CN110309037A - A kind of selection method of data center's efficiency correlated characteristic - Google Patents
A kind of selection method of data center's efficiency correlated characteristic Download PDFInfo
- Publication number
- CN110309037A CN110309037A CN201811469430.2A CN201811469430A CN110309037A CN 110309037 A CN110309037 A CN 110309037A CN 201811469430 A CN201811469430 A CN 201811469430A CN 110309037 A CN110309037 A CN 110309037A
- Authority
- CN
- China
- Prior art keywords
- sample
- feature
- data center
- pue
- efficiency
- 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.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/30—Monitoring
- G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
- G06F11/3409—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment for performance assessment
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/30—Monitoring
- G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
- G06F11/3452—Performance evaluation by statistical analysis
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Hardware Design (AREA)
- General Physics & Mathematics (AREA)
- Quality & Reliability (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Probability & Statistics with Applications (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention proposes a kind of selection methods of data center's efficiency correlated characteristic, for the feature selection issues of data center's efficiency, present invention employs a kind of feature selection approach based on k nearest neighbor Classification Loss function and class interval, this method is by collecting consumption of data center data and corresponding PUE value, then PUE value is classified, corresponding class interval is found by sample, and updates feature weight and sorts to feature weight, feature selecting result is obtained according to the threshold value of setting with this.The method of the invention can extract feature relevant to data center's efficiency and handle noise data well, to improve the precision of subsequent efficiency prediction, effectively prevent overfitting.
Description
Technical field
The invention belongs to cloud computings and machine learning, and in particular to a kind of selecting party of data center's efficiency correlated characteristic
Method.
Background technique
Data center is the infrastructure for executing round-the-clock extensive key operation task, is the weight for supporting the operating of IT industry
Want facility.Demand with the large-scale cloud service of network operator and Internet company to data calculating, processing and storage is not
Disconnected to increase, the large-scale data center for possessing thousands of servers is increased sharply.Secondly, the cloud of high performance computation is with network
The dilatation of bandwidth and continue to develop, this expand building large-scale calculations architecture demand.Therefore, data center becomes fast
One of the critical infrastructures of the IT industry of speed development.
In recent years, due to the high economic benefit and environmental dependence of data center, the optimization of the energy efficiency of data center
Problem has become most important.Firstly, data center brings many economic benefits, this makes the scale sum number of data center
Amount also constantly increases.With the rising of electricity consumption sharply increased with power cost, the electricity charge have become current data center
Capital expenditure.In some cases, the power cost of data center could possibly be higher than the cost of original capital investment.Secondly, data
The energy use at center can generate many environmental problems, such as the greenhouse gases row of a large amount of power consumption, air-conditioning refrigeration equipment
Put the discharge with cooling water.Even if the server of data center is in idle condition, a large amount of energy can be equally consumed.Out
In these reasons, its energy efficiency needs are paid the utmost attention to during data center's operation at present.
The most frequently used index for measuring data center's efficiency is energy use efficiency, i.e. PUE.The definition of this index is input
The energy consumption that the total energy consumption of data center is used divided by information technoloy equipment.Total energy consumption includes the energy consumption that uses of information technoloy equipment plus any non-computational
Any expense power consumption consumed by equipment (i.e. cooling, lighting apparatus etc.) with data communication purposes.If the PUE value of data center
It is 2.0, it means that the energy consumption of every 1 watt of information technoloy equipment of the supply of the facility, other non-information technoloy equipments can also consume 1 watt of energy consumption.Most
Ideal PUE is 1.0, i.e., the hypothesis situation in addition to information technoloy equipment without other energy consumptions.This kind of situation is that can not reach in practical applications
It arrives, so PUE all makes every effort in advanced data center levels off to 1.0.
