CN107038252A - A kind of generation method of the route metric based on multi-modal data - Google Patents
A kind of generation method of the route metric based on multi-modal data Download PDFInfo
- Publication number
- CN107038252A CN107038252A CN201710307115.9A CN201710307115A CN107038252A CN 107038252 A CN107038252 A CN 107038252A CN 201710307115 A CN201710307115 A CN 201710307115A CN 107038252 A CN107038252 A CN 107038252A
- Authority
- CN
- China
- Prior art keywords
- data
- label
- label data
- modal
- mode
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/28—Databases characterised by their database models, e.g. relational or object models
- G06F16/284—Relational databases
- G06F16/285—Clustering or classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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"
- G06Q10/047—Optimisation of routes or paths, e.g. travelling salesman problem
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- Game Theory and Decision Science (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Development Economics (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention belongs to for radio network technique field, Internet route search technical field especially in vehicle self-organizing network, specifically propose a kind of generation method of the route metric based on multi-modal data, by analyze include represent environment, driver, vehicle, transport information multiple modalities data, route metric TDR is set up, the judgement precision to the reliable of intermediate node is improved:With a kind of semi-supervised multi-modal machine learning framework, there is label data study to a small amount of, coorinated training, coorinated training process is divided into positive training and label two iterative process of upgrading, constructs route metric.A kind of mechanism that route metric is generated based on machine learning algorithm is proposed, magnanimity route restriction point is considered, influence of the related data to route, the accurate route metric for judging route reliability of generation are analyzed with data mining angle.There is label data without label data and on a small quantity using semi-supervised multi-modal Architecture Analysis magnanimity, training modeling process is completed with minimum cost.
Description
Technical field
The present invention relates to a kind of semi-supervised multi-modal machine learning framework, by having label data study to a small amount of,
Coorinated training, the method for constructing route metric, the generation method of specially a kind of route metric based on multi-modal data.This point
Analysis includes representing environment, driver, vehicle, the multiple modalities data of transport information, sets up route metric TDR (Traffic-Data
Based Routing Metric), improve the judgement precision to the reliable of intermediate node (next-hop node).
Background technology
Since 21 century, the automobile sum of China rises to 2016 by more than 2,000 ten thousand and is close on 1.9 hundred million, and will
It is incident the problems such as be congestion in road, traffic accident with this with 10% speedup continuous rise.Car networking can be provided such as car
The service such as lane change prompting, intersection prompting, accident prompting, round-the-clock road condition analyzing, active automatic Pilot, is strengthening driving peace
On the basis of complete, mitigate congestion in road.In car networking system, vehicle self-organizing network (hereinafter referred to as VANET) is for it
The optimal selection that communication is supported is provided, future will realize vehicle-to-vehicle (V2V), car to infrastructure (V2I), car to all devices
Etc. (V2X) interconnecting.
In techniques known, researcher generally relies on experience to select the route restriction point in route metric to join
Number and respective weights coefficient, it is impossible to Comprehensive consideration numerous obligatory points influential on Route Selection and each obligatory point weighted value.Such as
Shown in Fig. 2, vehicle node is among complicated physical environment, therefore it is also many-sided to influence the factor of its route reliability
's.
In known technology, Xu Wenjun of Beijing University of Post & Telecommunication et al. exists《A Stable Routing Protocol for
Highway Mobility over Vehicular Ad-Hoc Networks》The routing algorithm proposed is moved using vehicle
And the information such as electronic map constructs highway mobility model, its route metric considered node relative distance, signal and received
Quality and vehicle acceleration etc..Wang new recruit of Shanghai Communications University et al.《History based Predictive Routing
in Multi-lane Delay Tolerable VANETs》Route metric used in carried routing algorithm introduces vehicle
The information such as the conventional driving path and vehicle class of node, to judge the reliability for routeing intermediate node.Wu et al. exists
《Routing in VANETs:A Fuzzy Constraint Q-Learning Approach》With《Flexible,
Portable,and Practicable Solution for Routing in VANETs:A Fuzzy Constraint Q-
Learning Approach》A kind of routing algorithm FPQ-AODV is proposed, used road is predicted and selected with fuzzy logic
By measuring.FPQ-AODV part scalabilities are strong, but the weight of each parameter needs root in the route metric produced by fuzzy logic
It is adjusted according to network environment, such design has certain limitation in actual applications.
