CN109978238B - Destination prediction method based on deep echo state network - Google Patents

Destination prediction method based on deep echo state network Download PDF

Info

Publication number
CN109978238B
CN109978238B CN201910162617.6A CN201910162617A CN109978238B CN 109978238 B CN109978238 B CN 109978238B CN 201910162617 A CN201910162617 A CN 201910162617A CN 109978238 B CN109978238 B CN 109978238B
Authority
CN
China
Prior art keywords
training
destination
echo state
state network
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.)
Expired - Fee Related
Application number
CN201910162617.6A
Other languages
Chinese (zh)
Other versions
CN109978238A (en
Inventor
邵杰
宋作华
申恒涛
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.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
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 University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN201910162617.6A priority Critical patent/CN109978238B/en
Publication of CN109978238A publication Critical patent/CN109978238A/en
Application granted granted Critical
Publication of CN109978238B publication Critical patent/CN109978238B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • 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)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Health & Medical Sciences (AREA)
  • Economics (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • General Business, Economics & Management (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Marketing (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Tourism & Hospitality (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Development Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Operations Research (AREA)
  • Game Theory and Decision Science (AREA)
  • Quality & Reliability (AREA)
  • Primary Health Care (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a destination prediction method based on a deep echo state network, and belongs to the technical field of track destination prediction. In order to improve the prediction accuracy of a destination, the invention provides a new deep echo state network variant, namely a dual-input deep echo state network, and the variant overcomes the defect of high time consumption of an original deep echo model and the existing variants thereof under large-scale training data and achieves the best known prediction performance; in addition, when the prediction model for destination prediction is trained, only the historical track data of the target is used for training, no other additional information is utilized, the application condition is wide, and compared with the traditional neural network model, the model has stronger time sequence data processing capacity and is easy to train.