Based on the above situation, the efficiency forecasting problem for solving data center is extremely urgent, this problem becomes the country
Outer research hotspot.And the core missions in efficiency prediction first is that select relevant to data center's efficiency determinant attribute (special
Sign).Current most of efficiency forecasting researches are all based on extremely simple data center model to realize, such as simple server
Cpu frequency or performance counter index, therefore feature selecting is easier.And for large-scale data center, feature is various
And it is complicated, the research of correlated characteristic selection is fewer, and only model is all based on the black-box model of deep neural network mostly, can
Explanatory difference.
Summary of the invention
Goal of the invention: insufficient for selectivity of the above-mentioned prior art for characteristic, the present invention provides a kind of data
The selection method of center efficiency correlated characteristic can carry out the selection of correlated characteristic for all data centers.
A kind of technical solution: selection method of data center's efficiency correlated characteristic, comprising the following steps:
(1) consumption of data center data and corresponding PUE value are collected；
(2) PUE value is classified by grade scale；
(3) it randomly chooses sample and searches its k nearest neighbor, while calculating class interval corresponding to the sample；
(4) it establishes based on Classification Loss-interval feature selecting interpretational criteria；
(5) feature weight is updated by the designed interpretational criteria of gradient decline optimization；
(6) it sorts to feature weight, and obtains feature selecting result by given threshold.
Further, since the PUE that step 1 obtains is successive value, so needing to convert PUE to by grade scale
Discrete value.PUE value is classified by grade scale in the step (2), every number is calculated according to efficiency hierarchical table
According to xiCorresponding PUE grade yi∈ { 1,2,3 }, xiIndicate the n D feature vectors of the i-th data, x thereinijThen indicate i-th
J-th of real number characteristic value of data, expression formula are as follows:
Random selection sample described in step (3) simultaneously searches its k nearest neighbor, while calculating class interval corresponding to the sample
Specific step is as follows:
(31) two-dimentional two-value label corresponding relationship matrix B and target neighbor relational matrix T, element in the matrix B are obtained
bij∈ { 0,1 } indicates PUE grade yiAnd yjIt is whether identical, element t in matrix Tij∈ { 0,1 } indicates sample xjIt whether is xiMesh
Mark neighbour；
(32) it is and x by the definition of target neighboriThe similar sample of the identical k nearest neighbor of PUE grade, wherein K > 2；
(33) selection sample x is not put back to from N samplei, find and sample xiArest neighbors and the identical sample of PUE grade
nearhit(xi) and with sample xiArest neighbors and the different sample nearmiss (x of PUE gradei), and calculate class interval θi, use
Formula indicates are as follows:
θi=| ‖ xi-nearmiss(xi)‖2-‖xi-nearhit(xi)‖2|。
Step (4) includes by sample xiLoss function L based on feature weight ws(w,xi) as feature selecting evaluation it is quasi-
Then, is defined as:
Wherein, c is normal number, is usually obtained by cross validation；H is hinge loss, is indicated are as follows:
[a]+=max (a, 0)
Wherein, the Euclidean distance calculation formula of weighting are as follows:
The step (5) includes the gradient for calculating the loss function of each feature fIt finally obtains about all features
N ties up gradient vectorPass throughUpdate feature weight vector w；It is as follows for the loss function gradient calculation expression of feature f:
Wherein, the gradient of hinge loss is defined as follows:
g(wf)=2wf((xif-xjf)2-(xif-xpf)2)
Feature weight vector w more new formula is as follows:
Finally based on set the number of iterations, repeat step (3)-step (5).
The step (6) includes that final character subset, institute are determined by given threshold after feature sorts by weight w
Stating all features in character subset is key feature relevant to data center's efficiency.
The utility model has the advantages that compared with prior art, the present invention its significant effect is: first, the present invention only needs to calculate n
The weight of a feature simultaneously sorts them, lower relative to conventional method computation complexity, and can effectively prevent overfitting；The
Two, it is of the present invention that cloud data center noise data that may be present can preferably be handled based on k nearest neighbor algorithm, improve number
According to accuracy.