In addition, in known technology, Jerbi et al. exists《An Improved Vehicular ad hoc Routing
Protocol for City Environments》VANET routing algorithms (GyTAR) are proposed, its route metric uses car first
The electronic map of node judges intersection, the magnitude of traffic flow and the from home information such as distance, significantly reduces packet loss
Rate.Li Changle et al. exists《LSGO:Link State Aware Geographic Opportunistic Routing
Protocol for VANETs》A kind of geographical link routing algorithm of chance type is proposed, the modified version ETX roads based on distance are used
Link reliability between vehicle node is assessed by measuring.ETX and its derivative route metric have only monitored the reliability of link,
Though handling capacity is added, due to not considering the influent factor of other routes in route metric, it is impossible to provide preferably transmission
Reliability.
The content of the invention
To solve problems of the prior art, the present invention includes representing environment, driver, vehicle, traffic by analysis
The multiple modalities data of information, set up route metric TDR (Traffic-Data based Routing Metric), raising pair
The judgement precision of the reliable of intermediate node (next-hop node), specifically provides a kind of route metric based on multi-modal data
Generation method, comprise the following steps,
Step one:Multi-modal data sample is collected, has label data and other are big comprising a small amount of in data sample herein
Amount is without label data;
Step 2:Based on semi-supervised Multimodal Learning framework to being labelled in step one without label data, now count
Then updated according to sample, there is label data to increase, no label data is reduced, obtain new having label data with new without label
Data;
Step 3:Using semi-supervised Multimodal Learning framework, to without label data and having label obtained by step 2
Data carry out coorinated training, construct route amount;
Further, it is described multi-modal including environment, driver, traffic data, vehicle.
Further, the described collection for having label data is labeled as L={ (E1,D1,V1,T1,y1),...,(Em,Dm,Vm,
Tm,ym), this set includes m and data sample is mixed, Ei、Di、Vi、TiRepresent respectively and be labeled as yi∈ { 1 ..., C } environment
Mode, driver's mode, vehicle mode and traffic data modal information;U={ (E are labeled as without label data collection1,D1,V1,T1,
y1),...,(En,Dn,Vn,Tn,yn), this set includes n and data sample is mixed, Ei、Di、Vi、TiRepresent respectively and be labeled as yi∈
Environment mode, driver's mode, vehicle mode and the traffic data modal information of { 1 ..., C }.
Further, in step 3, to it is described without label data and have label data learn coorinated training during,
Coorinated training process is divided into step A and two iterative process of step B:
Step A, positive training process, described positive training process includes, first by PCA algorithms to there is label data
Each modal data carry out dimension-reduction treatment, secondly classification model construction is carried out to each modal data after dimensionality reduction with SVM algorithm, formation
The provisional classifications device of each modal data, i.e. environment classifier, driver's grader, vehicle classification device, traffic data grader;
Step B, label escalation process, using convex clustering algorithm, converges to global minimum, finds the classification of optimum cluster
Number, extracts data sample with there is label data to carry out feature clustering, feature confidence is high (i.e. similar from being concentrated without label data
Degree is high) will be labeled with new feature label Z without label data sample, obtain individual features and belong to what is obtained after class, cluster
New feature belongs to similar original feature and belongs to class, and constitute new feature has label data collection This new feature there is into label data collection L after completetStep A positive training is carried out, if from without mark
The high data sample of confidence can not be selected in label data set U, then step A and step B iteration terminate.
At this moment whether there is label data is all tagged by above step B, as posting the new of new feature label Z
There is label data, then, have label data collection to new feature Progress is final to be respectively trained, i.e., have label data to carry out positive training new feature, first by PCA algorithms to new
Each modal data for having label data of feature carries out dimension-reduction treatment, secondly with SVM algorithm to each modal data after dimensionality reduction
Classification model construction is carried out, i.e., the provisional classifications device of each mode is carried out after last time refresh data, the final environment classifier of acquisition,
Driver's grader, vehicle classification device, traffic data grader.
