CN109902880A - A kind of city stream of people's prediction technique generating confrontation network based on Seq2Seq - Google Patents
A kind of city stream of people's prediction technique generating confrontation network based on Seq2Seq Download PDFInfo
- Publication number
- CN109902880A CN109902880A CN201910195736.1A CN201910195736A CN109902880A CN 109902880 A CN109902880 A CN 109902880A CN 201910195736 A CN201910195736 A CN 201910195736A CN 109902880 A CN109902880 A CN 109902880A
- Authority
- CN
- China
- Prior art keywords
- people
- data
- network
- stream
- prediction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Traffic Control Systems (AREA)
Abstract
The invention discloses a kind of city stream of people's prediction techniques that confrontation network is generated based on Seq2Seq, comprising: the urban human flow data of different time is abstracted into " picture frame " first, flow of the people is indicated with thermodynamic chart；It is divided into training data and label by data are observed according to the time, converts image problem for this problem；The overall thought that confrontation network is generated using WGAN, generator generates the flow of the people in certain following a period of time using Seq2Seq method based on historical data, while the external factor such as weather are added；Arbiter uses the true and false data of Wasserstein distance discrimination；It combines to generate to fight to lose in training process and continues to optimize generator and arbiter with backpropagation.Finally, when arbiter can not differentiate true and false, following city flow of the people is predicted using optimised generator.Method proposed by the present invention has used generation confrontation network to carry out urban human volume forecasting for the first time, combines external environmental factor, solves the problems, such as that fuzzy prediction and algorithmic statement are slow.
Description
Technical field
The present invention provides a kind of city stream of people's prediction technique that confrontation network is generated based on Seq2Seq, is related to intelligent friendship
Logical field is mainly used for city stream of people prediction, has in Urban Traffic Planning, citizens' activities and in terms of reducing traffic risk important
Effect.
Background technique
Contradiction with the acceleration of Urbanization in China, between ever-increasing urban population and limited space resources
Increasingly sharpen, traffic jam issue is caused to become a great problem for hindering urban development.Since the sixties in last century, the world
Urban Traffic Planning and urban traffic control are studied by various countries, but with the continuous expansion of city size and traffic condition
It is increasingly sophisticated, it is no longer feasible to carry out effective traffic administration against both measures, intelligent transportation system (ITS,
Intelligent Transportation System) it comes into being.Intelligent transportation system combines advanced physical communication equipment
And intelligent computer technology, it is established that for the information prediction and management system of entire transportation network, it is comprehensively effective at present
Solve the problems, such as the optimal path of the traffic and transport field including traffic congestion.
City stream of people prediction is the important component of intelligent transportation system, there is important research application in many fields
Value carries out the interest that city stream of people prediction increasingly causes researcher using machine learning techniques, and classical is used for urban human
The method of stream prediction is all based on statistical method mostly, there is the methods of ARIMA, Kalman filter, these are based on statistical side
Method can only cannot all predict the stream of people in entire city for specific region, and such method cannot all embody number very well
According to it is spatiotemporal.With advances in technology, the collection of the promotion of hardware and mass data, neural network is due to its excellent performance
Performance and be widely used, with convolutional neural networks, circulation nerve net and its a series of variants network structure proposition, very
More researchers have been used in the stream of people's prediction of city, a series of new method occur, such as: ST-Resnet, STDN and
The methods of DeepTransport, these methods neural network based learn from a large amount of data to feature, and well
The spatiotemporal of data is utilized, and obtains superior performance.Above research is all existing in city stream of people prediction
The technology exploration of aspect and further optimization, but these above-mentioned having some limitations property of method.Firstly, major part
Preceding research all only focuses on stream of people's prediction of next period, not can be carried out multi-step prediction；Secondly existing method all exists
The problem of fuzzy prediction, is predicted using the method that average value obtains, and it is not very accurate for predicting the data come；Most
It is exactly exactly afterwards simply to be seen as external spy to the utilization of much external informations such as weather, vacation and road information
Sign, if it is possible to these information be used carefully, the prediction for the region stream of people will be substantially improved.
