CN111626490A - Multitask city space-time prediction method based on counterstudy - Google Patents

Multitask city space-time prediction method based on counterstudy Download PDF

Info

Publication number
CN111626490A
CN111626490A CN202010433080.5A CN202010433080A CN111626490A CN 111626490 A CN111626490 A CN 111626490A CN 202010433080 A CN202010433080 A CN 202010433080A CN 111626490 A CN111626490 A CN 111626490A
Authority
CN
China
Prior art keywords
time
private
tasks
public
features
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
Application number
CN202010433080.5A
Other languages
Chinese (zh)
Inventor
缪浩
王森章
杜金龙
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.)
Nanjing University of Aeronautics and Astronautics
Original Assignee
Nanjing University of Aeronautics and Astronautics
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 Nanjing University of Aeronautics and Astronautics filed Critical Nanjing University of Aeronautics and Astronautics
Priority to CN202010433080.5A priority Critical patent/CN111626490A/en
Publication of CN111626490A publication Critical patent/CN111626490A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/40—Business processes related to the transportation industry

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Economics (AREA)
  • Evolutionary Computation (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Strategic Management (AREA)
  • Data Mining & Analysis (AREA)
  • Human Resources & Organizations (AREA)
  • Marketing (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Game Theory and Decision Science (AREA)
  • Development Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Primary Health Care (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)

Abstract

The utility model discloses a multitask city space-time prediction method based on counterstudy, including: abstracting city space-time data at different times into image frames and dynamic images, and dividing observation data according to time; then combining ConvLSTM and GcnLSTM to form a heterogeneous space-time network HSTN, and extracting low-dimensional space-time implicit characteristics of each task; the general method is characterized in that the idea of multi-task learning is adopted, the common characteristics of multiple tasks are obtained by using counterlearning, the common characteristics and the private characteristics are separated by using the idea of matrix orthogonality, and finally the common characteristics and the private characteristics are combined and input into a time sequence queue, and the traffic in a period of time in the future is generated by using ConvLSTM in combination with an attention mechanism. The method provided by the invention is used for predicting the urban space-time task by using multi-task learning based on counterlearning for the first time, and combines external environmental factors, so that the accuracy of prediction is improved.