Description

Destination prediction method based on deep echo state network
Technical Field
The invention belongs to the technical field of track destination prediction, and particularly relates to a method for predicting a final destination based on an initial segment of a track.
Background
In the past decade, with the popularity of mobile devices with positioning capabilities, a large amount of trajectory data has been generated. Taxi destination prediction is to predict a final destination based on the initial track segment. The work has important significance for research and development in other fields, can be used for improving commercial operations such as taxi allocation efficiency and regional advertisement putting, can also be used for city management, and can predict large-scale meeting events in advance to strengthen police guard and the like.
Some of the earliest studies on destination prediction problems have resorted to additional information such as driver personal identity, behavioral habits, and social networking, to name a few. Obviously, the additional information can improve the accuracy of prediction, but the additional information is difficult to obtain in most cases, and the trained model has no strong generalization capability.
Most existing solutions are based on various markov chain models. However, such models are markov (the conditional probability distribution of the future states of the process depends only on the current state, and not on the historical state), and the implicit assumption is that the vehicle is traveling in a memory-free random walk, which makes the historical trajectory ineffective in such models.
In general, Neural Network-based solutions are more efficient than the above methods, in which a Recurrent Neural Network (RNN) provides effective time-processing capability for destination prediction, but the conventional RNN model is difficult to train due to gradient disappearance and explosion. Reservoir Computation (RC), represented by Echo State Networks (ESNs), has proven to be the most advanced method of improving the training effect of conventional RNNs. ESNs have been applied to the problem of destination prediction, performing well on small-scale data, but have limitations on large-scale data.
Disclosure of Invention
The invention aims to: in order to overcome the defects of the existing destination prediction method and give play to the prediction effectiveness of an Echo State Network (ESN) on large-scale data, the invention discloses a new destination prediction method based on a deep Echo State Network (deep ESN), and the invention improves the traditional Network structure, reduces the training complexity of the traditional deep ESN and improves the prediction accuracy.
The invention relates to a destination prediction method based on a deep echo state network, which comprises the following steps:
setting a prediction model for a target object predicted by a destination, wherein the network structure of the prediction model is a double-input deep echo state network;
the dual input deep echo state network comprises NLA layer; wherein the first layer comprises two neurons
Figure BDA0001985177650000021
Reservoir of (2) to (N)LThe layers respectively comprise a neuron with the number of NRThe water reservoir; combining the outputs of the two reservoirs of the first layer as the output of the first layer of the dual-input deep echo state network, and taking the output of each layer as the input of the next layer; linearly combining the outputs of all layers to obtain an output result, namely a prediction result, of the dual-input deep echo state network;
taking historical moving track data of a target object in a specified area as a training sample set, and carrying out track data preprocessing on the training sample set: taking the first point of each track in the training sample set as a starting point and the last point as a destination;
slicing each track, namely, reserving a section of track segment starting from the starting point as a training segment for each track; and storing destination information of each training segment;
extracting prefixes and suffixes of each training segment based on a training sample set subjected to trajectory data preprocessing according to a preset ratio (an empirical value can be generally set to be 20% -30%), respectively inputting the extracted prefixes and the suffix of the suffixes into two water reservoirs of a first layer of a dual-input deep echo state network one by one (the prefixes and the suffix locus points of the suffixes are respectively input into one water reservoir of the first layer, and prediction is input in the same mode as training), and performing deep echo state network training on a prediction model to obtain a trained prediction model;
and extracting prefixes and suffixes of the current travel track segment containing the starting point of the target object in the designated area according to a preset proportion, respectively inputting the extracted prefixes and the track points of the suffixes into two reservoirs of a first layer of a trained prediction model one by one, and obtaining a current destination prediction result of the target object based on the output of the prediction model.
In summary, due to the adoption of the technical scheme, the invention has the beneficial effects that:
in order to improve the prediction accuracy of a destination, the invention provides a new variant of a deep echo state network (deep ESN), namely a dual-input deep echo state network (deep ESN-DI), and the variant overcomes the defect of high time consumption of an original deep ESN model and the existing variants thereof under large-scale training data and achieves the best known existing prediction performance; in addition, when the prediction model (the prediction model based on depesn-DI) for destination prediction is trained, the invention only uses the historical trajectory data of the target for training, does not utilize any other additional information, has wide application conditions, has stronger time sequence data processing capability compared with the traditional neural network model, and is easy to train.
Drawings
FIG. 1 is a prior art underlying deep echo state network model;
FIG. 2 is two variations of a prior art deep echo state network model, wherein FIG. 2-a is a full input echo state network;
FIG. 2-b is a diagram of an echo state network group;
fig. 3 is a dual-input deep echo state network model of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the following embodiments and accompanying drawings.
The deep echo state network (deepsesn) consists of several Recurrent Neural Networks (RNNs) called "reservoirs (reservoirs)". Each reservoir contains a number of randomly connected neurons and remains unchanged during training. The reservoirs are stacked in layers, the first layer is directly input by external information, the output of each layer is used as the input of the next layer, and the transmission of state information between the successive layers has no time delay. The output is generated by a linear combination of the signals generated by the neurons from the water reservoir.
As shown in FIG. 1, a basic deepsesn model contains NLA plurality of reservoirs, each reservoir containing NRA neuron, and NUAn input unit and NYAnd an output unit. A weight matrix W is arranged between the input unit and the first reservoirinIs connected with W(i)An input weight matrix representing the reservoir of the ith floor, an input weight matrix WinThat is W(1)Neuron in each reservoir is defined by a weight matrix
Figure BDA0001985177650000031
Connected, between reservoir and output layer by weight matrix WoutConnected, there is no output feedback signal in this model.
The state update function for depesn is as follows. For the first layer, the update formula is:
Figure BDA0001985177650000032
Figure BDA0001985177650000033
when i > 1, the state update formula of the subsequent layer is as follows:
Figure BDA0001985177650000034
Figure BDA0001985177650000035
wherein the content of the first and second substances,
Figure BDA0001985177650000036
is the status of the reservoir at the ith floor,
Figure BDA0001985177650000037
is an upgrade of the state of it,
Figure BDA0001985177650000038
the representation of the real number field is performed,
Figure BDA0001985177650000039
representing the external inputs to the model, f is the activation function for the application of the element (a logical sigmoid or tanh function may typically be employed),
Figure BDA00019851776500000310
is a matrix of external input weights,
Figure BDA00019851776500000311
is an input weight matrix input from the (i-1) layer to the ith layer,
Figure BDA00019851776500000312
is a weight matrix in the reservoir of the ith layer, [ -; a]Representing a cascade of perpendicular vectors (or matrices), alpha e (0; 1)]For the leak rate, the project (parameter) associated with the ith floor reservoir is indicated using superscript (i).
For the output, the state of all reservoirs is used as input to the output layer, which in a typical linear case is calculated by the following formula:
Figure BDA00019851776500000313