To sum up, existing city stream of people prediction model often has ignored the spatial coherence in stream of people's prediction between section,
There is defect in terms of regional prediction, while the problem of there is also fuzzy predictions.Therefore, it is accurate often to there is prediction in existing issue
Degree and the lower defect of efficiency.
Summary of the invention
Goal of the invention: the purpose of the present invention is in view of the drawbacks of the prior art, provide a kind of generate based on Seq2Seq to fight
City stream of people's prediction technique of network, for solving the several defect problems proposed in terms of background technique.Using of the invention public
The method opened can efficiently use the temporal correlation of transport data stream to realize stream of people's prediction of entire city scope, solve
The problem of fuzzy prediction, and can guarantee there is higher precision of prediction in varied situations.
A kind of technical solution: city stream of people's prediction technique generating confrontation network based on Seq2Seq, the specific steps are as follows:
Step 1: data prediction
1) city: being divided into the grid chart of a m*n according to longitude and latitude by region division, and each grid is known as sub-district
Domain, all subregions form a set R={ r1,1... rI, j..rm*n, wherein rI, jRepresent be located at grid chart in the i-th row,
The subregion of jth column；
2) stream of people flows into Inflow and outflow Outflow: definitionFor the stream of people track of moment t, then for moment t sub-district
Domain rI, jInflow, Outflow can such as be given a definition respectively:
Wherein Tr:It isThe track at place；
3) weather, time and road information external information: are combined into a surface tensor；
The flowing in and out of the above-mentioned stream of people is combined into city stream of people's historical data tensor required for our, we by some
The stream of people at moment regards the frame of picture as, regards whole stream of people's data as a video.
Step 2: training neural network
The city stream of people's historical data tensor training constructed using step 1 generates confrontation network.There are two parts for model:
Generator and arbiter.For the sake of convenient, we respectively represent generator and arbiter using G and D, and their input then divides
It is not historical data tensor matrix X1:nWith true stream of people's tensor matrix XN+1:n+k。G(X1:n) it is that G is given birth to by learning sample distribution
The sample of production.If the input of D is distributed from truthful data, its output valve is 1 (it is true for representing judgement sample)；Such as
The input of fruit D derives from G (X1:n), then its output valve is 0 (representing judgement sample as vacation).Neural network model is totally adopted
Confrontation network is generated with WGAN, if using XiTruthful data is represented,The false data generated is represented, gives N to true and false data
Amount is lost in conjunction with mean square deviation, then the objective function of generation confrontation network may finally be described as following form:
Wherein λ is a parameter of balance confrontation loss and mean square deviation loss；
We use Seq2Seq model as generator G, in the training process for generating confrontation network, D are trained to maximize
Its discrimination precision to data source；Meanwhile the discrimination precision of D is minimized by training G；By using RMSProp algorithm
D and G are optimized respectively with back-propagation algorithm, finally when algorithmic statement, obtain optimal solution.
Step 3: prediction result is generated
By stream of people's tensor matrix { X at preceding t momentt| t=1 ... n } it inputs in trained neural network model, it is raw
The prediction result of the city flow of the people of the region at k moment, i.e. city stream of people tensor matrix { X after the G that grows up to be a useful person is generatedt| t=n
+ 1 ... n+k }.
As a further preferred embodiment of the present invention, in above-mentioned steps two, the specific design side of arbiter D and generator G
Method is as follows:
Generator contains a Seq2Seq structure, and Seq2Seq structure includes a compressor Encoder and a solution
Depressor Decoder.The part Encoder contains double-layer structure, and one CNN layers of convolutional neural networks, one long short-term memory net
LSTM layers of network.Firstly, training data is input to a convolutional neural networks layer (CNN) to learn the online stream of people's number of urban road
According to space characteristics；Then, one long memory network in short-term will be used to capture the temporal correlation of traffic data sequence；Final warp
The training for crossing CNN layers He LSTM layers exports the feature vector extracted.The feature that the part Decoder extracts Encoder layers
Vector is as input, LSTM layers of memory network in short-term one long it comprises three-decker, one R-CNN layers and an outside
EC-gate layers of information, EC-gate layers of effect is to combine the external information comprising weather, time and road information, by three layers
Transformation, utilize Encoder layers extraction feature generate after k moment urban human flow data tensor.
