount are all adjusted and terminated, first cycle of training just completes, and imports second
Individual training sample, training is iterated to neutral net;After neutral net reaches required precision, you can by the pre- of numerical model
Report result is input into the amount of the correcting Δ AQI that the neutral net obtains AQI.
The invention has the beneficial effects as follows:
Compared with prior art, it is of the invention that system is corrected based on the intelligent numerical forecast of artificial neural network, adopt
CUDA frameworks are built, and by CPU process, by GPU process, can increase substantially software performs effect to concurrent program to serial program
Rate, reduces operation time and fund cost;After being corrected by time module, space module and steep drop module, the time of obtaining is corrected
Amount, the space amount of correcting and the experience amount of correcting, are weighted flat to the time amount of correcting, the space amount of correcting and the experience amount of correcting three
, obtain correcting result for each grid in forecast area, overcome and correct in traditional method that efficiency is low, correct in each gas
As the shortcomings of key element is separate to be considered and ignore the variation tendency difference of different regions, the basis of accuracy rate can be corrected in raising
Upper raising corrects efficiency, reduces person works' amount, provides high-resolution, depth analysis and corrects for the future lattice point forecast that becomes more meticulous
Service.
Description of the drawings
Fig. 1 is that the present invention builds signal based on the CUDA frameworks that the intelligent numerical forecast of artificial neural network corrects system
Figure;
Fig. 2 is the square frame software overall architecture that the present invention corrects system based on the intelligent numerical forecast of artificial neural network
Figure;
Fig. 3 is the side of correcting of the space module that the present invention corrects system based on the intelligent numerical forecast of artificial neural network
Method schematic diagram;
Fig. 4 be the present invention based on artificial neural network intelligent numerical forecast correct system neutral net accuracy rate and
The graph of a relation of number of training.
Main element symbol description is as follows:
1st, time module 2, space module
3rd, module drops suddenly.
Specific embodiment
In order to more clearly state the present invention, the present invention is further described below in conjunction with the accompanying drawings.
Refering to Fig. 1-3, a kind of intelligent numerical forecast based on artificial neural network of the present invention corrects system, using CUDA
Framework is built, and serial program transfers to CPU process, concurrent program to transfer to GPU process, and including following three big modules:
Time module 1:The linear relationship of each meteorological element and target component is found to forecast by the method for statistical analysiss
Region carries out correcting for integration;
Space module 2:Using the hidden of the method for artificial neural network deep learning, each meteorological element of analysis and target component
Sexual intercourse, and carry out the differentiation of pointwise and correct;
Steep drop module 3:Based on the experience of local forecast person, to specific meteorological element given threshold, once reach threshold value
Specifically correcting rule a bit will be activated, and artificially be adjusted to correcting result;
Time module 1 obtains the time amount of correcting after correcting, space module 2 must be to the space amount of correcting after correcting, steep drop module 3
The experience amount of correcting is obtained after correcting, and the time amount of correcting, the space amount of correcting and the experience amount of correcting three are weighted averagely, obtain
Result is corrected for each grid in forecast area.
Fig. 1 is referred to, Fig. 1 is the schematic diagram that the present invention is built using CUDA frameworks, and Cache is Cache,
GPU is image processor;Developed using the CUDA programming models for being exclusively used in CPU and GPU " collaboration process ", wherein CPU bears
Program flow control and serial arithmetic task that duty instruction variation and execution efficiency have high demands, and GPU is then responsible for data volume Pang
The big but more single concurrent operation task of instruction.The lattice point forecast that becomes more meticulous in forecast lattice point quantity in terms of million or even hundred million
In, a large amount of arithmetic cores of GPU can greatly improve the data throughout of concurrent operation, greatly improve the efficiency for correcting work.
Fig. 2 is referred to, the foundation corrected of logarithm value forecast result can be divided into two kinds of major ways:First, meteorology will
Element trend over time within one period for closing on;Second, each meteorological element in different spatial it is interior phase
Guan Xing.By taking the correcting of air quality index AQI as an example, computing flow process is as shown in Figure 2.
