CN110505098A - A kind of trans-regional traffic forecasting method based on isomery model reusability - Google Patents
A kind of trans-regional traffic forecasting method based on isomery model reusability Download PDFInfo
- Publication number
- CN110505098A CN110505098A CN201910803496.9A CN201910803496A CN110505098A CN 110505098 A CN110505098 A CN 110505098A CN 201910803496 A CN201910803496 A CN 201910803496A CN 110505098 A CN110505098 A CN 110505098A
- Authority
- CN
- China
- Prior art keywords
- model
- new
- feature
- data
- communication
- 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
- 238000013277 forecasting method Methods 0.000 title claims abstract description 12
- 238000004891 communication Methods 0.000 claims abstract description 60
- 238000012549 training Methods 0.000 claims abstract description 20
- 238000000034 method Methods 0.000 claims abstract description 16
- 238000013480 data collection Methods 0.000 claims abstract description 5
- 239000011159 matrix material Substances 0.000 claims description 13
- 238000012417 linear regression Methods 0.000 claims description 6
- 238000012546 transfer Methods 0.000 claims description 6
- 230000007704 transition Effects 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 abstract description 5
- 230000008569 process Effects 0.000 abstract description 4
- 238000005259 measurement Methods 0.000 abstract 1
- 238000012512 characterization method Methods 0.000 description 2
- 238000013526 transfer learning Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 230000004907 flux Effects 0.000 description 1
- 210000004744 fore-foot Anatomy 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000010287 polarization Effects 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/142—Network analysis or design using statistical or mathematical methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/147—Network analysis or design for predicting network behaviour
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- Algebra (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Mathematical Physics (AREA)
- Probability & Statistics with Applications (AREA)
- Pure & Applied Mathematics (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention discloses a kind of trans-regional traffic forecasting method based on isomery model reusability, including data collection steps, model reusability step, new region model training step and new region model measurement step；Different zones communication data is collected first, and the metamessage including characteristic and feature describes data；This isomery model, is mapped in the feature space of current region by then one model of training in area data before using model reusability and the optimal technology transported；A new model finally is established using a small amount of communication data on new region, to complete to there is one performance of flag data training workable model well on a small quantity.The present invention can solve the difficult point that new and old provincial characteristics space difference and new region only have data to be difficult to set up a new model, and take up less resources in implementation process of the present invention, strong applicability.
Description
Technical field
Method for predicting in the case of changing the present invention relates to different zones communication feature under open dynamic environment,
Specifically a kind of trans-regional traffic forecasting method based on isomery model reusability.
Background technique
In the communications field, how rapidly and accurately to predict a region following a period of time communication flows to Guan Chong
It wants, for example can turn off the service of certain communication base stations in the case that a region coming few hours communication flows is very low, from
And reaching reduces energy consumption, reduces the purpose of cost.However, actual environment is mostly to open dynamically, the base station facility of different zones,
Biggish difference can all occur in the service of offer and communication quality etc., and the feature that this results in different zones that can obtain is different
It causes.So there was only a small amount of communication feature data in new region and feature probably changes, then how to reuse
The previous good model-aided current region model of regional training quickly and well is established, and is that this invention will solve the problems, such as.
Transfer learning technology and model reusability technology have certain similarity, but transfer learning lays particular emphasis on feature distribution
Alignment, and model reusability technology emphasizes the knowledge migration of model level.Previous Model Reuse Technique is confined to new old area more
Application scenarios with same characteristic features are not suitable for the changed open dynamic environment of feature, especially as trans-regional flow
Predict such application scenarios.Therefore, a kind of isomery Model Reuse skill is needed in trans-regional traffic forecasting application scenarios
Art.
Summary of the invention
Goal of the invention: previous Model Reuse Technique is confined to isomorphism feature for traffic forecasting more, and real ring
Border is often open dynamic, therefore the present invention provides a kind of trans-regional traffic forecasting side based on isomery Model Reuse
Method.Specifically, the metamessage of the characteristic and feature of collecting new and old area communication business first indicates, then uses model
Multiplexing technology and optimal transport technique complete isomery Model Reuse, and a new model is finally established in the numeric field data of new district, from
And it completes quickly to establish the good flux prediction model of performance using a small amount of communication data.
