CN104900063B - A kind of short distance running time Forecasting Methodology - Google Patents
A kind of short distance running time Forecasting Methodology Download PDFInfo
- Publication number
- CN104900063B CN104900063B CN201510345977.1A CN201510345977A CN104900063B CN 104900063 B CN104900063 B CN 104900063B CN 201510345977 A CN201510345977 A CN 201510345977A CN 104900063 B CN104900063 B CN 104900063B
- Authority
- CN
- China
- Prior art keywords
- mrow
- self
- encoding encoder
- data
- msub
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 34
- 238000012549 training Methods 0.000 claims abstract description 28
- 238000013528 artificial neural network Methods 0.000 claims abstract description 7
- 238000013480 data collection Methods 0.000 claims abstract description 4
- 238000012360 testing method Methods 0.000 claims description 14
- 238000010606 normalization Methods 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000004422 calculation algorithm Methods 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 2
- 238000012545 processing Methods 0.000 claims 1
- 230000006870 function Effects 0.000 description 8
- 230000008569 process Effects 0.000 description 6
- 230000004913 activation Effects 0.000 description 5
- 239000011159 matrix material Substances 0.000 description 4
- 230000006399 behavior Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000007935 neutral effect Effects 0.000 description 3
- 230000000644 propagated effect Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 1
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000001373 regressive effect Effects 0.000 description 1
- 230000001932 seasonal effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
-
- 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
- G06N3/048—Activation functions
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Chemical & Material Sciences (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Analytical Chemistry (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention discloses a kind of short distance running time Forecasting Methodology, comprises the following steps:Step S1:Pre-processed using the progress data acquisition of traffic data collection device and the data to collection using method for normalizing;Step S2:Using the data pre-processed, training stacks self-encoding encoder depth network, obtains optimum network structure and correspondence parameter, generates optimal stacking self-encoding encoder depth network;Step S3:Call the optimal stacking self-encoding encoder depth neural network forecast vehicle short distance running time trained.The present invention can depth excavate non-linear relation and other internal characteristicses between input variable, precision of prediction is high, and robustness is good.
Description
Technical field
The invention belongs to intelligent transport system field, more particularly to a kind of short distance running time Forecasting Methodology.
Background technology
In preferential (Transit Signal Priority, the TSP) system of bus signals, generally require previous at crossing
The arrival of bus is detected at set a distance, and predicts that public transport reaches the time at crossing, it is excellent to give timely, accurate signal
First control.Many researchers and engineer recognize that accurate running time prediction is that successful implementation bus signals are preferential
It is crucial.
Existing time of vehicle operation forecast model mainly has four classes, is naive model, the mould based on traffic theory respectively
Type, model and mixed model based on data.Naive model does not need any training and parameter Estimation, simply simply will be nearest
The running time of generation is as next predicted value, or history running time is carried out simply to screen and weight as prediction
Value, or the combination of the two;This method does not adapt to traffic environment complicated and changeable.Model based on traffic theory attempts
Traffic behavior is rebuild in each time step, then running time is speculated from state variable, this class model is generally in all kinds of traffic
Realized inside simulation software, such as CORSIM, PARAMICS and DynaMIT, they can at large describe traffic behavior
Details, but set up this class model and generally require substantial amounts of professional knowledge.The precision of prediction of this class model depends on the sum rebuild
Similitude between actual traffic behavior, but be due to the complexity of real world, this similitude tends not to ensure.
Beneficial to counting the development with machine learning method, emerge many forecast models based on data and be used widely, it is this kind of
The purpose of model is to find a kind of appropriate data to explain function f and its parameter, and input is mapped into output with minimum error,
Wherein artificial neural network (Artificial Neural Network, ANN) model has non-linear in powerful presentation data
The ability of relation, has obtained extensive research, and derivation goes out a variety of variants.And mixed model is general by two or more inhomogeneities
The model combination of type, collection takes respective advantage further to improve estimated performance, such as by seasonal auto regressive moving average mould
Type and adaptive Kalman filter are bonded multi-step prediction device, and the mixed of multiple ANN model outputs is explained using Bayes rule
Close ANN model etc..The research for being related to mixed model is relatively fewer, and the estimated performance of this class model need further checking.
ANN model has just been paid attention to and fast-developing since being suggested in prediction field, because setting up this class model
Too many field relevant professional knowledge is not needed, and estimated performance is being improved constantly always.Traditional shallow-layer ANN model needs big
The training data of amount is to improve precision, but data volume is excessive and is easily trapped into over-fitting.Increase network layer can solve this
Individual problem, but traditional training method is slow in face of deep layer network convergence speed.Deep learning network research recent years
Breakthrough solves this contradiction, and it is proved to be able to efficiently solve prediction and other many problems.