Compared with prior art, the present invention is to correct system based on the intelligent numerical forecast of artificial neural network, is adopted
CUDA frameworks are built, and can increase substantially the execution efficiency of software, reduce operation time and fund cost;By time module
1st, after space module 2 and steep drop module 3 are corrected, the time amount of correcting, the space amount of correcting and the experience amount of correcting are obtained, the time is corrected
Amount, the space amount of correcting and the experience amount of correcting three are weighted averagely, obtain correcting knot for interior each grid of forecast area
Really, overcome and correct that efficiency is low in traditional method, correct in the separate consideration of each meteorological element and ignore the change of different regions
The shortcomings of change trend difference, can improve to correct improved on the basis of accuracy rate and correct efficiency, reduce person works' amount, can be future
The lattice point that becomes more meticulous forecast provides high-resolution, depth analysis and corrects service.
With air quality index AQI as embodiment, certainly, correcting for time module 1 is not limited to air to the present embodiment
Performance figure AQI is corrected, it is also possible to which correcting for other indexes, time module 1 corrects detailed process and be:
The first historical data forecast in first three day is chosen as foundation is corrected, different weather key element and air quality are referred to
Number AQI carries out correlation analysiss, the sample for reference that the Weather Elements checked by correlation analysiss are retained as correcting;Pass through
Method of least square is fitted respectively to the linear relationship between different weather key element and AQI, such as formula (1):
In formula (1), x and y is respectively two groups of variables being fitted;
A certain meteorological element and AQI, coefficient R can be tried to achieve by formula (2), and formula (2) is:
The meteorological field at forecast moment is obtained by numerical simulation, is obtained and each reference further according to the linear relationship set up before
The predictive value AQI of the corresponding AQI of key elementi, by AQIiMake difference with the AQI of observation and obtain Δ AQIi, further according to each reference feature with
The coefficient R of AQI as weight, to Δ AQIiTime module 1 being obtained after being weighted averagely, parameter, Δ is corrected to AQI
AQI。
Fig. 3 is referred to, space module 2 includes input layer, output layer, the first hidden layer and the second hidden layer, and each weather will
Plain pass order is followed successively by input layer, the first hidden layer, the second hidden layer and output layer, each node and next layer of last layer
Each node between have a corresponding weighting function w, it is refreshing that data follow sigmoid in every one-level going down
Jing units excitation function, formula (3), (4) are as follows:
Ij=∑iwijxi+θj (4)
Wherein, sigmoid be excitation function title, xiFor last layer i-node, xjFor the data of next layer of j node;wijFor
The i-node of last layer to next layer of j node transmit when weight, θjFor last layer i-node to next layer of j node transmit when
Bias amount.
In the present embodiment, space module 2 needs the second historical data to be trained, and selected second historical data is 1 year
And the data set of the above, for each Weather Elements is normalized respectively in the second historical data, such as formula (5):
Y in formula1For the result after normalized, x1For the data before normalized, by the training after normalized
In data input neutral net, the initial value of w and θ is set to the random number between [- 1,1].
In the present embodiment, the number of nodes of the second hidden layer is arranged according to formula (6):
In formula, m is the second node in hidden layer;N is the first node in hidden layer;α is a certain constant of 1-10.
In the present embodiment, the AQI of prediction for the first time is obtained using untrained artificial neural networkp1With actual AQIt1It
Between error E rro1, such as formula (7):
Erro1=AQIP1(1-AQIP1)(AQIt1-AQIP1) (7)
After trying to achieve output layer error, the error in data Err of each node of the second hidden layer is calculatedjo1, such as formula (8):
Errjo1=AQIP1(1-AQIP1)∑kErro1wjo (8)
Wherein wjoFor weight from the second hidden layer j nodes to output node;According to the error in data meter of second layer hidden layer
Calculate the error in data of the first hidden layer.
In the present embodiment, obtaining between the first hidden layer, the second hidden layer and output layer after the error in data of each node,
Adjust weight w between each nodeijWith bias amount θj, can be calculated according to formula (9):
Δwij=(l) Errjoi (9)
Wherein l is learning rate;ErrjFor the error in data of j nodes in next layer;oiFor the data of i-node in last layer.
Bias variation delta θ j then can be calculated according to formula (10):
Δθj=(l) Errj (10)
Weight and bias amount after adjustment is all that the numerical value of its script is added with its variable quantity and obtains.
In the present embodiment, the weight and bias amount of all nodes are all adjusted after terminating, and first cycle of training just completes, and leads
Enter second training sample, training is iterated to neutral net;After neutral net reaches required precision, you can by Numerical-Mode
The forecast result of formula is input into the amount of the correcting Δ AQI that the neutral net obtains AQI.