Description

Multitask city space-time prediction method based on counterstudy
Technical Field
The invention provides a multitask urban space-time prediction method based on antagonistic learning, relates to the field of intelligent transportation, is mainly used for urban people flow prediction, and has important effects in urban traffic planning, citizen traveling and traffic risk reduction.
Background
With the acceleration of the urbanization process in China, the contradiction between the growing urban population and the limited space resources is increasingly intensified, so that the problem of traffic jam becomes a great problem for hindering urban development. Since the sixties of the last century, urban traffic planning and urban traffic control have been studied in countries around the world, but with the continuous expansion of urban scales and the increasing complexity of traffic conditions, effective traffic management against these two measures is no longer feasible, and Intelligent Traffic Systems (ITS) have been developed. The intelligent traffic system combines advanced physical communication equipment and intelligent computer technology to establish an information prediction and management system aiming at the whole traffic network, and is the best way for comprehensively and effectively solving the problems in the field of traffic transportation including traffic jam at present.
Urban people flow prediction is an important component of an intelligent traffic system, has important research application value in many fields, and the urban people flow prediction by utilizing a machine learning technology increasingly arouses the interest of researchers. With the advancement of technology, hardware improvement and collection of large amount of data, neural networks are widely used due to their excellent performance, and with the network structure of convolutional neural networks, cyclic neural networks and a series of variants thereof, many researchers have used them in urban traffic prediction, and a series of new methods are emerging, such as: ST-Resnet, STDN, deep transport and the like, which learn characteristics from a large amount of data, well utilize the spatiotemporal characteristics of the data and obtain more excellent performance. The above studies are all the existing technical exploration and further optimization in urban pedestrian flow prediction, but the above methods have some limitations. First, most previous studies have ignored associations between multiple tasks; secondly, the existing method cannot well model time and space simultaneously; finally, how to extract the relationship between tasks has not been well studied.
In summary, the existing urban people flow prediction model often ignores the relevance among a plurality of tasks, and has a defect in performing time and space modeling on data simultaneously. Therefore, the existing problems often have the defects of low prediction accuracy and efficiency.
Disclosure of Invention
The purpose of the invention is as follows: the invention aims to provide a multitask city space-time prediction method based on counterstudy aiming at the defects of the prior art, which is used for solving the problems of the defects in the background art. By adopting the method disclosed by the invention, the relevance among a plurality of space-time tasks can be effectively utilized, the space-time relevance of the traffic data stream is effectively modeled, and the problem of prediction accuracy is solved.
The technical scheme is as follows: a city people flow prediction method for generating a confrontation network based on Seq2Seq comprises the following specific steps:
the method comprises the following steps: data pre-processing
1) Area division: dividing the city into a grid map of m x n according to the longitude and latitude, calling each grid as a sub-region,
all sub-regions form a set R ═ R1，1，...ri，j，..rm*nIn which r isi，jRepresenting the subregions positioned in the ith row and the jth column in the grid map;
2) inflow of people into the Inflow and Outflow of the Outflow: definition of
Figure BSA0000209256420000021
For a trajectory of people stream at time t, then sub-region r for time ti，jThe infiflow and the Outflow of (c) can be defined as follows:
Figure BSA0000209256420000022
Figure BSA0000209256420000023
wherein T isr：
Figure BSA0000209256420000024
Is that
Figure BSA0000209256420000025
The trajectory of (d);
3) external information: combining weather, time and road information into an external feature tensor;
4) start-stop OD matrix: at time t, the OD matrix between the two sub-regions may be defined as follows:
Figure BSA0000209256420000026
the inflow and outflow of people streams and start and stop OD matrixes are combined into a city people stream historical data tensor required by us, people streams at a certain moment are regarded as frames of pictures, and the whole people stream data is regarded as a video.
Step two: training neural networks
And training the multi-task learning deep network based on the counterstudy by using the urban people stream historical data tensor constructed in the step one. The model comprises three parts: a private signature encoder, a public signature encoder, and a private decoder. Generally, using the Seq2Seq model, each task shares a public feature encoder, while having a private feature encoder and private decoder, the three types of encoders consist of an underlying network HSTN. For convenience, we use Encoder for task mm_private、EncodercommonAnd Decoderm_privateRespectively representing a private signature encoder, a public signature encoder and a private decoder. Firstly, inputting a historical in-out flow matrix and an OD matrix into a private encoder for a task m, regarding each sub-area as a node, establishing a hierarchical Dynamic space-time diagram (Dynamic Spatio-temporal graph) according to the OD matrix, regarding the historical in-out flow matrix and the OD matrix as pictures, and extracting the private features of data by using a countermeasure and HSTN network; secondly, the data of all tasks are input into the common code through the same processing procedureThe device extracts the common characteristics of a plurality of tasks by utilizing the countermeasure idea and the HSTN network; then, separating the private characteristics from the public characteristics by using the idea of matrix orthogonality; and finally, adding the public characteristics and the private characteristics, inputting the public characteristics and the private characteristics into a private decoder of the task m, and generating data needing to be predicted by using a convolution long-time memory network (ConvLSTM) by using an attention mechanism thought. By optimizing each encoder and decoder separately by stochastic gradient descent and back propagation algorithms, an optimal solution is obtained when the algorithms converge.