Arbiter D contains double-layer structure, one CNN layers of convolutional neural networks, LSTM layers of memory network in short-term one long.
The G false data generated and truthful data are put into D, are first inputted to the CNN layers of potential spatial character of learning data, it is then defeated
Enter to LSTM layers of learning time characteristic, the true and false of data is judged using Wasserstein distance, by way of backpropagation
Generator and arbiter are continued to optimize, until algorithmic statement, i.e. D can not divide the data of input to be true or false
's.
The utility model has the advantages that the present invention is directed to city stream of people forecasting problem, proposes and have been based on Seq2Seq generation confrontation network
City stream of people's prediction technique.The invention adopts the above technical scheme compared with prior art, has following technical effect that
1) present invention generates confrontation network progress city stream of people prediction using based on Seq2Seq for the first time, by entire road network
Urban human flow data is modeled as tensor matrix, realizes and predicts the other stream of people of City-level；
2) autocoder model is put into confrontation to generate in network, stream of people's data is regarded as image, stream of people's data sequence
Regard video as, the space-time characteristic of data is effectively extracted using convolutional neural networks and long memory network in short-term, while also adding
Enter and external information is considered, so that model is more accurate；
3) it solves the problems, such as that convergence present in the prior art is slow and fuzzy prediction, realizes and predicted from single point in time
The promotion of more moment predictions.
Detailed description of the invention
Fig. 1 is method flow diagram；
Fig. 2 is the specific design structure of neural network model；
Fig. 3 is to generate confrontation network structure.
Specific embodiment
With reference to the accompanying drawing, technical solution of the present invention is described in further detail.
The city stream of people's prediction technique overall procedure for generating confrontation network based on Seq2Seq is as shown in Figure 1.Number after modeling
According to being input in a generation confrontation network model, the prediction to the city stream of people of following a period of time is generated.Existing research table
It is bright, such as these external informations of weather, time and road information etc. for traffic flow data prediction play a significant role, because
This input data not only includes for trained history stream of people's data, also includes external information data tensor.Specifically, this hair
It is bright to construct following several groups of data as input:
Xt: stream of people's data at n moment before predicted time point.Xt={ xt| t=1 ... n }
EC-gate: the external information tensor being made of weather, time and POI road information
City stream of people's prediction technique disclosed by the invention that confrontation network is generated based on Seq2Seq, detailed process is as follows:
Step 1: data prediction
1) city: being divided into the grid chart of a m*n according to longitude and latitude by region division, and each grid is known as sub-district
Domain, all subregions form a set R={ r1,1... rI, j..rm*n, wherein rI, jRepresent be located at grid chart in the i-th row,
The subregion of jth column；
2) stream of people flows into Inflow and outflow Outflow: definitionFor the stream of people track of moment t, then for moment t sub-district
Domain rI, jInflow, Outflow can such as be given a definition respectively:
Wherein Tr:It isThe track at place；
3) weather, time and road information external information: are combined into a surface tensor；
The above-mentioned stream of people flows into Inflow and outflow Outflow is combined into city stream of people's historical data tensor required for us
As the input of whole network, while the stream of people sometime is regarded as the frame of picture by we, and whole stream of people's data are seen
Work is a video.