Fig. 4 is referred to, difference according to the actual requirements, output node layer and the implicit number of plies can all have different degrees of increasing
Plus, therefore the BP neural network that service operation is set up can be more complicated than example above many, but principle is all similar.
In addition, it is contemplated that the different generation Different Effects that can be to each Weather Elements of underlying surface type, therefore can be according to the ground for correcting region
Reason characteristic is respectively directed to the ground surface types such as Plain, hills, mountain region and sets up multiple artificial neural networks, further improves and corrects accurately
Rate.
Carrying out before space module 2 corrects, needing the assessment under different training samples numbers to correct accuracy rate, preferably building
Mould sample number is preferred with slightly above correcting required precision, and rationally arranges neural network learning speed according to the sample size, both
Ensure that accuracy rate also shortens as much as possible cycle of training.
Advantage of the invention is that:
It is a certain area neutral net be once successfully established, can in longer period of time without training again, therefore
The businessization of the system is run during main workload all concentrates on and correct, due to the lattice point quantity of numerical forecast result it is non-
Often huge and each lattice point correcting is independent, therefore corrects process with very high degree of parallelism;CUDA frameworks are exclusive
GPU accelerates the operation time that can significantly shorten process of correcting, so as to greatly improve the ageing of numerical forecast.Assay shows
Show, corrected based on time module 1 and the accuracy rate that can effectively improve forecast in more than 4 hours, wherein time are corrected with space module 2
It is slightly higher that module 1 corrects accuracy rate, but mortality outline is corrected higher than space module 2, particularly short forecasting.Therefore,
Both combinations can improve forecast accuracy, reduce forecast mortality.
Disclosed above is only several specific embodiments of the present invention, but the present invention is not limited to this, any ability
What the technical staff in domain can think change should all fall into protection scope of the present invention.
Claims (8)
1. a kind of intelligent numerical forecast based on artificial neural network corrects system, it is characterised in that taken using CUDA frameworks
Build, serial program transfers to CPU process, concurrent program to transfer to GPU process, and including following three big modules:
Time module:The linear relationship for finding each meteorological element and target component by the method for statistical analysiss enters to forecast area
Row integration is corrected;
Space module:Using the method for artificial neural network deep learning, the recessive pass of each meteorological element and target component is analyzed
It is, and carries out the differentiation of pointwise and corrects;
Steep drop module:Based on the experience of local forecast person, to specific meteorological element given threshold, once some are specific to reach threshold value
Correct rule and will be activated, artificially adjusted to correcting result;
The time module obtains the time amount of correcting after correcting, the space module must be described steep to the space amount of correcting after correcting
Drop module obtains the experience amount of correcting after correcting, and the time amount of correcting, the space amount of correcting and the experience amount of correcting three be weighted it is flat
, obtain correcting result for interior each grid of forecast area.
2. the intelligent numerical forecast based on artificial neural network according to claim 1 corrects system, it is characterised in that
The time module corrects detailed process:
The first historical data forecast in first three day is chosen as foundation is corrected, by different weather key element and air quality index
AQI carries out correlation analysiss, the sample for reference that the Weather Elements checked by correlation analysiss are retained as correcting;By most
Little square law is fitted respectively to the linear relationship between different weather key element and AQI, such as formula (1):
In formula (1), x and y is respectively two groups of variables being fitted;
A certain meteorological element and AQI, coefficient R can be tried to achieve by formula (2), and formula (2) is:
The meteorological field at forecast moment is obtained by numerical simulation, is obtained and each reference feature further according to the linear relationship set up before
The predictive value AQI of corresponding AQIi, by AQIiMake difference with the AQI of observation and obtain Δ AQIi, further according to each reference feature and AQI
Coefficient R as weight, to Δ AQIiTime module being obtained after being weighted averagely, parameter, Δ AQI is corrected to AQI.
3. the intelligent numerical forecast based on artificial neural network according to claim 2 corrects system, it is characterised in that
The space module includes input layer, output layer, the first hidden layer and the second hidden layer, and each Weather Elements pass order is followed successively by
Input layer, the first hidden layer, the second hidden layer and output layer, between each node of last layer and next layer each node
There is a corresponding weighting function w, data follow sigmoid neuron excitation functions, formula in every one-level going down
(3), (4) are as follows:
Ij=∑iwijxi+θj (4)
Wherein, xiFor last layer i-node, xjFor the data of next layer of j node;wijFor last layer i-node to next layer of j node
Weight during transmission, θjFor last layer i-node to next layer of j node transmit when bias amount.