The content of the invention
The practical application request that the present invention is predicted for short distance running time, proposes a kind of short distance running time prediction side
Method, can depth excavate non-linear relation and other internal characteristicses between input variable, precision of prediction is high, and robustness is good.
A kind of short distance running time Forecasting Methodology proposed by the present invention, comprises the following steps:
Step S1:Entered using the progress data acquisition of traffic data collection device and the data to collection using method for normalizing
Row pretreatment;
Step S2:Using the data pre-processed, training stacks self-encoding encoder depth network, obtain optimum network structure and
Correspondence parameter, generates optimal stacking self-encoding encoder depth network;
Step S3:Call the optimal stacking self-encoding encoder depth neural network forecast vehicle short distance running time trained.
It is preferred that, it is flat that the data that step S1 is gathered include Vehicle Speed, the magnitude of traffic flow, traffic current density or track
The actual travel time of equal vehicle number, signal time and vehicle between prediction terminal.
It is preferred that, step S1 the data of collection are used the specific method that method for normalizing is pre-processed for:Selection is treated
Normalize the minimum value v of variableminWith maximum vmax, to [vmin,vmax] between arbitrary value v, its normalize after value vnFor
It is preferred that, the step of step S2 generates optimal stacking self-encoding encoder depth network includes:
Step S21:Data after normalized are divided into training set and test set in proportion;
Step S22:The network structure for stacking self-encoding encoder depth network is set:The self-editing of self-encoding encoder is stacked including composition
Code device number L, and each self-encoding encoder hidden layer node number nl;
Step S23:Using training dataset, the parameter of each self-encoding encoder is successively trained non-supervisoryly;
Step S24:The self-encoding encoder trained in step S23 is stacked up, and fallout predictor is added in top layer, it is sharp again
With training data, each layer parameter of depth network is finely tuned with having supervision;
Step S25:Using test data set, forward calculation is carried out on the depth network trained, is obtained on test set
Average absolute predicated error;
Step S26:Change self-encoding encoder number L and each self-encoding encoder hidden layer node number nlNumerical value, weight
Multiple step S22~S25;The corresponding network structure of minimal error and parameter are chosen as the optimal of stacking self-encoding encoder depth network
Network structure and parameter.
It is preferred that, the step S3 further comprises the steps:
Step S31:When new vehicle is reached, gather the input data needed and done with same parameter at normalization
Reason;
Step S32:Input data after normalization is brought into the depth network trained and carries out forward calculation, correspondence is obtained
Normalization predicted value;
Step S33:The normalization predicted value renormalization that step S32 is obtained, obtains vehicle between prediction terminal
Prediction running time.
The present invention makes full use of the big of actual generation by constructing based on the depth Network Prediction Model for stacking self-encoding encoder
Measure traffic data training pattern, can automatically depth excavate input variable between non-linear relation and other internal characteristicses.
The present invention has versatility, can be deployed in different prediction places easily, and the precision of prediction with second level and higher
Robustness.
Brief description of the drawings
Fig. 1 is short distance running time Forecasting Methodology block schematic illustration of the present invention;
Fig. 2 is the optimal schematic flow sheet for stacking self-encoding encoder depth network of present invention generation;
Fig. 3 is single hidden layer self-encoding encoder structural representation;
Fig. 4 is the depth schematic network structure that the present invention is used to predict.
Embodiment
The present invention is described in detail below in conjunction with accompanying drawing, it is noted that described embodiment is only intended to just
In the understanding of the present invention, and any restriction effect is not played to it.
As shown in figure 1, a kind of short distance running time Forecasting Methodology of the present invention specifically comprises the following steps:
Step S1:Entered using the progress data acquisition of traffic data collection device and the data to collection using method for normalizing
Row pretreatment.
This method training stage needs the data gathered to include Vehicle Speed, the magnitude of traffic flow, traffic current density or car
The actual travel time of road average traffic number, signal time and vehicle between prediction terminal.Wherein Vehicle Speed
It can be directly obtained with the magnitude of traffic flow by ring coil detector.Traffic current density or track average traffic number can by
Ring coil detector is set respectively at prediction terminal, then statistics obtains (car by the difference of the vehicle number of two detectors
Road number is known constant).Signal time can be obtained by connecting crossing signals controller.Predict the actual row between terminal
The time of sailing can be obtained by detector timestamp information.In addition to ring coil detector, the inspection of higher level can also be used
Survey device and obtain data above, such as video detector, GPS device and AVL devices.