Step 2: training neural network
The city stream of people's historical data tensor training constructed using step 1 generates confrontation network.There are two parts for model:
Generator and arbiter.For the sake of convenient, we respectively represent generator and arbiter using G and D, and their input then divides
It is not historical data tensor matrix X1:nWith true stream of people's tensor matrix XN+1:n+k。G(X1:n) it is that G is given birth to by learning sample distribution
The sample of production.If the input of D is distributed from truthful data, its output valve is 1 (it is true for representing judgement sample)；Such as
The input of fruit D derives from G (X1:n), then its output valve is 0 (representing judgement sample as vacation).Neural network model is totally adopted
Confrontation network is generated with WGAN, network structure is as shown in figure 3, if use XiTruthful data is represented,The false data generated is represented,
Given N loses true and false data tensor in conjunction with mean square deviation, then generate confrontation network objective function may finally be described as it is as follows
Form:
Wherein λ is a parameter of balance confrontation loss and mean square deviation loss.
Use Seq2Seq model as generator G, in the training process for generating confrontation network, training D is right to maximize its
The discrimination precision of data source；Meanwhile the discrimination precision of D is minimized by training G；By using RMSProp algorithm and instead
D and G are optimized respectively to propagation algorithm, finally when algorithmic statement, obtain optimal solution.
As shown in Fig. 2, the generator G generated in confrontation network contains a Seq2Seq structure, Seq2Seq structure packet
Containing a compressor Encoder and a decompression machine Decoder.The part Encoder contains double-layer structure, a convolutional Neural
It is NN layers of network C, LSTM layers of memory network in short-term one long.Firstly, training data is input to a convolutional neural networks layer
(CNN) learn the space characteristics of the online stream of people's data of urban road；Then, one long memory network in short-term will be used to capture friendship
The temporal correlation of logical data sequence；The training for eventually passing through CNN layers He LSTM layers exports the feature vector extracted.
The feature vector that the part Decoder extracts Encoder layers is one long to remember in short-term it comprises three-decker as input
Recall LSTM layers of network, one R-CNN layers and one EC-gate layers of external information, EC-gate layers of effect is combined comprising day
Gas, the external information of time and road information, by three layers of transformation, when using k after the Encoder layers of feature extracted generation
The urban human flow data tensor at quarter.Arbiter D contains double-layer structure, one CNN layers of convolutional neural networks, one long to remember in short-term
Recall LSTM layers of network.The G false data generated and truthful data are put into D, it is potentially empty to be first inputted to CNN layers of learning data
Between characteristic, be then input to LSTM layers of learning time characteristic, the true and false of data judged using Wasserstein distance, is passed through
The mode combination RMSProp gradient descent algorithm of backpropagation continues to optimize generator and arbiter, until algorithmic statement, i.e. D
It is true or false for can not dividing the data of input.
Step 3: prediction result is generated
By stream of people's tensor matrix { X at preceding t momentt| t=1 ... n } it inputs in trained neural network model, it is raw
The prediction result of the city flow of the people of the region at k moment, i.e. city stream of people tensor matrix { X after the G that grows up to be a useful person is generatedt| t=n
+ 1 ... n+k }.
Embodiments of the present invention are explained in detail above in conjunction with attached drawing, but the present invention is not limited to above-mentioned implementations
Mode within the knowledge of a person skilled in the art can also be without departing from the purpose of the present invention
It makes a variety of changes.
Claims (2)
1. a kind of city stream of people's prediction technique for being generated confrontation network based on Seq2Seq, main feature are included the following steps:
(1) data observed are pre-processed: city being divided into the grid of a m*n based on longitude and latitude, uses rI, jIt indicates to be located at the
The grid of i row, jth column, defines t moment rI, jThe inflow Inflow and outflow Outflow of the region stream of people:
WhereinIt isFlowing in and out for the stream of people is finally combined into a tensor by the track at place
Input as whole network；
(2) problem definition: stream of people's tensor information of the preceding n time interval of given area R, while between predicting the following k time
Every interior flow of the people；
(3) using the method for generating confrontation network, the overall thought for using WGAN, wherein Web vector graphic Seq2Seq model is generated,
Whole network is known as SeqST-GAN, objective function is as follows:
Wherein λ is a parameter of balance confrontation loss and mean square deviation loss；
(4) generator network model G, arbiter network model D are defined: using Seq2Seq model buildings generator G, uses convolution
The feature of neural network and the long compression of memory network in short-term image reuses long memory network in short-term, R-CNN and combines external
The feature of information decompression compression, arbiter D carry out true and false differentiation using Wasserstein distance；Random initializtion generator G and
The weight parameter of arbiter D network；
(5) conception of history measured data is updated to and generates false prediction data in G, sentenced with false data and the truthful data update of generation
Other device D, it is excellent with arbiter progress to generator using the RMSProp algorithm of minimum batch processing, and by way of backpropagation
Change, generator is enable to generate more life-like prediction data, arbiter can more accurately identify true and false data, reach pair
Anti- purpose.If arbiter D does not restrain, i.e., it can differentiate the true and false of prediction data, then repeat the above process and be trained；
(6) if arbiter D restrains, that is, illustrate that arbiter cannot determine the true and false of prediction data, illustrate the prediction knot of generator
Fruit extremely approaches with legitimate reading, then uses the generator network G of final optimization pass as final fallout predictor.