Step three: generating a prediction result
The people flow tensor matrix and the od matrix (X) at the first t momentstI t 1, i.n is input into the trained neural network model, and a prediction result of the urban pedestrian flow of the area where k moments are located, namely an urban pedestrian flow tensor matrix { X) is generatedt|t＝n+1，...n+k}。
As a further preferable aspect of the present invention, in the second step, a specific design method of the encoder and the decoder is as follows:
the encoder is composed of a basic heterogeneous space-time network (HSTN), the input of the HSTN is divided into two parts, namely a tensor matrix (ST-Image) formed by an incoming and outgoing flow or an OD matrix, the other part is a space-time Graph (ST-Graph) constructed according to the OD matrix and the incoming and outgoing flow matrix, each sub-area is regarded as a node, if the two sub-areas go forward, an edge is established between the two nodes, then the original ST-Image is subjected to down sampling for multiple times, the sub-areas with different scales after the down sampling are regarded as nodes, and if the nodes go forward or have interaction with the previous layer, an edge is established between the two nodes. And extracting the implicit characteristics of the ST-Image by utilizing a multilayer ConvLSTM network, and extracting the implicit characteristics of the ST-Graph by utilizing a multilayer multi-scale GcnLSTM network. The encoder is divided into a private encoder and a public encoder, the public encoder is used for extracting public characteristics of a plurality of tasks, HSTN is used as a generator, then a multilayer full-connection network is used as a discriminator, and the discriminator cannot discriminate which task the generated characteristics come from by virtue of the ideas of gradient inversion and countermeasure; the private encoder is used for extracting the private features of each task, the HSTN is used as a generator, as the overlap possibly exists between the public features and the private features, the private features and the public features are separated by adopting a matrix orthogonal idea, and the private features extracted by utilizing the matrix orthogonal idea possibly have problems, a discriminator is added in each private encoder, the countermeasure idea is utilized to discriminate which task the private features come from so that the features generated by the private encoder belong to the current task, and finally the private features and the public features are combined and stored in a time sequence queue and input into a decoder through an attention mechanism.
The decoder is used for predicting a plurality of tasks, and generates a people stream data tensor of k times after the generation by multi-layer transformation by using ConvLSTM as a basic network and combining external data such as weather, time, road information and the like.
Has the advantages that: aiming at the problem of urban people stream prediction, the invention provides a multitask urban people stream prediction method for generating a countermeasure network based on Seq2 Seq. Compared with the prior art, the invention adopting the technical scheme has the following technical effects:
1) the invention firstly utilizes a multitask urban people stream prediction method for generating an antagonistic network based on Seq2Seq to model urban people stream data of the whole road network into a tensor matrix, and consideration on external data is added to realize urban-level multitask people stream prediction;
2) adding the countermeasure idea into space-time multi-task learning, extracting public features of a plurality of tasks by using the idea of gradient inversion, and extracting private features of the tasks by means of the idea of orthogonal countermeasure and matrix;
3) the heterogeneous spatiotemporal network HSTN is provided by combining the traditional image neural network and the current popular image neural network;
drawings
FIG. 1 is a method flow diagram;
FIG. 2 is a specific design structure of a neural network model;
FIG. 3 is a diagram of a heterogeneous spatiotemporal network architecture;
FIG. 4 is a spatiotemporal queue structure diagram.
Detailed Description
The technical scheme of the invention is further explained in detail with reference to the attached drawings.
The general flow of the multi-task city space-time prediction method based on the counterstudy is shown in figure 1. The modeled data is input into a multitask Seq2Seq model to generate a flow prediction for multiple tasks over a period of time in the future. Existing research shows that external information such as weather, time, and road information has an important role in data prediction of traffic flow because input data includes not only historical pedestrian flow data used for training but also external information data tensors. Specifically, the present invention is able to take as input several sets of data as follows:
Figure BSA0000209256420000031
and predicting the people flow data at n moments before the time point.
Figure BSA0000209256420000032
External: external information tensor composed of weather, time and POI road information