Technical solution: a kind of trans-regional traffic forecasting method based on isomery Model Reuse, including data collection step
Suddenly, model reusability step, new region model training step and new region traffic forecasting step；
The data collection step specifically:
Step 100, new, old area communication data is collected；
Step 101, collecting new, old area communications field feature metamessage indicates, if collected less than using new, old area
The public characteristic data reconstruction of communication service is new, the specific characteristic of old area, by reconstruction weights coefficient as the member of feature
Information indicates；
Model reusability step specifically:
Step 200, using communication feature data one linear regression model (LRM) M0 of training of old area, mode input is communication special
Sign, output is communication flows；
Step 201, indicate to obtain the transport distance matrix between new and old feature by new, old area feature metamessage, it is new special
Using being uniformly distributed between old feature between sign, algorithm is transported using optimal, learns one from new feature to old feature most
Excellent transfer matrix T；
Step 202, the prior model M1 for being converted to new region is carried out using model M 0 of the optimal transfer matrix T to old area,
The parameter of hypothesized model M0 is [W0a, W0b], and wherein W0a is the weight of old area characteristic feature, and W0b is that public feature is corresponding
Weight, using transition matrix obtain M1 parameter be [W0b, d2TW0a], d2 indicate new region characteristic feature quantity, TW0a
Indicate the matrix multiplication of T and W0a；
The step of new region model training specifically:
Step 300, communication feature data and prior model M1 in new region are collected；
Step 301, in a small amount of communication feature data of new region, the data volume that refers to working as on a small quantity here is not enough to training one
When new model, for example amount of training data is less than 100, using there is inclined regular terms about priori M1, and is aided with coefficient weights
The parameter of adjusting establishes new model M.
The step of new region traffic forecasting specifically:
Step 400, communication feature data to be predicted in new region are collected；
Step 401, it is predicted using established model M, input communication feature data, i.e., exportable coming few hours lead to
How much is letter flow amount.
The method of the present invention is suitable for change the field of (newly-increased feature, disappearance feature) in different area communication features
Scape, but there is Partial Feature to be consistent.
Communication data needs to collect communication feature data metamessage during collecting indicates；If collected less than can be with
The weight that new old area special characteristic reconstructs is indicated as metamessage using the feature of public characteristic part.
Optimal transport technique has been used to obtain the optimal transfer matrix of feature/model in model reusability step, so as to
Using by the isomery model conversion of old area as a prior model of new district domain model.
After obtaining model priori using transport technique, also need to establish using a small amount of communication feature data of new region new
Model, include polarization then with the weight coefficient of model adjust.
The utility model has the advantages that compared with prior art, the trans-regional communication stream provided by the present invention based on isomery Model Reuse
Prediction technique is measured, can solve the case where communication data feature changes, and take up less resources in implementation process, applicability
By force.
Detailed description of the invention
Fig. 1 is the new and old area communication collecting characterization data flow chart of the embodiment of the present invention；
Fig. 2 is the model reusability flow chart of the embodiment of the present invention；
Fig. 3 is the model training flow chart of the embodiment of the present invention；
Fig. 4 is the flow chart of new region traffic forecasting.
Specific embodiment
Combined with specific embodiments below, the present invention is furture elucidated, it should be understood that these embodiments are merely to illustrate the present invention
Rather than limit the scope of the invention, after the present invention has been read, those skilled in the art are to various equivalences of the invention
The modification of form falls within the application range as defined in the appended claims.
The following examples are illustrated using 4G signal communication volume forecasting as specific example, and new old area is respectively with cell
B and cell A indicates that the feature of communication data includes period, cell number, base station location, the bit error rate, Block Error Rate and bandwidth
Occupancy etc., the target of traffic forecasting are how much can to predict certain following several hour possible communication flows, here
Traffic forecasting model is exactly a simple linear regression model (LRM), and input is communication feature, and output is communication flows.Small
Area A has had a good historical models, can predict well communication flows, but cell B is built not yet
Corresponding model, and only a small amount of communication data are found, it is generally desirable to which one can just be constructed by collecting trimestral data
A new model.And the base station facility due to different community and geographical environment difference, having some features may be cell A/B special
Have, for example the distinctive service switch of cell A/B inside of base station opens situation, but two cells still there are some public spies
Sign, such as period, bandwidth usage etc..