Fusion point is built with the final environment classifier, driver's grader, vehicle classification device, traffic data grader
Class device, by the final environment classifier, driver's grader, vehicle classification device, each grader of traffic data grader it is pre-
Accuracy rate is surveyed to use at integrated classification device as the weight of ballot.Integrated classification device is ultimately constructed route metric, its
Formula is as follows:
That is f=A Σ ejEj+B∑dkDk+C∑vpVp+H∑tqTq
Wherein, Ej、Dk、Vp、TqIt is a feature in terms of environment, driver, vehicle, traffic four respectively, and ej、dk、vp、tq
It is this weighted value of four features in this mode respectively, these weighted values are by finally training the final classification device of each mode
Established during accuracy rate, i.e. ∑ ejEj、∑dkDk、∑vpVp、∑tqTqThe final classification device of respectively four mode.And A, B,
C, H use prediction accuracy when this four weighted values are by finally constituting integrated classification device, obtained each mode grader
Shared weight, accuracy rate is higher, and weight is bigger.
Further, for A, B, C, H, this four weighted values are established in the following manner, i.e., predicted using integrated classification device
When, each modal data is by first by a new training dataset, and this new training dataset is when being different from us to train
One beyond the training dataset used new independent to have label data collection.Built before directly this new data set is put into
In each mode final classification device of that vertical, the predictablity rate of final classification device can be tested out.Here, we use each mould
The prediction accuracy of state is voted as weight in integrated classification device, herein, alternatively, and we can also
Each mode weight is trained using Adaboost, to obtain optimal weighted value.
Further, it is the part of data sample described in step one without label using GPRS-T to network environment simulating
Data sample label so that it, which turns into, label data.
A kind of advantage of the generation method of route metric based on multi-modal data provided by the present invention is:By right
Environment, driver, vehicle, the multiple modalities data of transport information, set up route metric TDR (Traffic-Data based
Routing Metric), the judgement precision to the reliable of intermediate node (next-hop node) is improved, Optimization route is selected.It is existing
Have in route metric, the method that route restriction point is selected based on expertise has when judging route reliability the degree of accuracy not
Sufficient the problem of.By contrast, this project proposes a kind of mechanism that route metric is generated based on machine learning algorithm, considers magnanimity road
By obligatory point, influence of the related data to route is analyzed with data mining angle, generation can accurately judge route reliability
Route metric.Wherein, using the Unlabeled data and a small amount of flag data of semi-supervised multi-modal Architecture Analysis magnanimity, with most
Low cost completes training modeling process.
Brief description of the drawings
Fig. 1 sets up the basic process figure of route metric.
Fig. 2 is to influence each modal information of route reliability and carry out dimension-reduction treatment to data.
Fig. 3 is a kind of semi-supervised study framework, and by having label data study to a small amount of, coorinated training constructs road
By measuring, Fig. 3 is another form of expression to flow chart in Fig. 1.
The GPSR training versions GPSR-T that Fig. 4 is carried for the present invention
Embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, it is right below in conjunction with drawings and Examples
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
With reference to Fig. 1-4, the invention provides a kind of generation method of the route metric based on multi-modal data, including it is as follows
Step:
Step one:Collect multi-modal data sample, including environment, driver, vehicle, traffic data etc., each modal data
Source as shown in Fig. 2 201-207 be environmental information source, 209-215 be driver information source, 217-223 is information of vehicles
Source, 225-231 originates for traffic data information, for example:Weather, regional rank, building, other environment etc. are to belong to environment
Several features of modal data, they belong to environment modal data, and this belongs to class, other modal datas source detailed annotation reference rings
Border modal data carrys out Source Description.There is label data and other are a large amount of without label data comprising a small amount of in the data sample of collection.