The invention discloses a multitask city space-time prediction method based on antagonistic learning, which comprises the following specific processes:
the method comprises the following steps: data pre-processing
1) Area division: dividing a city into a grid graph of m x n according to longitude and latitude, wherein each grid is called as a sub-region, and all the sub-regions form a set R ═ { R ═ R1，1，...ri，j，..rm*nIn which r isi，jRepresenting the subregions positioned in the ith row and the jth column in the grid map;
2) inflow of people into the Inflow and Outflow of the Outflow: definition of
Figure BSA0000209256420000041
For a trajectory of people stream at time t, then sub-region r for time ti，jThe infiflow and the Outflow of (c) can be defined as follows:
Figure BSA0000209256420000042
Figure BSA0000209256420000043
wherein T isr：
Figure BSA0000209256420000044
Is that
Figure BSA0000209256420000045
The trajectory of (d);
3) external information: combining weather, time and road information into an external feature tensor;
4) start-stop OD matrix: at time t, the OD matrix between the two sub-regions may be defined as follows:
Figure BSA0000209256420000046
the inflow and outflow of people streams and start and stop OD matrixes are combined into a city people stream historical data tensor required by us, people streams at a certain moment are regarded as frames of pictures, and the whole people stream data is regarded as a video.
Step two: training neural networks
And training the multi-task learning deep network based on the counterstudy by using the urban people stream historical data tensor constructed in the step one. The model comprises three parts: a private signature encoder, a public signature encoder, and a private decoder. Generally, using the Seq2Seq model, each task shares a public feature encoder, while having a private feature encoder and private decoder, the three types of encoders consist of an underlying network HSTN, as shown in fig. 3. For convenience, we use Encoder for task mm_private、EncodercommonAnd Decoderm_privateRespectively representing a private signature encoder, a public signature encoder and a private decoder. First, for task m, input history outInputting the inflow matrix and the OD matrix into a private encoder, regarding each sub-area as a node, establishing a hierarchical dynamic space-time diagram (dynamic spatial-Temporal Graph) according to the OD matrix, regarding the historical inflow and outflow matrix and the OD matrix as pictures, and extracting the private features of data by using a countermeasure and HSTN network; secondly, inputting data of all tasks into a common encoder through the same processing process, and extracting common characteristics of a plurality of tasks by utilizing a countermeasure idea and an HSTN network, wherein the countermeasure idea is as follows:
Figure BSA0000209256420000047
then, the private characteristics and the public characteristics are separated by using the idea of matrix orthogonality.
And finally, adding the public characteristics and the private characteristics, storing the sum into a predefined time sequence queue, inputting a private decoder of the task m, and generating data needing to be predicted by using a convolution long-time memory network (ConvLSTM) by using an attention mechanism idea. By optimizing each encoder and decoder separately by stochastic gradient descent and back propagation algorithms, an optimal solution is obtained when the algorithms converge.
As shown in fig. 2, assuming there are m tasks, there are m private encoders, 1 public encoder, and m private decoders in the entire network. The structure of the private encoder is composed of two parts, one is a heterogeneous space-time network and is composed of a convolution long-time space-time network and a graph convolution long-time space-time network. Firstly, training data is input into a multilayer convolution long-term memory network to learn the geographical space-time characteristics of the pedestrian data on the urban road network, and the formula of the convolution long-term memory network is as follows:
it＝σ(Wxi*Xt+Whi*Ht-1+WciοCt-1+bi)，
ft＝σ(Wxf*Xt+Whf*Ht-1+WcfοCt-1+bf)，
Ct＝ftοCt-1+itοtanh(Wxc*Xt+Whc*Ht-1+bc)，
Ot＝σ(Wxo*Xt+Who*Ht-1+WcoοCt-1+bo).
Ht＝otοtanh(Ct).
where, represents convolution, o represents the hadamard product, σ represents the activation function, it, ft, Ct, ot, Ht represent the input gate, forgetting gate, memory cell and hidden feature, respectively. Then, a space-time diagram is established according to training data of the data, as shown in fig. 3, the space-time diagram is input into a graph convolution long-time memory network to learn semantic space-time characteristics of the human stream data on the urban road network, and a graph convolution long-time memory network formula is as follows:
Figure BSA0000209256420000051
Figure BSA0000209256420000052
Figure BSA0000209256420000053
Figure BSA0000209256420000054
Ht＝otοtanh(Ct)，
wherein
Figure BSA0000209256420000055
Representing graph convolution. Finally, as shown in fig. 3, the geographical spatio-temporal features and the semantic spatio-temporal features are combined, and the combined features are input into respective decoders to obtain a prediction result, where a multi-head attention mechanism is used, and the formula is as follows:
Figure BSA0000209256420000056
headi＝Attention(QWi Q，KWi K，VWi V)
Figure BSA0000209256420000057
where Q, K, V represent the query, key, and value, respectively.
Step three: generating a prediction result
The people stream tensor matrix (X) of the first t moments of a plurality of taskstI, t is 1, i.e. n, is input into a trained neural network model, and a decoder generates a prediction result of urban pedestrian flow of an area where k moments are located by using a convolution long-time and short-time memory network, namely an urban pedestrian flow tensor matrix { X [ -X ]t|t＝n+1，...n+k}。
The embodiments of the present invention have been described in detail with reference to the drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the gist of the present invention.