4. the intelligent numerical forecast based on artificial neural network according to claim 3 corrects system, it is characterised in that
The space module needs the second historical data to be trained, and selected second historical data is the data set of a year and the above,
For each Weather Elements is normalized respectively in the second historical data, such as formula (5):
Y in formula1For the result after normalized, x1For the data before normalized, by the training data after normalized
In input neutral net, the initial value of w and θ is set to the random number between [- 1,1].
5. the intelligent numerical forecast based on artificial neural network according to claim 3 corrects system, it is characterised in that
The number of nodes of second hidden layer is arranged according to formula (6):
In formula, m is the second node in hidden layer;N is the first node in hidden layer;α is a certain constant of 1-10.
6. the intelligent numerical forecast based on artificial neural network according to claim 3 corrects system, it is characterised in that
The AQI of prediction for the first time is obtained using untrained artificial neural networkp1With actual AQIt1Between error E rro1, such as formula
(7):
Erro1=AQIP1(1-AQIP1)(AQIt1-AQIP1) (7)
After trying to achieve output layer error, the error in data Err of each node of the second hidden layer is calculatedjo1, such as formula (8):
Errjo1=AQIP1(1-AQIP1)∑kErro1wjo (8)
Wherein wjoFor weight from the second hidden layer j nodes to output node;The is calculated according to the error in data of second layer hidden layer
The error in data of one hidden layer.
7. the intelligent numerical forecast based on artificial neural network according to claim 6 corrects system, it is characterised in that
Obtaining between the first hidden layer, the second hidden layer and output layer after the error in data of each node, adjusting weight w between each nodeij
With bias amount θj, can be calculated according to formula (9):
Δwij=(l) Errjoi (9)
Wherein l is learning rate;ErrjFor the error in data of j nodes in next layer;oiFor the data of i-node in last layer.Bias
Variation delta θjThen can be calculated according to formula (10):
Δθj=(l) Errj (10)
Weight and bias amount after adjustment is all that the numerical value of its script is added with its variable quantity and obtains.
8. the intelligent numerical forecast based on artificial neural network according to claim 7 corrects system, it is characterised in that
The weight and bias amount of all nodes is all adjusted after terminating, and first cycle of training just completes, and imports second training sample, right
Neutral net is iterated training;After neutral net reaches required precision, you can should by the forecast result input of numerical model
Neutral net obtains the amount of the correcting Δ AQI of AQI.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610980927.5A CN106570592A (en) | 2016-11-08 | 2016-11-08 | Artificial-neural-network-based intelligent numerical value forecasting correction system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610980927.5A CN106570592A (en) | 2016-11-08 | 2016-11-08 | Artificial-neural-network-based intelligent numerical value forecasting correction system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106570592A true CN106570592A (en) | 2017-04-19 |
Family
ID=58540102
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610980927.5A Pending CN106570592A (en) | 2016-11-08 | 2016-11-08 | Artificial-neural-network-based intelligent numerical value forecasting correction system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106570592A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107609702A (en) * | 2017-09-15 | 2018-01-19 | 郑州云海信息技术有限公司 | A kind of process meteorological data method and device |
CN108399469A (en) * | 2018-01-31 | 2018-08-14 | 中国人民解放军空军研究院战场环境研究所 | A kind of weather phenomenon forecasting procedure based on deep learning and numerical weather forecast |
WO2019128514A1 (en) * | 2017-12-29 | 2019-07-04 | 阿里巴巴集团控股有限公司 | Method and device for pushing information and method and device for determining default input value |
RU2696121C1 (en) * | 2018-07-13 | 2019-07-31 | федеральное государственное бюджетное образовательное учреждение высшего образования "Нижегородский государственный технический университет им. Р.Е. Алексеева" (НГТУ) | Method for 3d printing on cnc equipment with intelligent optimization of modes |
CN111736148A (en) * | 2020-06-28 | 2020-10-02 | 国家海洋环境预报中心 | Method for correcting sea wave effective wave height of satellite radar altimeter and related device |
CN114330641A (en) * | 2021-11-09 | 2022-04-12 | 国网山东省电力公司应急管理中心 | Method for establishing short-term wind speed correction model based on deep learning of complex terrain |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105631537A (en) * | 2015-12-23 | 2016-06-01 | 南京信息工程大学 | Air quality forecast service system based on meteorological service platform |