Step 2:Based on semi-supervised Multimodal Learning framework, network environment is emulated using improved GPSR-T, is
Part is labelled without label data sample in step one, and now data sample is then updated in step one, there is number of tags
According to increasing, no label data is reduced, obtain it is new have label data with new without label data, their set is respectively to have mark
Sign data set L={ (E1,D1,V1,T1,y1),...,(Em,Dm,Vm,Tm,ym), this set includes m and data sample is mixed, Ei、
Di、Vi、TiRepresent respectively and be labeled as yi∈ { 1 ..., C } environment mode, driver's mode, vehicle mode and traffic data mode
Information, no label data collection U={ (E1,D1,V1,T1,y1),...,(En,Dn,Vn,Tn,yn), this set includes n to data sample
This mixing, Ei、Di、Vi、TiRepresent respectively and be labeled as yi∈ { 1 ..., C } environment mode, driver's mode, vehicle mode and traffic
Data modality information.The improved GPSR-T being previously mentioned is GPSR modified version, GPSR-T agreements be by GPSR agreements with
Distance carries out the greedy part of greedy forwarding for route metric and is modified, after modification when next-hop is selected, to every
Individual neighbor node sends a packet and forwarded, and neighbor node is done in the same fashion forwarding, until packet is reached
Destination node so track, this process is as shown in Figure 4.
Step 3:Using semi-supervised Multimodal Learning framework, to without label data and having label obtained by step 2
Data carry out coorinated training, construct route amount.To it is described without label data and have label data learn coorinated training during, point
For step A and two iterative process of step B:
Step A, is positive training process, including three steps:First by PCA algorithms to having number of tags in step 2
According to each mode carry out dimension-reduction treatment, reduce data coupling and redundancy, improve arithmetic speed, secondly with SVM algorithm to drop
Each modal data after dimension carries out classification model construction, forms the provisional classifications device of each modal data, i.e. environment classifier, Si Jifen
Class device, vehicle classification device, traffic data grader.
Step B, is label escalation process, using convex clustering algorithm, converges to global minimum, find the class of optimum cluster
Not Shuo, from without label data concentrate extract low volume data sample and have label data carry out feature clustering, feature confidence
High (feature confidence height refers to that similarity is high) will be labeled with new feature label Z without label data sample, obtain corresponding special
Levy the label data collection that the similar original category class of the implicit category obtained after category class, cluster constitutes new feature The process upgraded by label is herein referred to, these can obtain new without label data
Feature tag and the category class based on these new feature labels.For example, certain original has label data there was only 4 features, by this
Individual process, it is possible that 2 newly-increased features, and the new classification based on all 6 features.
At this moment whether there is label data is all tagged by above step B, as posting the new of new feature label Z
There is label data, then, have label data collection to new feature Progress is final to be respectively trained, i.e., have label data to carry out positive training new feature, first by PCA algorithms to new
Each modal data for having label data of feature carries out dimension-reduction treatment, secondly with SVM algorithm to each modal data after dimensionality reduction
Classification model construction is carried out, i.e., the provisional classifications device of each mode is carried out after last time refresh data, the final environment classifier of acquisition,
Driver's grader, vehicle classification device, traffic data grader.It may be said that every time from progress step B operations from without label data U
Select the high data sample of feature confidence and enter row label upgrading and label, it is complete after form interim point of each mode by step A again
Class device, each mode now divides the once refreshing that interim class device is each mode provisional classifications device to a upper process.By many
Secondary step B operation labels, and the provisional classifications device of step A has then carried out multiple refreshing, obtains final each mode grader.