Next the data to collection are pre-processed, using method for normalizing.Normalized variable is treated for some, is taken
Its minimum value vminWith maximum vmax, then to [vmin,vmax] between arbitrary value v, its normalize after value vnSuch as formula (1) institute
Show.
Data set after above normalized is normalized is done to all variables (including output variable).
Step S2:Using the data pre-processed, training stacks self-encoding encoder depth network, obtain optimum network structure and
Correspondence parameter, generates optimal stacking self-encoding encoder depth network.The process flow diagram flow chart is as shown in Fig. 2 wherein AE represents automatic volume
Code device (Auto-Encoders), is specifically comprised the steps of:
Data after normalized are divided into training set and test set by step S21.
If data volume is sufficiently large, the data that can take 80% are training set, and 20% data are test set.Here
20%th, 80% simply recommendation is divided, suitable division should be taken during practical application as the case may be.If data volume is too
It is few, N can be used to roll over cross validation method and divide training set and test set to make full use of data.
Step S22, sets the network structure for stacking self-encoding encoder depth network:The self-editing of self-encoding encoder is stacked including composition
Code device number L, and each self-encoding encoder hidden layer node number nl。
Step S23, using training dataset, successively trains the parameter of each self-encoding encoder non-supervisoryly;" successively " mean
The input that the coding output of previous self-encoding encoder is used as to latter self-encoding encoder, the input of first self-encoding encoder comes from
Training set data.
Self-encoding encoder is a kind of input node and output node number identical neutral net, and simplest self-encoding encoder is only
There is a hidden layer, its structure is as shown in Figure 3.Kth layer directly input z(k)(overstriking body surface shows vector or matrix, similarly hereinafter)
Activation value a is transformed to by function f ()(k), k=1,2,3.Wherein first layer is that input layer need not activate conversion, and z(1)
Equal to input x, therefore x=z(1)=a(1).Function f () is referred to as activation primitive, sigmoid functions is taken, shown in such as formula (2).
One vectorial function f (z) is still a vector, and it represents to find a function after value each element of vector respectively
The vector of composition, i.e. f (z)=[z1,z2,…,zD]T, D is vector z dimension.
Later layer directly inputs z(k)By the activation value a of preceding layer(k-1)Obtained by linear combination, linear combination parameter
It is stored in matrix W(k-1)With vectorial b(k-1)In.The output of modelEqual to the activation value a of last layer(3)。
It is the propagated forward process of neutral net above, formula (3), (4), (5) is expressed as with the form of vector sum matrix:
X=z(1)=a(1) (3)
a(2)=f (z(2))=f (W(1)a(1)+b(1)) (4)
Wherein, W(1)And W(2)Respectively n1× M and M × n1Matrix, b(1)And b(2)Respectively n1The column vector tieed up with M, M
Represent the dimension of input vector, n1Represent that first self-encoding encoder hides the number of node layer.
Self-encoding encoder is different from general neural network part and is, it is reconstructed by encoding further decoding to input data, from
And find the form for expressing input data again.Therefore, the target of training self-encoding encoder is minimized between input and output
Error, adds the minimum target after regularization term (purpose is to avoid over-fitting) and is represented by formula (6):
Wherein W={ W(1),W(2), b={ b(1),b(2), m is training sample number, and λ is weight parameter, for adjusting mesh
The weight of reconstructed error and regularization term in scalar functions.
When hidden layer node number is more than input node number, self-encoding encoder training result is not often unique, and result
Depend critically upon primary condition.In order to solve this problem, in addition it is also necessary to add openness limitation.We useRepresent
In the case that given input is x, self-encoding encoder hides node layer j activity, therefore node layer j average activation is hidden in definition
Degree is as shown in formula (7)
It is desirable that average activity and ρ are closer to better, wherein ρ is one close to 0 normal number, referred to as openness ginseng
Number.It can be represented with ρ degree of closeness with KL distances, therefore plus shown in the object function such as formula (8) of sparsity constraints:
Wherein,β is weight parameter, for adjust J (W, b) and sparse
Property limitation weight.J is minimized using neutral net Back Propagation Algorithmsparse(W can b) try to achieve parameter W and b.
After first self-encoding encoder is trained, the i.e. coding output of output of node layer is hidden as new input,
Second self-encoding encoder is trained with same method.The like, until L self-encoding encoder, all training is completed.