2. a kind of city stream of people's prediction technique for generating confrontation network based on Seq2Seq according to claim 1, utilizes volume
Product neural network CNN and long memory network LSTM in short-term learns the space-time characterisation of urban human flow data, which is characterized in that when us
When possessing sufficient stream of people's data, in conjunction with external informations such as weather, recreation and road informations, Seq2Seq mould is utilized
Type is predicted the flow of the people in some area in next time interval using the method that confrontation generates network, solves fuzzy prediction
The problem of, the accuracy rate of prediction is improved, is provided more for Urban Traffic Planning, Path selection and traffic risk profile etc.
Added with the auxiliary tool of power, more convenient, more accurate method is provided.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910195736.1A CN109902880A (en) | 2019-03-13 | 2019-03-13 | A kind of city stream of people's prediction technique generating confrontation network based on Seq2Seq |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910195736.1A CN109902880A (en) | 2019-03-13 | 2019-03-13 | A kind of city stream of people's prediction technique generating confrontation network based on Seq2Seq |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109902880A true CN109902880A (en) | 2019-06-18 |
Family
ID=66953237
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910195736.1A Pending CN109902880A (en) | 2019-03-13 | 2019-03-13 | A kind of city stream of people's prediction technique generating confrontation network based on Seq2Seq |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109902880A (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110414747A (en) * | 2019-08-08 | 2019-11-05 | 东北大学秦皇岛分校 | A kind of space-time shot and long term urban human method for predicting based on deep learning |
| CN110580548A (en) * | 2019-08-30 | 2019-12-17 | 天津大学 | Multi-step traffic speed prediction method based on class integration learning |
| CN110689184A (en) * | 2019-09-21 | 2020-01-14 | 广东毓秀科技有限公司 | Method for predicting rail traffic stream of people through deep learning |
| CN110781838A (en) * | 2019-10-28 | 2020-02-11 | 大连海事大学 | Multi-modal trajectory prediction method for pedestrian in complex scene |
| CN110913033A (en) * | 2019-11-19 | 2020-03-24 | 广东电力信息科技有限公司 | IDCIP address allocation method based on CNN convolutional neural network learning |
| CN110992354A (en) * | 2019-12-13 | 2020-04-10 | 华中科技大学 | Abnormal region detection method for countering self-encoder based on introduction of automatic memory mechanism |
| CN111343650A (en) * | 2020-02-14 | 2020-06-26 | 山东大学 | Urban scale wireless service flow prediction method based on cross-domain data and loss resistance |
| CN111626490A (en) * | 2020-05-20 | 2020-09-04 | 南京航空航天大学 | Multitask city space-time prediction method based on counterstudy |
| CN112070270A (en) * | 2020-08-05 | 2020-12-11 | 杭州未名信科科技有限公司 | Network model for time sequence prediction and use method |
| CN112766600A (en) * | 2021-01-29 | 2021-05-07 | 武汉大学 | Urban area crowd flow prediction method and system |
| CN112910690A (en) * | 2021-01-18 | 2021-06-04 | 武汉烽火技术服务有限公司 | Network traffic prediction method, device and equipment based on neural network model |
| CN113128772A (en) * | 2021-04-24 | 2021-07-16 | 中新国际联合研究院 | Crowd quantity prediction method and device based on sequence-to-sequence model |
| CN113408772A (en) * | 2020-03-16 | 2021-09-17 | 广东毓秀科技有限公司 | Subway tramcar pedestrian flow prediction method based on deep learning |
| CN114819253A (en) * | 2022-03-02 | 2022-07-29 | 湖北大学 | Urban crowd gathering hotspot area prediction method, system, medium and terminal |