As shown in Figure 1, new and old area communication collecting characterization data process: firstly, starting to collect data (step 10), data
Collecting includes two parts, and first part is communication feature data (step 11), and second part is the metamessage data (step of feature
15).Collection for first part's communication feature data, first collects the data of cell A, and whether judgement stores (step before
12), if there is storage, directly bringing use can (step 14a)；Otherwise it needs to collect (step 13a) again and obtains cell A's
Communication feature data (step 14a).Then the data (step 13b) for collecting cell B obtain the communication feature data (step of cell B
Rapid 14b).Second part data are the metamessage data (step 15) of communication feature, if can be collected into, are directly used
It can (step 18)；Otherwise the public characteristic data (such as the characteristic informations such as period, bandwidth usage) of characteristic are used
(step is indicated as metamessage to reconstruct the weight of the special characteristic (service release situation of certain base station etc.) of cell A, B
17).The metamessage of the characteristic and feature that finally export collection indicates (step 19).
As shown in Fig. 2, Model Reuse step stream are as follows: firstly, starting model reusability (step 20), be then ready for being collected into
Communication feature data and feature metamessage indicate (step 21).First judge that the model of cell A whether there is (step 22), such as
Fruit exists, and directly bringing use can (step 24)；Otherwise one mould of communication feature data re -training using cell A is needed
Type (step 23) then obtains the model M 0(step 24) of cell A.Then one is practised using the meta information table dendrography of cell A, B
Optimal matrix (the step 25,26) transported.Then the model M 0 of cell A is converted on cell B using optimal transfer matrix T
Model priori M1(step 27,28,29).
As shown in figure 3, model training steps flow chart are as follows: firstly, starting model training (step 30), be then ready for collecting
The communication feature data (a small amount of) and model priori M1(step 31) arrived, then by model priori M1 as there is inclined regular terms,
And the optimal way training pattern (step 32) that coefficient of utilization weight is adjusted, finally obtain in new region using low volume data and
Forefoot area on the obtained model (step 33,34) of isomery Model Reuse.
As shown in figure 4, the process of new region traffic forecasting are as follows: firstly, beginning preparing the communication flows of prediction cell B
(step 40) collects communication feature data (step 41) to be predicted, is then predicted (step using trained model M
42), finally export communication flows number (step 43,44).
Claims (6)
1. a kind of trans-regional traffic forecasting method based on isomery Model Reuse, which is characterized in that collect new and old area first
The characteristic of field communication business and the metamessage of feature indicate, are then completed using Model Reuse Technique and optimal transport technique
Isomery Model Reuse finally establishes a new model in the numeric field data of new district；Specifically include data collection step, model reusability
Step, new region model training step and new region model training step.
2. the trans-regional traffic forecasting method based on isomery Model Reuse as described in claim 1, which is characterized in that institute
State data collection step specifically:
Step 100, new, old area communication data is collected；
Step 101, collecting new, old area communications field feature metamessage indicates, if collected less than using new, old area
The public characteristic data reconstruction of communication service is new, the specific characteristic of old area, by reconstruction weights coefficient as the member of feature
Information indicates.
3. the trans-regional traffic forecasting method based on isomery Model Reuse as described in claim 1, which is characterized in that mould
Type reuses step specifically:
Step 200, using communication feature data one linear regression model (LRM) M0 of training of old area, mode input is communication special
Sign, output is communication flows；
Step 201, indicate to obtain the transport distance matrix between new and old feature by new, old area feature metamessage, it is new special
Using being uniformly distributed between old feature between sign, algorithm is transported using optimal, learns one from new feature to old feature most
Excellent transfer matrix T；
Step 202, the prior model M1 for being converted to new region is carried out using model M 0 of the optimal transfer matrix T to old area,
The parameter of hypothesized model M0 is [W0a, W0b], and wherein W0a is the weight of old area characteristic feature, and W0b is that public feature is corresponding
Weight, using transition matrix obtain M1 parameter be [W0b, d2TW0a], d2 indicate new region characteristic feature quantity, TW0a
Indicate the matrix multiplication of T and W0a.