Fusion point is built with the final environment classifier, driver's grader, vehicle classification device, traffic data grader
Class device, by the final environment classifier, driver's grader, vehicle classification device, each grader of traffic data grader it is pre-
Accuracy rate is surveyed to use at integrated classification device as the weight of ballot.Integrated classification device is ultimately constructed route metric, its
Formula is as follows:
That is f=A Σ ejEj+BΣdkDk+C∑vpVp+H∑tqTq
Wherein, Ej、Dk、Vp、TqIt is a feature in terms of environment, driver, vehicle, traffic four respectively, and ej、dk、vp、tq
It is this weighted value of four features in this mode respectively, these weighted values are by finally training the final classification device of each mode
Established during accuracy rate, i.e. ∑ ejEj、∑dkDk、∑vpVp、∑tqTqThe final classification device of respectively four mode.And A, B,
This four weighted values of C, H are the prediction accuracies for finally constituting each mode final classification device after integrated classification device, and accuracy rate is higher
Weight is bigger.For A, B, C, H, this four weighted values are established in the following manner, i.e., when being predicted using integrated classification device, each mould
State data are by first by a new training dataset, and this new training dataset is to be different from the instruction used when we train
Practicing one beyond data set new independent has label data collection.This new data set is directly put into that set up before
In each mode final classification device, the predictablity rate of final classification device can be tested out.Here, we use the prediction of each mode
The degree of accuracy is voted as weight in integrated classification device, herein, alternatively, and we can also use
Adaboost is trained to each mode weight, to obtain optimal weighted value.Being also referred to as without label data in the present invention program
For the data that do not label.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention
Any modifications, equivalent substitutions and improvements made within refreshing and principle etc., should be included in the scope of the protection.
Claims (6)
1. a kind of generation method of the route metric based on multi-modal data, it is characterised in that:Comprise the following steps,
Step one:Collect to include in multi-modal data sample, data sample herein and have label data and other a large amount of nothings on a small quantity
Label data;
Step 2:Based on semi-supervised Multimodal Learning framework to labelling without label data in step one, obtain new having
Label data is with new without label data;
Step 3:Using semi-supervised Multimodal Learning framework, to without label data and having label data obtained by step 2
Coorinated training is carried out, route amount is constructed.
2. a kind of generation method of the route metric based on multi-modal data as claimed in claim 1, it is characterised in that:It is described
It is multi-modal including environment, driver, traffic data, vehicle.
3. a kind of generation method of the route metric based on multi-modal data as claimed in claim 1, it is characterised in that:It is described
Have label data collection be labeled as L={ (E1,D1,V1,T1,y1),...,(Em,Dm,Vm,Tm,ym), this set includes m logarithms
Mixed according to sample, Ei、Di、Vi、TiRepresent respectively and be labeled as yi∈ { 1 ..., C } environment mode, driver's mode, vehicle mode and
Traffic data modal information;U={ (E are labeled as without label data collection1,D1,V1,T1,y1),...,(En,Dn,Vn,Tn,yn), this
Set includes n and data sample is mixed, Ei、Di、Vi、TiRepresent respectively and be labeled as yi∈ { 1 ..., C } environment mode, Si Jimo
State, vehicle mode and traffic data modal information.
4. a kind of generation method of the route metric based on multi-modal data as claimed in claim 1, it is characterised in that:In step
In rapid three, to it is described without label data and have label data learn coorinated training during, coorinated training process is divided into step A
With two iterative process of step B:
Step A, positive training process, described positive training process includes, first by PCA algorithms to there is each of label data
Individual modal data carries out dimension-reduction treatment, secondly carries out classification model construction to each modal data after dimensionality reduction with SVM algorithm, forms each mould
The provisional classifications device of state data, i.e. environment classifier, driver's grader, vehicle classification device, traffic data grader;
Step B, label escalation process, using convex clustering algorithm, converges to global minimum, finds the classification number of optimum cluster,
Data sample is extracted with there is label data to carry out feature clustering from being concentrated without label data, and feature confidence is high without label data sample
This will be labeled with new feature label Z, obtain individual features and belong to the new feature obtained after class, cluster and belong to similar original feature and belong to class,
Constitute new feature has label data collectionWill after complete
This new feature has a label data collection LtCarry out step A positive training;If can not select letter from without label data collection U
The high data sample of the heart, then step A and step B iteration terminate, have label data collection to new featureProgress is final to be respectively trained, and obtains final ring
Border grader, driver's grader, vehicle classification device, traffic data grader, are classified with the final environment classifier, driver
Device, vehicle classification device, traffic data grader build integrated classification device, by the final environment classifier, driver's grader,
Vehicle classification device, the predictablity rate of each grader of traffic data grader are used in integrated classification device as the weight of ballot
Place.Integrated classification device is ultimately constructed route metric, and its formula is as follows:
That is f=A Σ ejEj+B∑dkDk+CΣvpVp+H∑tqTq
Wherein, f is route metric, Ej、Dk、Vp、TqIt is a feature in terms of environment, driver, vehicle, traffic four respectively, and ej、
dk、vp、tqIt is this weighted value of four features in this mode respectively, these weighted values are by finally training the final of each mode
Established during the accuracy rate of grader, i.e. ∑ ejEj、∑dkDk、∑vpVp、ΣtqTqThe final classification of respectively four mode
Device, A, B, C, H use prediction accuracy when this four weighted values are by finally constituting integrated classification device, obtained each mode
Weight shared by grader, accuracy rate is higher, and weight is bigger.