wherein the content of the first and second substances,
Figure BDA0001985177650000041
is a matrix of the output weights,
Figure BDA0001985177650000042
is the output of the model, [ ·; h; …, respectively; a]Still representing a vertical vector (or matrix) concatenation.
Since the output of depesn is generally linear and feed forward, the above equation can be written in matrix form as:
Y＝WoutX
wherein the content of the first and second substances,
Figure BDA0001985177650000043
is the sum of all of y (n),
Figure BDA0001985177650000044
is all produced by the reservoir
Figure BDA0001985177650000046
T is the number of data points in the training data set.
WoutIs done by ridge regression, also called Tikhonov regularized regression, i.e.
Wout＝YXT(XXT+λI)-1
Where λ is the regularization coefficient and I is the identity matrix.
Proposed simultaneously with the deep echo state network (deepESN), there are two variants, full Input echo state network (deepESN Input to All, deepESN-IA) and echo state network group (groupedESN), respectively. As shown in fig. 2, compared with deepsesn, the two variants only modify the input mode, and add external input to each floor reservoir to obtain deepsesn-IA, and disconnect the floor-to-floor connections, and provide external input only to each floor to obtain groupedESN. These two variants differ from deepsesn only in the inputs in the state transitions, and the other formulas are equally applicable.
From this, it is possible to estimate the training time complexity of the model shown in fig. 1 and 2 to be O (N)TNLNR) In which N isTFor the number of input track points, NLNumber of reservoirs, NRThe number of neurons in each reservoir. Regardless of the length of the input trace sequence, the model shown in fig. 1 and 2 inputs trace points into the network to update the status x (n) of each reservoir. Thus, the training time complexity of the deep echo state network is several times that of the normal echo state network, so the high time complexityThe degree will result in a long model training process. In practical applications, in the case of a large training data scale, the training process of deepsesn and the two variants thereof is time-consuming, which affects the efficiency of model usage.
In order to solve the defect that the training of the traditional depESN and two variants thereof is time-consuming, and in consideration of the fact that the prefix and the suffix of a known track sequence play more roles in the comparison of the similarity of track segments, the invention provides a new variant of the depESN, which is called a Dual-Input Deep Echo State Network (Deep Echo State Network with Dual Input, depESN-DI).
Referring to fig. 3, in contrast to the standard deepsesn model, the present invention divides the first-level reservoir into two parts, each part containing
Figure BDA0001985177650000045
And (4) a neuron. The two input impounding reservoirs respectively process the prefix and the suffix of the known track sequence, and the outputs of the two impounding reservoirs are combined to be used as the input of the impounding reservoir of the next layer. The state transition function still applies, the only difference being the dimension of the input and output weight matrices.
Meanwhile, the length of the trajectory sequence used for training can be controlled by controlling the ratio of the prefix and the suffix. Assuming that the prefix and suffix each take 25% of the known trajectory sequence, the length of the sequence used for training is reduced to half of the original model. In addition, the prefix and the suffix are simultaneously input into two input water reservoirs to be processed, so that the training time is reduced by half. Therefore, the complexity of the training time of the depesn-DI is one fourth of the complexity of the original depesn and the two variants, which greatly shortens the training time of the network model of the invention, improves the utilization rate of the training data, and achieves better effect at the same time.
Experimental data prove that under large-scale training data, the final prediction effect of the deepsesn-DI is better than the prediction result of the existing best known destination prediction model, and meanwhile, the prediction result of the traditional deepsesn is better and the training time is short.
An evaluation metric for the destination prediction problem is the average Haversine Distance (MHD) between the predicted destination and the real destination, the Distance between two points on a sphere being measured in terms of latitude and longitude, commonly used in navigation. In the Kaggle challenge in 2015, the best performing model achieved an error of 2.39km (MHD). Under large-scale training data, the predicted performance of deepsesn is 3.04km (MHD) error, but the predicted performance of deepsesn-DI of the present invention is 1.86km (MHD) error. This performance indicates that deepsesn-DI achieves better prediction results than the traditional deepsesn model, and results are better than the best known prediction model. Meanwhile, the training time of the deePSEN-DI is greatly shortened compared with that of the original model, and the use efficiency of training data is greatly improved.
The destination prediction method based on the deep echo state network can be used for predicting the destination of a target with a running track, such as a taxi, a network appointment car, a robot and the like.
When the destination prediction method is used for taxi destination prediction, no additional information is used, the prediction effect is good, and the realization process of the taxi destination prediction method based on deePSEN-DI is as follows:
first, the deepsesn-DI specific network architecture (number of reservoirs N) of the present invention is set based on the forecasted demandLNeuron N contained in reservoir of non-first layerRNumber, number of input units NUAnd an output unit NYNumber, etc.) and using it as a predictive model of the rental car destination;
and then, inputting historical track data (a complete passenger carrying track segment of a taxi) of a certain area into a track database, wherein the first point of each track is a starting point, and the last point is a destination. Taking out the track data of the region in a complete period of time (for example, one week), slicing the extracted track data, keeping a track segment (no more than 75% in the present embodiment) from the starting point, and storing the accurate destination information of each training segment, wherein the track segment and the destination information are the training data of the deepsesn-DI model of the present invention;
then, the training data is input into the prediction model whose parameters (weight information) have been adjusted to meet the requirements, and the model extracts the prefixes and suffixes (set to 25% in this embodiment) of the input track segments according to a certain proportion. The prefix and the postfix track points are respectively input into the two input water reservoirs one by one for processing, the processed states of the two input water reservoirs are used as the input of the next layer of water reservoir, and the states of all the water reservoirs are updated in sequence.
Finally, combining the states of all the impounding reservoirs with the destination information ridge regression of the track segment to obtain an output weight matrix Wout. Until all track segments are input and trained.
And after the model training is finished, storing the states and the weight matrixes of all the water reservoirs. And taking a track segment of a taxi in any region, which contains an initial point, inputting the track segment into the prediction model, extracting a prefix and a suffix according to a certain proportion (which is consistent with that in training), and respectively inputting the prefix and the suffix into two input reservoirs for state upgrading. And finally, calculating the output of the prediction model according to the states of all the water reservoirs obtained by training and the weight matrix obtained by training, wherein the output is the destination of the taxi obtained by prediction.
In summary, the method for predicting the taxi destination based on deepsesn-DI of the present invention can at least provide the following technical effects:
firstly, a deep echo state network (deepsesn) is applied to a taxi destination prediction problem without using any additional information, and a good prediction effect is obtained;
secondly, a novel deep echo state network (deepsesn) variant is provided, which becomes a dual-input deep echo state network (deepsesn-DI), successfully solves the defect of high time consumption of the original deepsesn model and the variant thereof under large-scale training data, and achieves the best known and existing prediction performance.
While the invention has been described with reference to specific embodiments, any feature disclosed in this specification may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise; all of the disclosed features, or all of the method or process steps, may be combined in any combination, except mutually exclusive features and/or steps.