Detailed description of the invention
Fig. 1 is structural schematic diagram of the invention；
Fig. 2 is PUE structural schematic diagram；
Fig. 3 is embodiment class interval θiIndicate figure.
Specific embodiment
In order to which technical solution disclosed in this invention is described in detail, with reference to the accompanying drawings of the specification and specific embodiment is done
It is further elucidated above.
Firstly, being described below about correlated variables according to the present invention:
Assuming that having collected N data power consumption data and its PUE value, indicate are as follows:
Wherein xiIndicate the n D feature vectors of the i-th data, xijThen indicate the reality of j-th of characteristic of the i-th data
Number characteristic value, i.e.,
ziIt is then expressed as the corresponding PUE value of the i-th data, according to ziCorresponding PUE grade y can be obtainedi∈ { 1,2,3 }, because
And new sample can be obtained, it indicates are as follows:
The weight vectors that w is made of the weight of each feature in original efficiency data set, wherein at the beginning of each feature weight
Initial value is 1.
With xiThe similar sample of the identical k nearest neighbor of PUE grade is defined as target neighbor, wherein K > 2.
The distance that the present invention hereinafter mentions represents Euclidean distance.
Provided by the present invention is a kind of feature selection approach for all data center's forecasting problems, and process is as schemed
Shown in 1, the specific steps are as follows:
Step 1: the energy consumption data at collection N data center and its corresponding PUE value.
The feature of the consumption of data center data of acquisition is as follows: the total IT load of server；The total IT load in core network room；
Operational process water pump sum；Process pump variable frequency device average speed；Condensate pump sum；Condensate pump frequency converter average speed；
Run cooling tower sum；Cooling tower is discharged average set temperature；Run cooling-water machine sum；Run dry and cold machine sum；Operation freezing
Water water injecting pump sum；Chilled water water injecting pump is averaged set temperature；The mean temperature of heat exchanger；Outdoor air wet bulb temperature；It is outdoor
Air dry-bulb temperature；Outdoor air enthalpy；Outdoor air relative humidity；Outdoor wind speed；Outdoor wind direction etc..Different data centers
Feature that can be different according to the different acquisition of equipment or layout.
Step 2: every is obtained according to efficiency (PUE) hierarchical table that DB11/T 1139-2014 standard proposes
Data xiCorresponding PUE grade yi.The structural schematic diagram of PUE classification is as shown in Fig. 2, PUE hierarchical table is as shown in table 1.
Table 1PUE hierarchical table
| Rank | I grades | II grade | III grade |
| PUE value | 1 PUE≤1.5 < | 1.5 PUE≤1.8 < | 1.8 PUE≤2.0 < |
Such as the corresponding PUE value of certain data for the data center being collected into is 1.4, then available data center
Corresponding PUE grade is 1.
Step 3: random selection sample xiAnd its k nearest neighbor is searched, while calculating the corresponding class interval θ of the samplei.Specifically
, it is necessary first to obtain two-dimentional two-value label corresponding relationship matrix B and target neighbor relational matrix T；Wherein, element in matrix B
bij∈ { 0,1 } indicates PUE grade yiAnd yjIt is whether identical, i.e., as sample xiWith sample xjCorresponding PUE grade yi,yjWhen identical,
bij=1, different then bij=0；Element t in matrix Tij∈ { 0,1 } indicates sample xjIt whether is xiTarget neighbor, i.e., as sample xj
It is and sample xiDistance most mutually nearby one of K sample and corresponding PUE grade yj,yiWhen identical, tij=1, otherwise, tij=0.