Claims (2)

1. A multitask city space-time prediction method based on counterstudy is mainly characterized by comprising the following steps:
(1) pretreatment observed data: dividing the city into m x n squares based on longitude and latitude, and using ri，jDenotes a cell located in the ith row and the jth column, and defines a time ri，jInflow and Outflow of regional flows infilow and Outflow:
Figure FSA0000209256410000011
Figure FSA0000209256410000012
wherein
Figure FSA0000209256410000013
Is that
Figure FSA0000209256410000014
The inflow and outflow of people flow are combined into a tensor as the input of the whole network finally;
(2) problem definition: given m correlated spatio-temporal data mining tasks
Figure FSA0000209256410000015
(e.g., OD prediction, in-out flow prediction, etc.), each task has a data set { D }t＝{dt1，dt2，...dtnGiven external data E (such as weather, POI and the like), all tasks are jointly learned, and the flow rates of the tasks in k time intervals in the future are predicted simultaneously { D }1k，D2k，...Dtk}K＝{t+K}；
(3) Establishing m +1 encoders and m decoders for simultaneously predicting all tasks by combining an Attention mechanism and utilizing a Seq2Seq model, wherein the m encoders are used for extracting the private features of the m tasks, establishing a public encoder for extracting the public features of the m tasks by virtue of a countermeasure idea, separating the private features of each task from the public features according to the idea of matrix similarity, and establishing the m decoders for simultaneously predicting the m tasks by utilizing whether the generated private features are reasonable or not by utilizing countermeasure resolution and combining the public features and the private features;
(4) inputting historical observation data of m tasks into an encoder, wherein an encoder network consists of ConvLSTM and GCNLSTM, the ConvLSTM extracts the characteristics of adjacent regions, the GCNLSTM extracts the characteristics of semantic adjacent regions, the characteristics of the semantic adjacent regions are combined, a public encoder is regarded as a generator network model G by virtue of an antagonistic learning thought, a discriminator network model D is defined, the characteristics obtained by the public encoder are input into a discriminator D, the task from which the data comes is discriminated by using Wassertein distance, the common characteristics of multiple tasks are generated until the discriminator cannot discriminate, the common characteristics and the private characteristics of the tasks are separated by virtue of matrix similarity and antagonism, and finally the common characteristics and the private characteristics are combined and input into a time sequence queue;
(5) performing an attention operation on the features in the queue, and predicting the weighted and summed features as initial features of a decoder;
(6) a gradient random descent method is used, and the model is optimized in a back propagation mode, so that data generation is more accurate.
2. The multi-task city space-time prediction method based on antagonistic learning according to claim 1, which utilizes convolution long-and-short-term memory network ConvLSTM, graph convolution long-and-short-term memory network GcnLSTM and Attention mechanism to learn the space-time characteristics of city people stream data, and is characterized in that when we have various sufficient people stream data, the Seq2Seq model is utilized to predict the people stream flow of a certain area in the next time interval, the antagonistic learning and matrix similarity ideas are utilized to generate the public characteristics and the private characteristics of tasks, and the public characteristics and the private characteristics are separated, and the public characteristics and the private characteristics are combined to store the characteristic time sequence into a queue, so as to capture the long-term trend, improve the prediction accuracy, and provide a powerful auxiliary tool for the aspects of city traffic planning, path selection, traffic risk prediction and the like, provides a more convenient and accurate method.
CN202010433080.5A 2020-05-20 2020-05-20 Multitask city space-time prediction method based on counterstudy Pending CN111626490A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010433080.5A CN111626490A (en) 2020-05-20 2020-05-20 Multitask city space-time prediction method based on counterstudy