Claims (4)

1. The destination prediction method based on the deep echo state network is characterized by comprising the following steps:
setting a prediction model for a target object predicted by a destination, wherein the network structure of the prediction model is a double-input deep echo state network;
the dual input deep echo state network comprises NLA layer; wherein the first layer comprises two neurons
Figure FDA0001985177640000011
Reservoir of (2) to (N)LThe layers respectively comprise a neuron with the number of NRThe water reservoir; combining the outputs of the two reservoirs of the first layer as the output of the first layer, and using the output of each layer as the input of the next layer; linearly combining the outputs of all layers to obtain an output result of the double-input deep echo state network;
taking historical moving track data of a target object in a specified area as a training sample set, and carrying out track data preprocessing on the training sample set:
taking the first point of each track in the training sample set as a starting point and the last point as a destination;
for each track, keeping a section of track segment starting from the starting point as a training segment; and storing destination information of each training segment;
extracting prefixes and suffixes of each training fragment according to a preset ratio based on a training sample set subjected to trajectory data preprocessing, respectively inputting the extracted prefixes and the track points of the suffixes into two water reservoirs of a first layer of a double-input deep echo state network one by one, and performing deep echo state network training on a prediction model to obtain a trained prediction model;
and extracting prefixes and suffixes of the current travel track segment containing the starting point of the target object in the designated area according to a preset proportion, respectively inputting the extracted prefixes and the track points of the suffixes into two reservoirs of a first layer of a trained prediction model one by one, and obtaining a current destination prediction result of the target object based on the output of the prediction model.
2. The method of claim 1, wherein the preferred range of the ratio of the prefix to the suffix is 20% to 30%.
3. The method of claim 1, wherein the target object is a vehicle.
4. The method of claim 1, wherein the training segments comprise no more than 75% of the preferred proportion of each trajectory.
CN201910162617.6A 2019-03-05 2019-03-05 Destination prediction method based on deep echo state network Expired - Fee Related CN109978238B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910162617.6A CN109978238B (en) 2019-03-05 2019-03-05 Destination prediction method based on deep echo state network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910162617.6A CN109978238B (en) 2019-03-05 2019-03-05 Destination prediction method based on deep echo state network

Publications (2)

Publication Number Publication Date
CN109978238A CN109978238A (en) 2019-07-05
CN109978238B true CN109978238B (en) 2022-05-03

Family

ID=67077864

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910162617.6A Expired - Fee Related CN109978238B (en) 2019-03-05 2019-03-05 Destination prediction method based on deep echo state network