Selection sample x is not put back to from N samplei, find and sample xiArest neighbors and the identical sample of PUE grade
nearhit(xi) and with sample xiArest neighbors and the different sample nearmiss (x of PUE gradei), and calculate class interval θi, such as
Shown in Fig. 3.It is formulated are as follows:
θi=| ‖ xi-nearmiss(xi)‖2-‖xi-nearhit(xi)‖2|
Interval θiIt is expressed as and sample xiThe identical sample nearhit (x of the most close and PUE grade of distancei) and sample xi
Distance square subtract and sample xiSample nearmiss (the x that distance is most close and PUE grade is differenti) and sample xiAway from
From square absolute value.
Step 4: design is based on class interval θiFeature selecting interpretational criteria.
By sample xiLoss function L based on feature weight ws(w,xi) interpretational criteria as feature selecting, definition
Are as follows:
Wherein, c is normal number, is usually obtained by cross validation；H is hinge loss, is expressed as
[a]+=max (a, 0)
Wherein, the distance calculation formula of weighting is
The first item of loss function indicates sample xiTarget neighbor K sample and sample xiWeighted distance square
Sum, pass through more newly arrive minimum and the sample x to feature weightiWeighted distance apart from farther away target neighbor sample；And
Section 2 is then indicated for all sample xiTarget neighbor sample, with sample xiWeighted distance square add sorting room
Every subtracting again and sample xiK most close and different PUE grade sample xpWeighted distance square, if the value less than 0,
Then illustrate sample xpOpposite sample xiDistance has been more than current sample x farther outiTarget neighbor sample and sample xiBetween plus
Power distance adds the size of class interval, so losing by hinge, the value is allowed to be 0；If the value is greater than 0, illustrate sample xp
With sample xiDistance it is closer, need to minimize this by more newly arriving for weight, so that sample xpWith sample xiPlus
Power is apart from farther.
The hinge loss used by loss function Section 2, is converted to soft margin standard for evaluation function, so as to have
Effect reduces the influence of exceptional value.Simultaneously by the way that the K value of target neighbor is set as K > 2, so that can be very well for arest neighbors classification
Filtering noise data.
Step 5: optimizing evaluation criterion is declined by gradient to update feature weight w.
Calculate the gradient of the loss function of each feature fIt finally obtains and ties up gradient vector about the n of all featuresIt is logical
It crossesUpdate feature weight vector w.The loss function gradient of feature f is calculated as follows:
Wherein, the gradient of hinge loss is defined as follows:
g(wf)=2wf((xif-xjf)2-(xif-xpf)2)
Feature weight vector w more new formula is as follows:
Based on set the number of iterations, repeat Step 3: four, five.
The calculating time is mainly concerned with matrix B, the calculating of T and vector w, their time complexity difference in the present invention
For O (N2),O(N2) and O (N2Kn).Usually, K is a small constant.So the method total time complexity in the present invention is
2O(N2)+O(N2n)≈O(N2N), it is better than the time complexity O (N of conventional method2n2)。
Step 6: final character subset is determined by given threshold after feature is sorted by weight w.In character subset
All features are key feature relevant to data center's efficiency.
It is assumed that according to the feature after weight sort descending are as follows: the total IT load of server；Run cooling-water machine sum；Operation cooling
Tower sum；Cooling tower is discharged average set temperature；Operational process water pump sum；Process pump variable frequency device average speed；Operation is dry
Cold sum；Outdoor air wet bulb temperature；Outdoor air enthalpy；The total IT load in core network room；Condensate pump sum；Condensation
Pump variable frequency device average speed；Run chilled water water injecting pump sum；Chilled water water injecting pump is averaged set temperature；Heat exchanger is averaged
Temperature；Outdoor air dry-bulb temperature；Outdoor air relative humidity；Outdoor wind speed；Outdoor wind direction.
Assuming that setting threshold value is 14, then the character subset finally obtained are as follows: the total IT load of server；It is total to run cooling-water machine
Number；Run cooling tower sum；Cooling tower is discharged average set temperature；Operational process water pump sum；Process pump variable frequency device is average
Speed；Run dry and cold machine sum；Outdoor air wet bulb temperature；Outdoor air enthalpy；The total IT load in core network room；Condensed water
Pump sum；Condensate pump frequency converter average speed；Run chilled water water injecting pump sum；Chilled water water injecting pump is averaged set temperature.