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010433080.5A CN111626490A (en) 2020-05-20 2020-05-20 Multitask city space-time prediction method based on counterstudy

Publications (1)

Publication Number Publication Date
CN111626490A true CN111626490A (en) 2020-09-04

Family

ID=72272570

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010433080.5A Pending CN111626490A (en) 2020-05-20 2020-05-20 Multitask city space-time prediction method based on counterstudy

Country Status (1)

Country Link
CN (1) CN111626490A (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112541852A (en) * 2020-12-24 2021-03-23 南方科技大学 Urban people flow monitoring method and device, electronic equipment and storage medium
CN112562339A (en) * 2020-12-09 2021-03-26 福州大学 Robust traffic flow prediction method based on multitask graph convolutional network
CN112560981A (en) * 2020-12-24 2021-03-26 北京百度网讯科技有限公司 Training method, apparatus, device, program and storage medium for generating countermeasure model
CN112712169A (en) * 2021-01-08 2021-04-27 中国科学院地理科学与资源研究所 Model building method and application of full residual depth network based on graph convolution
CN112801411A (en) * 2021-02-26 2021-05-14 南京工业大学 Network flow prediction method based on generation countermeasure network
EP4040353A4 (en) * 2020-12-21 2022-08-10 Beijing Baidu Netcom Science And Technology Co., Ltd. Method for establishing risk prediction model, regional risk prediction method and corresponding apparatus

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109493599A (en) * 2018-11-16 2019-03-19 南京航空航天大学 A kind of Short-time Traffic Flow Forecasting Methods based on production confrontation network
CN109902880A (en) * 2019-03-13 2019-06-18 南京航空航天大学 A kind of city stream of people's prediction technique generating confrontation network based on Seq2Seq
CN110148296A (en) * 2019-04-16 2019-08-20 南京航空航天大学 A kind of trans-city magnitude of traffic flow unified prediction based on depth migration study

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109493599A (en) * 2018-11-16 2019-03-19 南京航空航天大学 A kind of Short-time Traffic Flow Forecasting Methods based on production confrontation network
CN109902880A (en) * 2019-03-13 2019-06-18 南京航空航天大学 A kind of city stream of people's prediction technique generating confrontation network based on Seq2Seq
CN110148296A (en) * 2019-04-16 2019-08-20 南京航空航天大学 A kind of trans-city magnitude of traffic flow unified prediction based on depth migration study