5. a kind of generation method of the route metric based on multi-modal data as claimed in claim 1, it is characterised in that:For
This four weighted values of A, B, C, H are established in the following manner, i.e., when being predicted using integrated classification device, each modal data will make first
With a new training dataset, this new training dataset be different from the above-mentioned training dataset that we are used when training with
Outer one it is new it is independent have label data collection, this new data set is directly put into each mode final classification of above-mentioned foundation
In device, the predictablity rate of each mode final classification device is tested out, we use the prediction accuracy of each mode to exist as weight
Voted in integrated classification device;Herein, alternatively, we can also use Adaboost to each mode weight
It is trained, to obtain weighted value.
6. a kind of generation method of the route metric based on multi-modal data as claimed in claim 1, it is characterised in that:Using
GPRS-T, to network environment simulating, is that the part of data sample described in step one without label data sample label making it
As there is label data.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710307115.9A CN107038252B (en) | 2017-05-04 | 2017-05-04 | Route measurement generation method based on multi-mode data |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710307115.9A CN107038252B (en) | 2017-05-04 | 2017-05-04 | Route measurement generation method based on multi-mode data |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107038252A true CN107038252A (en) | 2017-08-11 |
CN107038252B CN107038252B (en) | 2020-11-06 |
Family
ID=59538318
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710307115.9A Active CN107038252B (en) | 2017-05-04 | 2017-05-04 | Route measurement generation method based on multi-mode data |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107038252B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110674854A (en) * | 2019-09-09 | 2020-01-10 | 东软集团股份有限公司 | Image classification model training method, image classification method, device and equipment |
WO2020093714A1 (en) * | 2018-11-09 | 2020-05-14 | 长沙理工大学 | Data processing method, apparatus and device, and readable storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110311038A1 (en) * | 2007-03-26 | 2011-12-22 | Avraham Faizakov | Semi-supervised training of destination map for call handling applications |
CN103778782A (en) * | 2014-02-20 | 2014-05-07 | 银江股份有限公司 | Traffic state partitioning method based on semi-supervised machine learning |
CN103886330A (en) * | 2014-03-27 | 2014-06-25 | 西安电子科技大学 | Classification method based on semi-supervised SVM ensemble learning |
CN104318242A (en) * | 2014-10-08 | 2015-01-28 | 中国人民解放军空军工程大学 | High-efficiency SVM active half-supervision learning algorithm |
CN105245608A (en) * | 2015-10-23 | 2016-01-13 | 同济大学 | Telematics network node screening and accessibility routing construction method based on self-encoding network |
-
2017
- 2017-05-04 CN CN201710307115.9A patent/CN107038252B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110311038A1 (en) * | 2007-03-26 | 2011-12-22 | Avraham Faizakov | Semi-supervised training of destination map for call handling applications |
CN103778782A (en) * | 2014-02-20 | 2014-05-07 | 银江股份有限公司 | Traffic state partitioning method based on semi-supervised machine learning |
CN103886330A (en) * | 2014-03-27 | 2014-06-25 | 西安电子科技大学 | Classification method based on semi-supervised SVM ensemble learning |
CN104318242A (en) * | 2014-10-08 | 2015-01-28 | 中国人民解放军空军工程大学 | High-efficiency SVM active half-supervision learning algorithm |
CN105245608A (en) * | 2015-10-23 | 2016-01-13 | 同济大学 | Telematics network node screening and accessibility routing construction method based on self-encoding network |