Step S24, the self-encoding encoder trained in step S23 is stacked up, and adds fallout predictor in top layer, sharp again
With training data, each layer parameter of depth network is finely tuned with having supervision;
The hidden layer output of previous self-encoding encoder is directly connected to the hidden layer input i.e. structure of latter self-encoding encoder
Into stacking self-encoding encoder.The stacking self-encoding encoder of fallout predictor is added as shown in figure 4, in figure, x is some input sample;
For the coding output of l-th of self-encoding encoder, l=1,2 ..., L;For the kth group parameter of l-th of self-encoding encoder, k
=1,2;u0With the parameter that u is fallout predictor;For the fallout predictor output after renormalization.
Fallout predictor of the present invention uses logistic (Logistic) regression model, is expressed as formula (9):
WhereinTo stack the coding output of last self-encoding encoder in self-encoding encoder.Parameter u0Needed with u by instruction
Practice and obtain.Training data is reused herein, and further finely tuning depth network using the learning method for having supervision includes fallout predictor
Parameter, still using Back Propagation Algorithm.Suitable initial value has been obtained because stacking self-encoding encoder passes through pre-training,
Therefore supervised learning herein can restrain quickly.So far, the training process of depth network is fully completed.
Step S25, using test data set, carries out forward calculation on the depth network trained, obtains on test set
Average absolute predicated error;
Forward calculation only need to slightly extend on the basis of propagated forward process described in step S23, and by last
Multiple regression model is brought in the hidden layer output of individual self-encoding encoder into, you can try to achieve the predicted value of input sample, the predicted value
For normalized value, it is designated asIts actual value is obtained after renormalizationTherefore, test set
On average absolute predicated error such as formula (10) shown in:
Wherein m' is the number of samples of test set.
Step S26, changes self-encoding encoder number L in certain scope and each self-encoding encoder hides node layer
Number nlNumerical value, repeat step S22~S25;The corresponding network structure of minimal error and parameter are chosen as stacking self-encoding encoder
The optimum network structure and parameter of depth network.
Stack the possible structure number of self-encoding encoder with L increase exponentially to increase, in order to avoid " multiple shot array " is asked
Topic, when L is very big, it is not necessary to travel through each possible structure, but randomly generate the structure of certain amount (such as 1000),
Test checking is carried out on the depth network of these structures.
Step S3:Call the optimal stacking self-encoding encoder depth neural network forecast vehicle short distance running time trained.The step
Suddenly specifically include:
Step S31, when new vehicle is reached, is gathered the input data needed and is done with same parameter at normalization
Reason.
Data acquisition and method for normalizing are as it was previously stated, normalized parameter is obtained by step S1.
Step S32, the input data after normalization is brought into the depth network trained and carries out forward calculation, obtain correspondence
Normalization predicted value.
Forward calculation only need to slightly extend on the basis of propagated forward process described in step S23, and by last
Multiple regression model is brought in the hidden layer output of individual self-encoding encoder into, you can try to achieve the normalization predicted value of input sample.
Step S33, the normalization predicted value renormalization that step S32 is obtained obtains vehicle between prediction terminal
Prediction running time.
Renormalization process is as described in step S25.
It is described above, it is only the embodiment in the present invention, but protection scope of the present invention is not limited thereto, and appoints
What be familiar with the people of the technology disclosed herein technical scope in, it will be appreciated that the conversion or replacement expected, should all cover
Within the scope of the present invention.Therefore, protection scope of the present invention should be defined by the protection domain of claims.