Then this feature subset feature that data center's efficiency is predicted as after can be continued subsequent operation.
Claims (6)
1. a kind of selection method of data center's efficiency correlated characteristic, it is characterised in that: the following steps are included:
(1) consumption of data center data and corresponding PUE value are collected；
(2) PUE value is classified by grade scale；
(3) it randomly chooses sample and searches its k nearest neighbor, while calculating class interval corresponding to the sample；
(4) it establishes based on Classification Loss-interval feature selecting interpretational criteria；
(5) feature weight is updated by the designed interpretational criteria of gradient decline optimization；
(6) it sorts to feature weight, and obtains feature selecting result by given threshold.
2. according to a kind of selection method of data center efficiency correlated characteristic described in claim 1, it is characterised in that: the step
(2) PUE value is classified by grade scale in, every data x is calculated according to efficiency hierarchical tableiCorresponding PUE etc.
Grade yi∈ { 1,2,3 }, xiIndicate the n D feature vectors of the i-th data, x thereinijThen indicate j-th of real number of the i-th data
Characteristic value, expression formula are as follows:
3. according to a kind of selection method of data center efficiency correlated characteristic described in claim 1, it is characterised in that: step (3) institute
The random selection sample stated simultaneously searches its k nearest neighbor, while calculating class interval corresponding to the sample specific step is as follows:
(31) two-dimentional two-value label corresponding relationship matrix B and target neighbor relational matrix T, element b in the matrix B are obtainedij∈
{ 0,1 } PUE grade y is indicatediAnd yjIt is whether identical, element t in matrix Tij∈ { 0,1 } indicates sample xjIt whether is xiTarget it is close
It is adjacent；
(32) definition of target neighbor is and xiThe similar sample of the identical k nearest neighbor of PUE grade, wherein K > 2；
(33) selection sample x is not put back to from N samplei, find and sample xiArest neighbors and the identical sample of PUE grade
nearhit(xi) and with sample xiArest neighbors and the different sample nearmiss (x of PUE gradei), and calculate class interval θi, use
Formula indicates are as follows:
θi=| | | xi-nearmiss(xi)||2-||xi-nearhit(xi)||2|。
4. according to a kind of selection method of data center efficiency correlated characteristic described in claim 1, it is characterised in that: step (4) packet
It includes sample xiLoss function L based on feature weight ws(w, xi) interpretational criteria as feature selecting, is defined as:
Wherein, c is normal number, is usually obtained by cross validation；H is hinge loss, is indicated are as follows:
[a]+=max (a, 0)
Wherein, the Euclidean distance calculation formula of weighting are as follows:
5. the selection method of data center's efficiency correlated characteristic according to claim 1, which is characterized in that the step (5)
Gradient including calculating the loss function of each feature fIt finally obtains and ties up gradient vector about the n of all featuresPass through
Update feature weight vector w；It is as follows for the loss function gradient calculation expression of feature f:
Wherein, the gradient of hinge loss is defined as follows:
g(wf)=2wf((xif-xjf)2-(xif-xpf)2)
Feature weight vector w more new formula is as follows:
Finally based on set the number of iterations, repeat step (3) to step (5).