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112562339A (en) * 2020-12-09 2021-03-26 福州大学 Robust traffic flow prediction method based on multitask graph convolutional network
CN112562339B (en) * 2020-12-09 2022-01-04 福州大学 Robust traffic flow prediction method based on multitask graph convolutional network
EP4040353A4 (en) * 2020-12-21 2022-08-10 Beijing Baidu Netcom Science And Technology Co., Ltd. Method for establishing risk prediction model, regional risk prediction method and corresponding apparatus
JP2023510665A (en) * 2020-12-21 2023-03-15 ベイジン バイドゥ ネットコム サイエンス テクノロジー カンパニー リミテッド Risk prediction model establishment method, area risk prediction method, and response device
CN112541852A (en) * 2020-12-24 2021-03-23 南方科技大学 Urban people flow monitoring method and device, electronic equipment and storage medium
CN112560981A (en) * 2020-12-24 2021-03-26 北京百度网讯科技有限公司 Training method, apparatus, device, program and storage medium for generating countermeasure model
WO2022134649A1 (en) * 2020-12-24 2022-06-30 南方科技大学 Urban people flow monitoring method and apparatus, electronic device, and storage medium
CN112560981B (en) * 2020-12-24 2023-07-25 北京百度网讯科技有限公司 Training method, device, apparatus, program, and storage medium for generating countermeasure model
CN112541852B (en) * 2020-12-24 2024-04-12 南方科技大学 Urban people stream monitoring method and device, electronic equipment and storage medium
CN112712169A (en) * 2021-01-08 2021-04-27 中国科学院地理科学与资源研究所 Model building method and application of full residual depth network based on graph convolution
CN112801411A (en) * 2021-02-26 2021-05-14 南京工业大学 Network flow prediction method based on generation countermeasure network
CN112801411B (en) * 2021-02-26 2024-02-09 南京工业大学 Network traffic prediction method based on generation of countermeasure network

Similar Documents

Publication Publication Date Title
CN111626490A (en) Multitask city space-time prediction method based on counterstudy
Fang et al. Constgat: Contextual spatial-temporal graph attention network for travel time estimation at baidu maps
US11270579B2 (en) Transportation network speed foreeasting method using deep capsule networks with nested LSTM models
Liu et al. Contextualized spatial–temporal network for taxi origin-destination demand prediction
Liu et al. Urban traffic prediction from mobility data using deep learning
AU2020101023A4 (en) A Novel Deep Learning Approach for Distributed Traffic Flow Forecasting
Wang et al. Hierarchical traffic flow prediction based on spatial-temporal graph convolutional network
Zhang et al. Taxi demand prediction using parallel multi-task learning model
Niu et al. Real-time taxi-passenger prediction with L-CNN
Pan et al. Matrix factorization for spatio-temporal neural networks with applications to urban flow prediction
Zhang et al. Multistep speed prediction on traffic networks: A graph convolutional sequence-to-sequence learning approach with attention mechanism
Miao et al. MBA-STNet: Bayes-enhanced discriminative multi-task learning for flow prediction
Jin et al. Adaptive dual-view wavenet for urban spatial–temporal event prediction
Dai et al. Spatio-temporal deep learning framework for traffic speed forecasting in IoT
Gu et al. Exploiting interpretable patterns for flow prediction in dockless bike sharing systems
Jin et al. Deep multi-view spatiotemporal virtual graph neural network for significant citywide ride-hailing demand prediction
Dai et al. Attention based simplified deep residual network for citywide crowd flows prediction
Zhang et al. Dynamic auto-structuring graph neural network: A joint learning framework for origin-destination demand prediction
Wang et al. Traffic flow prediction based on spatiotemporal potential energy fields
CN104778355B (en) The abnormal track-detecting method of traffic system is distributed based on wide area
Lu et al. Bi-temporal attention transformer for building change detection and building damage assessment
Wu et al. Multi-task spatial-temporal graph attention network for taxi demand prediction
Fan et al. Global contextual attention for pure regression object detection
Wang et al. St-expertnet: A deep expert framework for traffic prediction
CN116665448A (en) Traffic speed real-time prediction method and system based on graph convolution network

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
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20200904

WD01 Invention patent application deemed withdrawn after publication