CN110290582B - Base station label track prediction method based on seq2seq frame - Google Patents

Base station label track prediction method based on seq2seq frame Download PDF

Info

Publication number
CN110290582B
CN110290582B CN201910427329.9A CN201910427329A CN110290582B CN 110290582 B CN110290582 B CN 110290582B CN 201910427329 A CN201910427329 A CN 201910427329A CN 110290582 B CN110290582 B CN 110290582B
Authority
CN
China
Prior art keywords
base station
label
label track
prediction model
seq2seq
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.)
Active
Application number
CN201910427329.9A
Other languages
Chinese (zh)
Other versions
CN110290582A (en
Inventor
吕明琪
曾大建
陈铁明
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University of Technology ZJUT
Original Assignee
Zhejiang University of Technology ZJUT
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Zhejiang University of Technology ZJUT filed Critical Zhejiang University of Technology ZJUT
Priority to CN201910427329.9A priority Critical patent/CN110290582B/en
Publication of CN110290582A publication Critical patent/CN110290582A/en
Application granted granted Critical
Publication of CN110290582B publication Critical patent/CN110290582B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/02—Services making use of location information
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/02—Services making use of location information
    • H04W4/029—Location-based management or tracking services
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W64/00—Locating users or terminals or network equipment for network management purposes, e.g. mobility management
    • H04W64/003—Locating users or terminals or network equipment for network management purposes, e.g. mobility management locating network equipment

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

A base station label track prediction method based on a seq2seq frame comprises the following steps: 1) continuously collecting labels of base stations currently connected with a user smart phone, and then dividing collected data to obtain a plurality of base station label tracks to form a base station label track data set; 2) analyzing the switching mode among the base stations, constructing a base station switching diagram, and then obtaining the vectorization representation of the base stations by adopting a diagram embedding algorithm; 3) constructing a base station label track prediction model by adopting a seq2seq frame, and providing a space loss function as a loss function of the seq2seq frame to finish the training of the base station label track prediction model; 4) inputting the label track of the current base station of the user into the trained prediction model, and predicting the most possible label track of the base station of the user in the future by the model. The method is only based on the base station label, does not need explicit position information, does not need to artificially define a formula and characteristics, and improves the accuracy of base station label track prediction.