Non-Patent Citations (1)
Title |
---|
LIANG ZHAO, ETC: "A SVM based routing scheme in VANETs", 《2016 16TH INTERNATIONAL SYMPOSIUM ON COMMUNICATIONS AND INFORMATION TECHNOLOGIES (ISCIT)》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020093714A1 (en) * | 2018-11-09 | 2020-05-14 | 长沙理工大学 | Data processing method, apparatus and device, and readable storage medium |
CN110674854A (en) * | 2019-09-09 | 2020-01-10 | 东软集团股份有限公司 | Image classification model training method, image classification method, device and equipment |
CN110674854B (en) * | 2019-09-09 | 2022-05-17 | 东软集团股份有限公司 | Image classification model training method, image classification method, device and equipment |
Also Published As
Publication number | Publication date |
---|---|
CN107038252B (en) | 2020-11-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Jiang et al. | Graph neural network for traffic forecasting: A survey | |
EP3726439B1 (en) | Edge learning | |
Kong et al. | Mobility dataset generation for vehicular social networks based on floating car data | |
US11373115B2 (en) | Asynchronous parameter aggregation for machine learning | |
Alsrehin et al. | Intelligent transportation and control systems using data mining and machine learning techniques: A comprehensive study | |
US11263549B2 (en) | Method, apparatus, and system for in-vehicle data selection for feature detection model creation and maintenance | |
Huang et al. | Tourist hot spots prediction model based on optimized neural network algorithm | |
US11049390B2 (en) | Method, apparatus, and system for combining discontinuous road closures detected in a road network | |
US10943152B2 (en) | Method, apparatus, and system for detecting a physical divider on a road segment | |
Jiang et al. | A destination prediction network based on spatiotemporal data for bike‐sharing | |
Kong et al. | RMGen: A tri-layer vehicular trajectory data generation model exploring urban region division and mobility pattern | |
Sampathkumar et al. | Majority voting based hybrid ensemble classification approach for predicting parking availability in smart city based on IoT | |
Zhu et al. | Development of destination choice model with pairwise district-level constants using taxi GPS data | |
US20190065869A1 (en) | Predicting features on a road network with repeating geometry patterns | |
CN107038252A (en) | A kind of generation method of the route metric based on multi-modal data | |
Servizi et al. | Mining User Behaviour from Smartphone data: a literature review | |
Almukhalfi et al. | Traffic management approaches using machine learning and deep learning techniques: A survey | |
Cloudin et al. | Challenges on Mobility Models Suitable to Vanet. | |
Zhang et al. | Dynamical topology analysis of VANET based on complex networks theory | |
Yu et al. | A dynamic line generation and vehicle scheduling method for airport bus line based on multi-source big travel data | |
US20220207993A1 (en) | Method, apparatus, and system for verifying a lane closure using probe data | |
Radi et al. | Enhanced Implementation of Intelligent Transportation Systems (ITS) based on Machine Learning Approaches | |
Wang et al. | Influential spatial facility prediction over large scale cyber-physical vehicles in smart city | |
Lehmann et al. | Using crowd sensed data as input to congestion model | |
Phan et al. | A recent connected vehicle-IoT automotive application based on communication technology |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right |
Effective date of registration: 20220831 Address after: 3002-02, No. 91, Nujiang Street, Huanggu District, Shenyang City, Liaoning Province 110000 Patentee after: Liaoning shulianda Technology Co.,Ltd. Address before: 110136, Liaoning, Shenyang, Shenbei New Area moral South Avenue No. 37 Patentee before: SHENYANG AEROSPACE University |
|
TR01 | Transfer of patent right |