4. the trans-regional traffic forecasting method based on isomery Model Reuse as described in claim 1, which is characterized in that new
The step of regional model training specifically:
Step 300, communication feature data and prior model M1 in new region are collected；
Step 301, in a small amount of communication feature data of new region, the data volume that refers to working as on a small quantity here is not enough to training one
When a new model, for example amount of training data is less than 100, using there is inclined regular terms about priori M1, and is aided with coefficient power
The parameter for resetting section, establishes new model M.
5. the trans-regional traffic forecasting method based on isomery Model Reuse as described in claim 1, which is characterized in that new
The step of area communication volume forecasting specifically:
Step 400, communication feature data to be predicted in new region are collected；
Step 401, it is predicted using established model M, i.e., how much is the communication flows of exportable coming few hours.
6. the trans-regional traffic forecasting method based on isomery Model Reuse as described in claim 1, which is characterized in that institute
State trans-regional traffic forecasting method be suitable for different area communication features can changed scene, but have part spy
Sign is consistent.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910803496.9A CN110505098A (en) | 2019-08-28 | 2019-08-28 | A kind of trans-regional traffic forecasting method based on isomery model reusability |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910803496.9A CN110505098A (en) | 2019-08-28 | 2019-08-28 | A kind of trans-regional traffic forecasting method based on isomery model reusability |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110505098A true CN110505098A (en) | 2019-11-26 |
Family
ID=68588666
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910803496.9A Pending CN110505098A (en) | 2019-08-28 | 2019-08-28 | A kind of trans-regional traffic forecasting method based on isomery model reusability |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110505098A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113822228A (en) * | 2021-10-27 | 2021-12-21 | 南京大学 | User expression recognition method and system based on continuous learning |
-
2019
- 2019-08-28 CN CN201910803496.9A patent/CN110505098A/en active Pending
Non-Patent Citations (1)
| Title |
|---|
| HAN-JIA YE ETC.: "Rectify Heterogeneous Models with Semantic Mapping", 《ICML》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113822228A (en) * | 2021-10-27 | 2021-12-21 | 南京大学 | User expression recognition method and system based on continuous learning |
| CN113822228B (en) * | 2021-10-27 | 2024-03-22 | 南京大学 | User expression recognition method and system based on continuous learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102377462B (en) | Method and system for interactively selecting auxiliary cell | |
| CN101541016A (en) | Method for predicting data and equipment | |
| CN108848520B (en) | Base station dormancy method based on flow prediction and base station state | |
| CN107992961A (en) | A kind of adaptive basin Medium-and Long-Term Runoff Forecasting model framework method | |
| CN107567079B (en) | A kind of base station energy-saving method of mobile Internet data driving | |
| CN105940722B (en) | A kind of AC, STA, AP and control method | |
| CN110381515B (en) | Method for realizing cell network flow resource index prediction based on combination and division mode | |
| CN101577918A (en) | Interference matrix establishment method based on mobile phone measurement report | |
| CN109815611B (en) | Basin boundary generating method based on digital basin | |
| CN105897811B (en) | A kind of method of data synchronization and device | |
| CN113382477B (en) | Method for modeling uplink interference between wireless network users | |
| CN110163426A (en) | A kind of multi-mode integrates precipitation forecast method and device | |
| US20130225156A1 (en) | Systems and Methods for Convergence and Forecasting for Mobile Broadband Networks | |
| CN113411817B (en) | Wireless system interference neural network prediction method based on wireless interference model | |
| EP3979563B1 (en) | Inter-domain data interaction method and apparatus | |
| CN109886445A (en) | A kind of tomorrow requirement prediction technique based on material requirements property quantification | |
| CN104155423A (en) | Variable weight ing combination landscape water quality early warning method | |
| CN104822162B (en) | Green base station shunt method and device in a kind of energy mix network | |
| CN114092769A (en) | Transformer substation multi-scene inspection analysis method based on federal learning | |
| CN114641015A (en) | Network evaluation method and device, electronic equipment and storage medium | |
| CN110505098A (en) | A kind of trans-regional traffic forecasting method based on isomery model reusability | |
| CN110824514A (en) | Fingerprint positioning method and device and computer readable storage medium | |
| CN110139315B (en) | Wireless network fault detection method based on self-learning | |
| CN112235821A (en) | Pilot frequency signal intensity prediction method based on AI model | |
| CN114118748B (en) | Service quality prediction method and device, electronic equipment and storage medium |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191126 |