Description

Base station label track prediction method based on seq2seq frame
Technical Field
The invention relates to mobile computing and machine learning technologies, in particular to a base station label track prediction method.
Background
The trajectory prediction means predicting the most probable future trajectory of a moving object (such as a vehicle and a pedestrian) based on the current trajectory, and the trajectory prediction is usually realized on the basis of mining and modeling the historical trajectory data of the moving object. Most of the current trajectory prediction methods require that the historical trajectory and the current trajectory data contain clear position information (i.e., longitude and latitude). For example, Y.Zheng reviewed the Mining and prediction of Trajectory data in "Trajectory data Mining: An overview" (ACM Transactions on Intelligent Systems and technology) and M.Lin and W.J.Hsu reviewed the Mining GPS data for mobility patterns: A surview "(personalized and Mobile Computing), all methods involved requiring that the Trajectory data contain explicit location information. However, the positioning method of the smart phone has great limitations, which hinders the application of these methods to the smart phone:
(1) smartphone accessible GPS equipment is fixed a position, and its limitation lies in: the GPS positioning energy consumption is too large, and the battery capacity of the smart phone can be quickly consumed by continuously using the GPS equipment.
(2) The smart phone can be positioned by a wireless signal source (such as a base station and WiFi), and has the limitations that: requiring frequent queries of the actual location of the wireless signal source over the network (since only a particular operator has this information), the continued use of wireless signal source location can greatly increase network traffic and render the method unusable offline.
In response to these problems, a small amount of mining and prediction work based on base station label trajectory data is currently performed. Laasonen, for example, proposes a base station label trajectory prediction method based on base station Clustering in "Clustering and prediction of mobile user routes from cellular data" (PKDD), G.
Figure BDA0002067930380000011
D.Katsaros、
Figure BDA0002067930380000012
Ulosoy et al in "A Data mining approach for location prediction in mobile environments" (Data)&Knowledge Engineering) proposes a base station label track prediction method based on base station switching association rule mining. However, these works consider a base station label trajectory directly as a string, ignoring spatial correlation between base stations, resulting in degraded trajectory prediction performance.
To address this problem, m.lv, l.chen, y.shen et al in "Measuring cell-id spatial mining for mobile phone route classification" (Knowledge-Based Systems) and m.lv, l.chen, t.chen et al in "Measuring induced virtual motion patterns from cell-id spatial data by exploiting mapping trajectory data" (Information Sciences) have mined base station label trajectory data in view of inter-base station spatial correlations. However, these works still suffer from the following drawbacks: (1) the spatial correlation calculation between the base stations depends too much on the field knowledge, and the formula or the characteristics need to be defined manually; (2) the trajectory model and the prediction model are fractured and cannot be jointly optimized.
Disclosure of Invention
In order to overcome the defects that the conventional base station label track prediction method is too dependent on domain knowledge and cannot be optimized in a combined mode, the invention provides a base station label track prediction method based on a seq2seq framework, which is only based on base station labels, does not need explicit position information, does not need manual definition of formulas and characteristics, and improves the accuracy of base station label track prediction.
The technical scheme adopted by the invention for solving the technical problems is as follows:
a base station label track prediction method based on a seq2seq frame comprises the following steps:
1) and (3) constructing a base station label track data set: continuously collecting labels of base stations currently connected with a user smart phone, and then dividing collected data to obtain a plurality of base station label tracks to form a base station label track data set;
2) base station embedding: analyzing the switching mode among the base stations and constructing a base station switching diagram, and on the basis, obtaining the vectorization representation of the base stations by adopting a diagram embedding algorithm;
3) training a base station label track prediction model: constructing a base station label track prediction model by adopting a seq2seq frame, and providing a space loss function as a loss function of the seq2seq frame so as to finish the training of the base station label track prediction model;