| 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 |
| CN111753519B (en) * | 2020-06-29 | 2024-05-28 | 鼎富智能科技有限公司 | Model training and identifying method and device, electronic equipment and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107103758A (en) * | 2017-06-08 | 2017-08-29 | 厦门大学 | A kind of city area-traffic method for predicting based on deep learning |
| WO2018028255A1 (en) * | 2016-08-11 | 2018-02-15 | 深圳市未来媒体技术研究院 | Image saliency detection method based on adversarial network |
| CN109215349A (en) * | 2018-10-26 | 2019-01-15 | 同济大学 | Traffic flow forecasting method when long based on deep learning |
-
2019
- 2019-03-13 CN CN201910195736.1A patent/CN109902880A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018028255A1 (en) * | 2016-08-11 | 2018-02-15 | 深圳市未来媒体技术研究院 | Image saliency detection method based on adversarial network |
| CN107103758A (en) * | 2017-06-08 | 2017-08-29 | 厦门大学 | A kind of city area-traffic method for predicting based on deep learning |
| CN109215349A (en) * | 2018-10-26 | 2019-01-15 | 同济大学 | Traffic flow forecasting method when long based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| 段宗涛等: "基于深度CNN-LSTM-ResNet组合模型的出租车需求预测", 《交通运输系统工程与信息》 * |
| 秦瑶等: "改进RNN的城市交通拥堵预测模型研究", 《电子世界》 * |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110414747B (en) * | 2019-08-08 | 2022-02-01 | 东北大学秦皇岛分校 | Space-time long-short-term urban pedestrian flow prediction method based on deep learning |
| CN110414747A (en) * | 2019-08-08 | 2019-11-05 | 东北大学秦皇岛分校 | A kind of space-time shot and long term urban human method for predicting based on deep learning |
| CN110580548A (en) * | 2019-08-30 | 2019-12-17 | 天津大学 | Multi-step traffic speed prediction method based on class integration learning |
| CN110689184A (en) * | 2019-09-21 | 2020-01-14 | 广东毓秀科技有限公司 | Method for predicting rail traffic stream of people through deep learning |
| CN110781838A (en) * | 2019-10-28 | 2020-02-11 | 大连海事大学 | Multi-modal trajectory prediction method for pedestrian in complex scene |
| CN110781838B (en) * | 2019-10-28 | 2023-05-26 | 大连海事大学 | Multi-mode track prediction method for pedestrians in complex scene |
| CN110913033A (en) * | 2019-11-19 | 2020-03-24 | 广东电力信息科技有限公司 | IDCIP address allocation method based on CNN convolutional neural network learning |
| CN110992354B (en) * | 2019-12-13 | 2022-04-12 | 华中科技大学 | Abnormal region detection method for countering self-encoder based on introduction of automatic memory mechanism |
| CN110992354A (en) * | 2019-12-13 | 2020-04-10 | 华中科技大学 | Abnormal region detection method for countering self-encoder based on introduction of automatic memory mechanism |
| CN111343650B (en) * | 2020-02-14 | 2021-11-30 | 山东大学 | Urban scale wireless service flow prediction method based on cross-domain data and loss resistance |
| CN111343650A (en) * | 2020-02-14 | 2020-06-26 | 山东大学 | Urban scale wireless service flow prediction method based on cross-domain data and loss resistance |
| CN113408772A (en) * | 2020-03-16 | 2021-09-17 | 广东毓秀科技有限公司 | Subway tramcar pedestrian flow prediction method based on deep learning |
| CN111626490A (en) * | 2020-05-20 | 2020-09-04 | 南京航空航天大学 | Multitask city space-time prediction method based on counterstudy |
| CN111753519B (en) * | 2020-06-29 | 2024-05-28 | 鼎富智能科技有限公司 | Model training and identifying method and device, electronic equipment and storage medium |
| CN112070270A (en) * | 2020-08-05 | 2020-12-11 | 杭州未名信科科技有限公司 | Network model for time sequence prediction and use method |
| CN112070270B (en) * | 2020-08-05 | 2022-12-20 | 杭州未名信科科技有限公司 | Time sequence prediction network model and use method |