-
2016
- 2016-11-08 CN CN201610980927.5A patent/CN106570592A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105631537A (en) * | 2015-12-23 | 2016-06-01 | 南京信息工程大学 | Air quality forecast service system based on meteorological service platform |
Non-Patent Citations (1)
Title |
---|
程水源 等: "《大气重污染过程预测技术研究》", 30 June 2014, 北京:中国环境科学出版社 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107609702A (en) * | 2017-09-15 | 2018-01-19 | 郑州云海信息技术有限公司 | A kind of process meteorological data method and device |
WO2019128514A1 (en) * | 2017-12-29 | 2019-07-04 | 阿里巴巴集团控股有限公司 | Method and device for pushing information and method and device for determining default input value |
US11087402B2 (en) | 2017-12-29 | 2021-08-10 | Advanced New Technologies Co., Ltd. | Method and device for pushing information and method and device for determining default input value |
CN108399469A (en) * | 2018-01-31 | 2018-08-14 | 中国人民解放军空军研究院战场环境研究所 | A kind of weather phenomenon forecasting procedure based on deep learning and numerical weather forecast |
RU2696121C1 (en) * | 2018-07-13 | 2019-07-31 | федеральное государственное бюджетное образовательное учреждение высшего образования "Нижегородский государственный технический университет им. Р.Е. Алексеева" (НГТУ) | Method for 3d printing on cnc equipment with intelligent optimization of modes |
CN111736148A (en) * | 2020-06-28 | 2020-10-02 | 国家海洋环境预报中心 | Method for correcting sea wave effective wave height of satellite radar altimeter and related device |
CN111736148B (en) * | 2020-06-28 | 2021-06-08 | 国家海洋环境预报中心 | Method for correcting sea wave effective wave height of satellite radar altimeter and related device |
CN114330641A (en) * | 2021-11-09 | 2022-04-12 | 国网山东省电力公司应急管理中心 | Method for establishing short-term wind speed correction model based on deep learning of complex terrain |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106570592A (en) | Artificial-neural-network-based intelligent numerical value forecasting correction system | |
CN111310889B (en) | Evaporation waveguide profile estimation method based on deep neural network | |
Zhang et al. | Short-term rainfall forecasting using multi-layer perceptron | |
CN108009674A (en) | Air PM2.5 concentration prediction methods based on CNN and LSTM fused neural networks | |
CN109657881A (en) | A kind of neural network photovoltaic power generation prediction technique and system suitable for small sample | |
CN102799772B (en) | Towards the sample optimization method of prediction of air quality | |
CN104616060B (en) | Insulator dirty degree Forecasting Methodology based on BP neural network and fuzzy logic | |
CN111783987A (en) | Farmland reference crop evapotranspiration prediction method based on improved BP neural network | |
CN107346459B (en) | Multi-mode pollutant integrated forecasting method based on genetic algorithm improvement | |
CN106022954A (en) | Multiple BP neural network load prediction method based on grey correlation degree | |
CN112288164B (en) | Wind power combined prediction method considering spatial correlation and correcting numerical weather forecast | |
Araujo et al. | Multilayer perceptron neural network for flow prediction | |
CN109635245A (en) | A kind of robust width learning system | |
CN107688825A (en) | A kind of follow-on integrated weighting extreme learning machine sewage disposal failure examines method | |
CN109143408B (en) | Dynamic region combined short-time rainfall forecasting method based on MLP | |
CN114626512A (en) | High-temperature disaster forecasting method based on directed graph neural network | |
CN111126704A (en) | Multi-region precipitation prediction model construction method based on multi-graph convolution and memory network | |
CN104933483A (en) | Wind power forecasting method dividing based on weather process | |
AU2021106540A4 (en) | Prediction method and system for river algal blooms | |
CN109345027A (en) | Micro-capacitance sensor short-term load forecasting method based on independent component analysis and support vector machines | |
CN111242380A (en) | Lake (reservoir) eutrophication prediction method based on artificial intelligence algorithm | |
Song et al. | Study on turbidity prediction method of reservoirs based on long short term memory neural network | |
CN105844334B (en) | A kind of temperature interpolation method based on radial base neural net | |
CN109214591B (en) | Method and system for predicting aboveground biomass of woody plant | |
CN107665376A (en) | A kind of Wetland Space changes in distribution framework analogue and Forecasting Methodology |
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: 20170419 |
|
RJ01 | Rejection of invention patent application after publication |