4) and (3) predicting the label track of the base station: inputting the label track of the current base station of the user into the trained prediction model, and predicting the most possible label track of the base station of the user in the future by the model.
Further, in step 2), the base station label track data set TS obtained in step 1) is given, and the step of embedding the base station label is as follows:
(2-1) base station switching diagram construction: first, for appearance in TSEach base station constructs a node; then, if the base station c1And base station c2Excessive direct switching occurs in TS, including c1Direct switch to c2And c2Direct switch to c1Then is c1Node and c2The nodes construct an edge, the weight w (c) of the edge1,c2) Is shown in formula (1), wherein, # h (c)1,c2) Is c1And c2Number of occurrences of inter-direct handover, # T (c)1,c2) Is c1And c2The number of simultaneous traces;
Figure BDA0002067930380000021
(2-2) base station vectorization based on graph embedding: processing the base station switching graph obtained in the step (2-1) by adopting an unsupervised graph embedding algorithm (such as Deepwalk and node2vec) to obtain a base station c corresponding to each graph nodeiK-dimensional dense vector x ofi。
Still further, in step 3), the step of training the base station label trajectory prediction model is as follows:
(3-1) constructing a prediction model based on the seq2seq framework: a base station label track prediction model is constructed by adopting a seq2seq framework, and the explanation is as follows:
an input layer: the input of the network is a base station label track, which is marked as TAFirst, T is putAAll base station labels are replaced by corresponding vectors to obtain a vector sequence vTAThen let vTAInputting a coding layer;
and (3) coding layer: coding layer processing vT with an LSTM networkAThe processing procedure is the current hidden state vector htFrom its previous hidden state vector ht-1And current base station index vector xtThe common generation is carried out, the output of the coding layer is a semantic vector C, and the semantic vector C is obtained by weighted average of all hidden state vectors through an attention mechanism;
a decoding layer: the decoding layer uses an LSTM network to process C, the processing process is the current hidden state vector stFrom the front of itA hidden state vector st-1And the previous base station index vector yt-1And (4) jointly generating. The output of the decoding layer is a base station label vector sequence vTB；
An output layer: output layer will vTBConverted into a base station label track TB；
(3-2) spatial loss function: first, an inter-base station loss coefficient is calculated from an inter-base station distance. The method for calculating the loss coefficient lw (u, v) between the base station u and the base station v is shown in formula (2), wherein d (u, v) is the shortest distance in the base station switching diagram obtained by the base station u and the base station v in the step (2-1), theta is a scale parameter and is a distance threshold, and CS is a set of all base stations in TS; then, considering the loss coefficient between the base stations, based on the cross entropy loss function, a spatial loss function is proposed as shown in formula (3) as the loss function of the seq2seq framework, where y1:t-1Represents the sequence y1,y2,...,yt-1，P(c|y1:t-1C) predicting the probability that the t base station is marked as C for the decoding layer;
Figure BDA0002067930380000031
Figure BDA0002067930380000032
(3-3) training a prediction model: firstly, dividing T at different positions of each base station label track T in TS, and obtaining two base station label tracks T by each divisionAAnd TBAs a training sample S ═ (T)A,TB) Wherein T isAAnd TBAnd respectively taking the training samples as the input and the output of the model, and recording the obtained training sample set as SS. And (4) inputting the SS into the prediction models obtained in the steps (3-1) and (3-2), and training the SS to obtain a final base station label track prediction model.
The invention has the following beneficial effects: 1. the method is based on the base station label only, and does not need clear position information, so that the limitation problem of a positioning mode of the smart phone is avoided; 2. considering the switching mode among the base stations, obtaining a vectorization representation capable of reflecting the spatial similarity among the base stations based on an unsupervised graph embedding algorithm without artificial feature design and base station spatial similarity marking; 3. and the accuracy of the base station label track prediction is improved by combining the base station vectorization representation and the space loss function.
Drawings
FIG. 1 is a flowchart of a base station label trajectory prediction method based on a seq2seq framework;
FIG. 2 is a base station switch graph according to an embodiment, wherein (a) is a base station label trace graph and (b) is a base station switch graph;
FIG. 3 is a prediction model based on the seq2seq framework.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