6. a kind of selection method of data center's efficiency correlated characteristic according to claim 1, which is characterized in that the step
It (6) include owning feature in the character subset by final character subset is determined after weight w sequence by given threshold
Feature is key feature relevant to data center's efficiency.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811469430.2A CN110309037A (en) | 2018-11-28 | 2018-11-28 | A kind of selection method of data center's efficiency correlated characteristic |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811469430.2A CN110309037A (en) | 2018-11-28 | 2018-11-28 | A kind of selection method of data center's efficiency correlated characteristic |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110309037A true CN110309037A (en) | 2019-10-08 |
Family
ID=68074192
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811469430.2A Withdrawn CN110309037A (en) | 2018-11-28 | 2018-11-28 | A kind of selection method of data center's efficiency correlated characteristic |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110309037A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110866528A (en) * | 2019-10-28 | 2020-03-06 | 腾讯科技（深圳）有限公司 | Model training method, energy consumption use efficiency prediction method, device and medium |
| CN117035562A (en) * | 2023-10-10 | 2023-11-10 | 云境商务智能研究院南京有限公司 | Environment-friendly intelligent monitoring method based on electric power big data and data analysis equipment |
-
2018
- 2018-11-28 CN CN201811469430.2A patent/CN110309037A/en not_active Withdrawn
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110866528A (en) * | 2019-10-28 | 2020-03-06 | 腾讯科技（深圳）有限公司 | Model training method, energy consumption use efficiency prediction method, device and medium |
| CN110866528B (en) * | 2019-10-28 | 2023-11-28 | 腾讯科技（深圳）有限公司 | Model training method, energy consumption use efficiency prediction method, device and medium |
| CN117035562A (en) * | 2023-10-10 | 2023-11-10 | 云境商务智能研究院南京有限公司 | Environment-friendly intelligent monitoring method based on electric power big data and data analysis equipment |
| CN117035562B (en) * | 2023-10-10 | 2024-01-30 | 云境商务智能研究院南京有限公司 | Environment-friendly intelligent monitoring method based on electric power big data and data analysis equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106845717B (en) | Energy efficiency evaluation method based on multi-model fusion strategy | |
| WO2023131215A1 (en) | Multi-scale aggregation mode analysis method for complex traffic network | |
| CN113378913B (en) | Semi-supervised node classification method based on self-supervised learning | |
| CN108805743A (en) | A kind of power grid enterprises' sale of electricity company operation Benefit Evaluation Method | |
| CN108460486A (en) | A kind of voltage deviation prediction technique based on improvement clustering algorithm and neural network | |
| CN110781068B (en) | Data center cross-layer energy consumption prediction method based on isomorphic decomposition method | |
| CN110309037A (en) | A kind of selection method of data center's efficiency correlated characteristic | |
| CN109492776A (en) | Microblogging Popularity prediction method based on Active Learning | |
| Yang et al. | Linearly decreasing weight particle swarm optimization with accelerated strategy for data clustering | |
| CN106126328B (en) | A kind of traffic metadata management method and system based on event category | |
| CN108664653A (en) | A kind of Medical Consumption client's automatic classification method based on K-means | |
| CN107563220A (en) | A kind of computer based big data analysis and Control system and control method | |
| CN103761286B (en) | A kind of Service Source search method based on user interest | |
| CN112633314A (en) | Active learning source tracing attack method based on multi-layer sampling | |
| Jie et al. | Review on the research of K-means clustering algorithm in big data | |
| CN105844334B (en) | A kind of temperature interpolation method based on radial base neural net | |
| Montusiewicz et al. | Looking for the optimal location for wind farms | |
| CN109376797A (en) | A kind of net flow assorted method based on binary coder and more Hash tables | |
| CN114897085A (en) | Clustering method based on closed subgraph link prediction and computer equipment | |
| Kim et al. | Hier: Metric learning beyond class labels via hierarchical regularization | |
| CN113658109A (en) | Glass defect detection method based on field loss prediction active learning | |
| Družeta et al. | Introducing languid particle dynamics to a selection of PSO variants | |
| Dang et al. | Interval type-2 fuzzy c-means approach to collaborative clustering | |
| CN109919401A (en) | A kind of multidimensional energy efficiency analysis method for air for system of providing multiple forms of energy to complement each other | |
| CN105162648B (en) | Corporations' detection method based on backbone network extension |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20191008 |