Country Status (1)

Country Link
CN (1) CN109978238B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112859924B (en) * 2021-01-27 2023-11-28 大连大学 Unmanned aerial vehicle track planning method combining artificial interference and ESN-PSO

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104346517A (en) * 2013-08-02 2015-02-11 杨凤琴 Echo state network based prediction method and prediction device
CN104463323A (en) * 2013-09-17 2015-03-25 北京邮电大学 Data prediction method and apparatus
CN107316103A (en) * 2017-06-04 2017-11-03 东南大学 Sight spot dynamic mass Forecasting Methodology based on echo state network
CN108710966A (en) * 2018-04-19 2018-10-26 重庆大学 A kind of photovoltaic power generation power prediction method based on more cluster ESN neural networks

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11188813B2 (en) * 2017-03-01 2021-11-30 Stc.Unm Hybrid architecture system and method for high-dimensional sequence processing

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104346517A (en) * 2013-08-02 2015-02-11 杨凤琴 Echo state network based prediction method and prediction device
CN104463323A (en) * 2013-09-17 2015-03-25 北京邮电大学 Data prediction method and apparatus
CN107316103A (en) * 2017-06-04 2017-11-03 东南大学 Sight spot dynamic mass Forecasting Methodology based on echo state network
CN108710966A (en) * 2018-04-19 2018-10-26 重庆大学 A kind of photovoltaic power generation power prediction method based on more cluster ESN neural networks

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
Deep-ESN: A Multiple Projection-encoding Hierarchical Reservoir Computing Framework;Qianli Ma;《journal of latex class files》;20171113;第14卷(第8期);1-15 *
基于改进回声状态网络的数据预测算法研究;朱毅等;《计算机与网络》;20180912(第17期);66-69 *
基于机器学习的智能出租车预测***;叶锋等;《计算机***应用》;20180915(第09期);61-67 *
基于神经网络的车辆轨迹目的地预测方法的研究;宋作华;《中国优秀硕士学位论文全文数据库 (工程科技Ⅱ辑) 信息科技》;20200815(第08期);C034-192 *
深度回声状态网络概述;程国建等;《电子科技》;20180727(第08期);92-95 *

Also Published As

Publication number Publication date
CN109978238A (en) 2019-07-05

Similar Documents

Publication Publication Date Title
CN110223517B (en) Short-term traffic flow prediction method based on space-time correlation
Tekouabou et al. Improving parking availability prediction in smart cities with IoT and ensemble-based model
Wang et al. DeepSD: Supply-demand prediction for online car-hailing services using deep neural networks
CN107862864B (en) Driving condition intelligent prediction estimation method based on driving habits and traffic road conditions
Xing et al. Energy oriented driving behavior analysis and personalized prediction of vehicle states with joint time series modeling
CN110415516B (en) Urban traffic flow prediction method and medium based on graph convolution neural network
CN111033535B (en) System and method for ride order scheduling
CN108629503B (en) Prediction method for taxi getting-on demand based on deep learning
CN104464304A (en) Urban road vehicle running speed forecasting method based on road network characteristics
CN111009134A (en) Short-term vehicle speed working condition real-time prediction method based on interaction between front vehicle and self vehicle
CN109272157A (en) A kind of freeway traffic flow parameter prediction method and system based on gate neural network
CN110979342B (en) Working condition information acquisition method for vehicle global energy management control
CN113554060B (en) LSTM neural network track prediction method integrating DTW
CN108091166A (en) Forecasting Methodology, device, equipment and the storage medium of available parking places number of variations
CN111695737B (en) LSTM neural network-based group target traveling trend prediction method
CN113159403B (en) Intersection pedestrian track prediction method and device
Shao et al. The Traffic Flow Prediction Method Using the Incremental Learning‐Based CNN‐LTSM Model: The Solution of Mobile Application
CN112396254A (en) Destination prediction method, destination prediction device, destination prediction medium, and electronic device
CN110543978A (en) Traffic flow data prediction method and device based on wavelet neural network
CN114780739A (en) Time sequence knowledge graph completion method and system based on time graph convolution network
CN113993172A (en) Ultra-dense network switching method based on user movement behavior prediction
CN110796317A (en) Urban taxi scheduling method based on demand prediction
CN109978238B (en) Destination prediction method based on deep echo state network
CN112559585A (en) Traffic space-time sequence single-step prediction method, system and storage medium
Huang et al. Short-term traffic prediction by two-level data driven model in 5G-enabled edge computing networks

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
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20220503