Referring to fig. 1 to 3, a method for predicting a base station label trajectory based on a seq2seq frame includes the following steps:
1) and (3) constructing a base station label track data set: continuously collecting labels of base stations currently connected with a user smart phone, and then dividing collected data to obtain a plurality of base station label tracks to form a base station label track data set;
2) base station embedding: analyzing the switching mode among the base stations and constructing a base station switching diagram, and on the basis, obtaining the vectorization representation of the base stations by adopting a diagram embedding algorithm;
3) training a base station label track prediction model: constructing a base station label track prediction model by adopting a seq2seq frame, and providing a space loss function as a loss function of the seq2seq frame so as to finish the training of the base station label track prediction model;
4) and (3) predicting the label track of the base station: inputting the label track of the current base station of the user into the trained prediction model, and predicting the most possible label track of the base station of the user in the future by the model.
Further, in step 2), the base station label track data set TS obtained in step 1) is given, and the step of embedding the base station label is as follows:
(2-1) base station switching diagram construction: first, for each occurrence in TSEach base station constructs a node; then, if the base station c1And base station c2Excessive direct switching occurs in TS, including c1Direct switch to c2And c2Direct switch to c1Then is c1Node and c2The nodes construct an edge, the weight w (c) of the edge1,c2) Is shown in formula (1), wherein, # h (c)1,c2) Is c1And c2Number of occurrences of inter-direct handover, # T (c)1,c2) Is c1And c2The number of simultaneous traces; one specific embodiment is shown in fig. 2.
Figure BDA0002067930380000041
(2-2) base station vectorization based on graph embedding: processing the base station switching graph obtained in the step (2-1) by adopting an unsupervised graph embedding algorithm (such as Deepwalk and node2vec) to obtain a base station c corresponding to each graph nodeiK-dimensional dense vector x ofi。
Still further, in step 3), the step of training the base station label trajectory prediction model is as follows:
(3-1) constructing a prediction model based on the seq2seq framework: a base station label track prediction model is constructed by adopting a seq2seq framework, the network structure of which is shown in FIG. 3 and is explained as follows:
an input layer: the input of the network is a base station label track, which is marked as TAFirst, T is putAAll base station labels are replaced by corresponding vectors to obtain a vector sequence vTAThen let vTAInputting a coding layer;
and (3) coding layer: coding layer processing vT with an LSTM networkAThe processing procedure is the current hidden state vector htFrom its previous hidden state vector ht-1And current base station index vector xtThe common generation is carried out, the output of the coding layer is a semantic vector C, and the semantic vector C is obtained by weighted average of all hidden state vectors through an attention mechanism;
a decoding layer: the decoding layer uses an LSTM networkC, processing the current hidden state vector stFrom its previous hidden state vector st-1And the previous base station index vector yt-1And (4) jointly generating. The output of the decoding layer is a base station label vector sequence vTB；
An output layer: output layer will vTBConverted into a base station label track TB；
(3-2) spatial loss function: first, an inter-base station loss coefficient is calculated from an inter-base station distance. The method for calculating the loss coefficient lw (u, v) between the base station u and the base station v is shown in formula (2), wherein d (u, v) is the shortest distance in the base station switching diagram obtained by the base station u and the base station v in the step (2-1), theta is a scale parameter and is a distance threshold, and CS is a set of all base stations in TS; then, considering the loss coefficient between the base stations, based on the cross entropy loss function, a spatial loss function is proposed as shown in formula (3) as the loss function of the seq2seq framework, where y1:t-1Represents the sequence y1,y2,...,yt-1，P(c|y1:t-1C) predicting the probability that the t base station is marked as C for the decoding layer;
Figure BDA0002067930380000051
Figure BDA0002067930380000052
(3-3) training a prediction model: firstly, dividing T at different positions of each base station label track T in TS, and obtaining two base station label tracks T by each divisionAAnd TBAs a training sample S ═ (T)A,TB) Wherein T isAAnd TBAnd respectively taking the training samples as the input and the output of the model, and recording the obtained training sample set as SS. And (4) inputting the SS into the prediction models obtained in the steps (3-1) and (3-2), and training the SS to obtain a final base station label track prediction model.