| 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 |
| CN112910690A (en) * | 2021-01-18 | 2021-06-04 | 武汉烽火技术服务有限公司 | Network traffic prediction method, device and equipment based on neural network model |
| CN112766600A (en) * | 2021-01-29 | 2021-05-07 | 武汉大学 | Urban area crowd flow prediction method and system |
| CN113128772A (en) * | 2021-04-24 | 2021-07-16 | 中新国际联合研究院 | Crowd quantity prediction method and device based on sequence-to-sequence model |
| CN114819253A (en) * | 2022-03-02 | 2022-07-29 | 湖北大学 | Urban crowd gathering hotspot area prediction method, system, medium and terminal |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109902880A (en) | A kind of city stream of people's prediction technique generating confrontation network based on Seq2Seq | |
| CN111612206B (en) | Neighborhood people stream prediction method and system based on space-time diagram convolution neural network | |
| CN110070713B (en) | Traffic flow prediction method based on bidirectional nested LSTM neural network | |
| Xu et al. | Satellite image prediction relying on GAN and LSTM neural networks | |
| CN109492822B (en) | Air pollutant concentration time-space domain correlation prediction method | |
| CN109410575B (en) | Road network state prediction method based on capsule network and nested long-time memory neural network | |
| CN109508360B (en) | Geographical multivariate stream data space-time autocorrelation analysis method based on cellular automaton | |
| CN109190795A (en) | A kind of interregional Travel Demand Forecasting method and device | |
| CN109887282A (en) | A kind of road network traffic flow prediction technique based on level timing diagram convolutional network | |
| CN111860951A (en) | Rail transit passenger flow prediction method based on dynamic hypergraph convolutional network | |
| Tang et al. | Medium-and long-term precipitation forecasting method based on data augmentation and machine learning algorithms | |
| CN110458048A (en) | Take population distribution Spatio-temporal Evolution and the cognition of town pattern feature into account | |
| CN109035779A (en) | Freeway traffic flow prediction technique based on DenseNet | |
| CN116187555A (en) | Traffic flow prediction model construction method and prediction method based on self-adaptive dynamic diagram | |
| CN110164129B (en) | Single-intersection multi-lane traffic flow prediction method based on GERNN | |
| CN115204478A (en) | Public traffic flow prediction method combining urban interest points and space-time causal relationship | |
| Deng et al. | The pulse of urban transport: Exploring the co-evolving pattern for spatio-temporal forecasting | |
| CN114565187B (en) | Traffic network data prediction method based on graph space-time self-coding network | |
| CN111242395B (en) | Method and device for constructing prediction model for OD (origin-destination) data | |
| CN107170234A (en) | A kind of method predicted based on SVM algorithm traffic in short-term | |
| CN116052427B (en) | Inter-city inter-regional mobility prediction method and device based on private car travel track data | |
| CN115376317A (en) | Traffic flow prediction method based on dynamic graph convolution and time sequence convolution network | |
| CN104850677A (en) | Deep learning cellular automaton simulation analysis method based on geographic information system | |
| CN112766240B (en) | Residual multi-graph convolution crowd distribution prediction method and system based on space-time relationship | |
| Cheng et al. | Analysis and forecasting of the day-to-day travel demand variations for large-scale transportation networks: a deep learning approach |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190618 |