Claims (2)

1. A base station label track prediction method based on a seq2seq frame is characterized by comprising the following steps:
1) and (3) constructing a base station label track data set: continuously collecting labels of base stations currently connected with a user smart phone, and then dividing collected data to obtain a plurality of base station label tracks to form a base station label track data set;
2) base station embedding: analyzing the switching mode among the base stations and constructing a base station switching diagram, and on the basis, obtaining the vectorization representation of the base stations by adopting a diagram embedding algorithm;
3) training a base station label track prediction model: constructing a base station label track prediction model by adopting a seq2seq frame, and providing a space loss function as a loss function of the seq2seq frame so as to finish the training of the base station label track prediction model;
4) and (3) predicting the label track of the base station: inputting the label track of the current base station of the user into a trained prediction model, and predicting the most probable label track of the base station of the user in the future by the model;
in the step 2), the base station label track data set TS obtained in the step 1) is given, and the step of embedding the base station label is as follows:
(2-1) base station switching diagram construction: firstly, constructing a node for each base station appearing in TS; then, if the base station c1And base station c2Excessive direct switching occurs in TS, including c1Direct switch to c2And c2Direct switch to c1Then is c1Node and c2The nodes construct an edge, the weight w (c) of the edge1,c2) Is shown in formula (1), wherein, # h (c)1,c2) Is c1And c2Number of occurrences of inter-direct handover, # T (c)1,c2) Is c1And c2The number of simultaneous traces;
Figure FDA0002594210620000011
(2-2) base station vectorization based on graph embedding: processing the base station switching graph obtained in the step (2-1) by adopting an unsupervised graph embedding algorithmThen, the corresponding base station c of each graph node is representediK-dimensional dense vector x ofi。
2. The method for predicting the label locus of the base station based on the seq2seq framework as claimed in claim 1, wherein in the step 3), the step of training the label locus prediction model of the base station is as follows:
(3-1) constructing a prediction model based on the seq2seq framework: a base station label track prediction model is constructed by adopting a seq2seq framework, and the explanation is as follows:
an input layer: the input of the network is a base station label track, which is marked as TAFirst, T is putAAll base station labels are replaced by corresponding vectors to obtain a vector sequence vTAThen let vTAInputting a coding layer;
and (3) coding layer: coding layer processing vT with an LSTM networkAThe processing procedure is the current hidden state vector htFrom its previous hidden state vector ht-1And current base station index vector xtThe common generation is carried out, the output of the coding layer is a semantic vector C, and the semantic vector C is obtained by weighted average of all hidden state vectors through an attention mechanism;
a decoding layer: the decoding layer uses an LSTM network to process C, the processing process is the current hidden state vector stFrom its previous hidden state vector st-1And the previous base station index vector yt-1Generated together, the output of the decoding layer is a base station label vector sequence vTB；
An output layer: output layer will vTBConverted into a base station label track TB；
(3-2) spatial loss function: firstly, calculating a loss coefficient between base stations according to the distance between the base stations, wherein a calculation method of the loss coefficient lw (u, v) between the base stations u and v is shown in a formula (2), d (u, v) is the shortest distance in a base station switching diagram obtained by the base stations u and v in the step (2-1), theta is a scale parameter and is a distance threshold value, and CS is a set of all base stations in TS; then, considering the loss coefficient between the base stations, based on the cross entropy loss function, a space loss function is proposed as the box of seq2seq as shown in formula (3)Loss function of the shelf, wherein y1:t-1Represents the sequence y1,y2,...,yt-1，P(c|y1:t-1C) predicting the probability that the t base station is marked as C for the decoding layer;
Figure FDA0002594210620000021
Figure FDA0002594210620000022
(3-3) training a prediction model: firstly, dividing T at different positions of each base station label track T in TS, and obtaining two base station label tracks T by each divisionAAnd TBAs a training sample S ═ (T)A,TB) Wherein T isAAnd TBAnd (3) respectively serving as the input and the output of the model, recording the obtained training sample set as SS, inputting the SS into the prediction models obtained in the steps (3-1) and (3-2), and training the SS to obtain the final base station label track prediction model.
CN201910427329.9A 2019-05-22 2019-05-22 Base station label track prediction method based on seq2seq frame Active CN110290582B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910427329.9A CN110290582B (en) 2019-05-22 2019-05-22 Base station label track prediction method based on seq2seq frame

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910427329.9A CN110290582B (en) 2019-05-22 2019-05-22 Base station label track prediction method based on seq2seq frame

Publications (2)

Publication Number Publication Date
CN110290582A CN110290582A (en) 2019-09-27
CN110290582B true CN110290582B (en) 2020-10-30

Family

ID=68002254

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910427329.9A Active CN110290582B (en) 2019-05-22 2019-05-22 Base station label track prediction method based on seq2seq frame

Country Status (1)

Country Link
CN (1) CN110290582B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110766227A (en) * 2019-10-24 2020-02-07 沈阳民航东北凯亚有限公司 Fare prediction method and device
CN111553543B (en) * 2020-05-18 2022-11-18 润联软件***(深圳)有限公司 TPA-Seq2 Seq-based power load prediction method and related components
CN114065884A (en) * 2021-10-20 2022-02-18 国网浙江省电力有限公司杭州供电公司 Infrared electronic tag position prediction method combined with local attention mechanism

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105678246A (en) * 2015-12-31 2016-06-15 浙江工业大学 Motion mode excavation method based on base station label locus
CN108446374A (en) * 2018-03-16 2018-08-24 北京三快在线科技有限公司 User view prediction technique, device, electronic equipment, storage medium
CN109685109A (en) * 2018-11-26 2019-04-26 浙江工业大学 A kind of base station label track classification method based on twin neural network

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9911043B2 (en) * 2012-06-29 2018-03-06 Omni Ai, Inc. Anomalous object interaction detection and reporting
CN107333232B (en) * 2016-04-29 2020-02-21 华为技术有限公司 Terminal positioning method and network equipment
CN106534350B (en) * 2016-12-09 2019-08-02 中国联合网络通信集团有限公司 A kind of method and device for prediction of meeting
CN108513251B (en) * 2018-02-13 2020-08-04 北京天元创新科技有限公司 Positioning method and system based on MR data
CN109034448B (en) * 2018-06-14 2022-02-11 重庆邮电大学 Trajectory prediction method based on vehicle trajectory semantic analysis and deep belief network

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105678246A (en) * 2015-12-31 2016-06-15 浙江工业大学 Motion mode excavation method based on base station label locus
CN108446374A (en) * 2018-03-16 2018-08-24 北京三快在线科技有限公司 User view prediction technique, device, electronic equipment, storage medium
CN109685109A (en) * 2018-11-26 2019-04-26 浙江工业大学 A kind of base station label track classification method based on twin neural network

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Discovering individual movement patterns from cell-id trajectory data by exploiting handoff features;Mingqi Lv;《Information Sciences》;20180920;参见正文第3节 *
Predicting a Pedestrian Trajectory Using Seq2Seq for Mobile Robot Navigation;Natsuki Sakata;《 IECON 2018 - 44th Annual Conference of the IEEE Industrial Electronics Society》;20181023;参见正文第3节 *

Also Published As

Publication number Publication date
CN110290582A (en) 2019-09-27

Similar Documents

Publication Publication Date Title
Gao et al. Identifying Human Mobility via Trajectory Embeddings.
Yuan et al. A survey of traffic prediction: from spatio-temporal data to intelligent transportation
CN110290582B (en) Base station label track prediction method based on seq2seq frame
He et al. Graph attention spatial-temporal network with collaborative global-local learning for citywide mobile traffic prediction
Zhao et al. Where are you heading? dynamic trajectory prediction with expert goal examples
CN107679558B (en) A kind of user trajectory method for measuring similarity based on metric learning
Gong et al. Potential passenger flow prediction: A novel study for urban transportation development
CN108153867B (en) User trajectory prediction technique and device based on temporal regularity
CN109858390A (en) The Activity recognition method of human skeleton based on end-to-end space-time diagram learning neural network
US20160125307A1 (en) Air quality inference using multiple data sources
CN106383868A (en) Road network-based spatio-temporal trajectory clustering method
Chen et al. Predicting next locations with object clustering and trajectory clustering
Xu et al. Predicting destinations by a deep learning based approach
CN108256590B (en) A kind of similar traveler recognition methods based on compound first path
Huang et al. A bimodal Gaussian inhomogeneous Poisson algorithm for bike number prediction in a bike-sharing system
CN105912683B (en) A kind of chronologically-based path matching method
Yue et al. Detect: Deep trajectory clustering for mobility-behavior analysis
CN105091890A (en) Method and apparatus used for recommending candidate path
CN111444243A (en) User behavior prediction image method and system based on track information
CN106339769A (en) User travel forecasting method for mobile social network
CN109685109A (en) A kind of base station label track classification method based on twin neural network
CN113298314A (en) Rail transit passenger flow prediction method considering dynamic space-time correlation
Cui et al. A road-aware neural network for multi-step vehicle trajectory prediction
CN114298187A (en) Target detection algorithm integrating improved attention mechanism
Chu et al. Simulating human mobility with a trajectory generation framework based on diffusion model

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