Claims (4)
1. a kind of short distance running time Forecasting Methodology, it is characterised in that comprise the following steps:
Step S1:Data acquisition is carried out using traffic data collection device and pre- using method for normalizing progress to the data of collection
Processing;
Step S2:Using the data pre-processed, training stacks self-encoding encoder depth network, obtains optimum network structure and correspondingly
Parameter, generates optimal stacking self-encoding encoder depth network;
Step S3:Call the optimal stacking self-encoding encoder depth neural network forecast vehicle short distance running time trained;
The step of step S2 generates optimal stacking self-encoding encoder depth network includes:
Step S21:Data after normalized are divided into training set and test set in proportion;
Step S22:The network structure for stacking self-encoding encoder depth network is set:The self-encoding encoder of self-encoding encoder is stacked including composition
Number L, and each self-encoding encoder hidden layer node number nl, l=1,2 ..., L;
Step S23:Using training dataset, the parameter of each self-encoding encoder is successively trained non-supervisoryly;
Step S24:The self-encoding encoder trained in step S23 is stacked up, and fallout predictor is added in top layer, instruction is reused
Practice data, using Back Propagation Algorithm, finely tune each layer parameter of depth network with having supervision;
Wherein, the fallout predictor uses Multiple regression model:
<mrow>
<mi>&sigma;</mi>
<mrow>
<mo>(</mo>
<msubsup>
<mi>a</mi>
<mi>L</mi>
<mrow>
<mo>(</mo>
<mn>2</mn>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mrow>
<mn>1</mn>
<mo>+</mo>
<mi>exp</mi>
<mrow>
<mo>(</mo>
<mo>-</mo>
<msub>
<mi>u</mi>
<mn>0</mn>
</msub>
<mo>-</mo>
<msup>
<mi>u</mi>
<mi>T</mi>
</msup>
<msubsup>
<mi>a</mi>
<mi>L</mi>
<mrow>
<mo>(</mo>
<mn>2</mn>
<mo>)</mo>
</mrow>
</msubsup>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
</mrow>
To stack the coding output of last self-encoding encoder in the self-encoding encoder, parameter u0Obtained with u by training;
Step S25:Using test data set, forward calculation is carried out on the depth network trained, obtains flat on test set
Equal absolute prediction error;
Step S26:Change self-encoding encoder number L and each self-encoding encoder hidden layer node number nlNumerical value, repeat step
S22~S25;The corresponding network structure of minimal error and parameter are chosen as the optimal network knot for stacking self-encoding encoder depth network
Structure and parameter.
2. a kind of short distance running time Forecasting Methodology as claimed in claim 1, it is characterised in that the data that step S1 is gathered
Predicted including Vehicle Speed, the magnitude of traffic flow, traffic current density or track average traffic number, signal time and vehicle
Actual travel time between stop.
3. a kind of short distance running time Forecasting Methodology as claimed in claim 1, it is characterised in that data of the step S1 to collection
Use the specific method that method for normalizing is pre-processed for:Choose the minimum value v of variable to be normalizedminWith maximum vmax,
To [vmin,vmax] between arbitrary value v, its normalize after value vnFor
<mrow>
<msub>
<mi>v</mi>
<mi>n</mi>
</msub>
<mo>=</mo>
<mfrac>
<mrow>
<mi>v</mi>
<mo>-</mo>
<msub>
<mi>v</mi>
<mrow>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
</mrow>
</msub>
</mrow>
<mrow>
<msub>
<mi>v</mi>
<mrow>
<mi>m</mi>
<mi>a</mi>
<mi>x</mi>
</mrow>
</msub>
<mo>-</mo>
<msub>
<mi>v</mi>
<mrow>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
</mrow>
</msub>
</mrow>
</mfrac>
<mo>.</mo>
</mrow>
4. a kind of short distance running time Forecasting Methodology as claimed in claim 1, it is characterised in that the step S3 is further wrapped
Include following steps:
Step S31:When new vehicle is reached, gather the input data needed and normalized is done with same parameter;
Step S32:Input data after normalization is brought into the depth network trained and carries out forward calculation, corresponding return is obtained
One changes predicted value;
Step S33:The normalization predicted value renormalization that step S32 is obtained, obtains vehicle pre- between prediction terminal
Survey running time.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510345977.1A CN104900063B (en) | 2015-06-19 | 2015-06-19 | A kind of short distance running time Forecasting Methodology |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510345977.1A CN104900063B (en) | 2015-06-19 | 2015-06-19 | A kind of short distance running time Forecasting Methodology |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104900063A CN104900063A (en) | 2015-09-09 |
CN104900063B true CN104900063B (en) | 2017-10-27 |
Family
ID=54032703
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510345977.1A Active CN104900063B (en) | 2015-06-19 | 2015-06-19 | A kind of short distance running time Forecasting Methodology |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104900063B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109887290A (en) * | 2019-03-30 | 2019-06-14 | 西安电子科技大学 | Traffic flow forecasting method based on equilibrium index exponential smoothing and stack self-encoding encoder |
Families Citing this family (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106096767A (en) * | 2016-06-07 | 2016-11-09 | 中国科学院自动化研究所 | A kind of link travel time prediction method based on LSTM |
CN106778571B (en) * | 2016-12-05 | 2020-03-27 | 天津大学 | Digital video feature extraction method based on deep neural network |
CN106875511A (en) * | 2017-03-03 | 2017-06-20 | 深圳市唯特视科技有限公司 | A kind of method for learning driving style based on own coding regularization network |
CN109417767B (en) | 2017-04-27 | 2020-09-18 | 北京嘀嘀无限科技发展有限公司 | System and method for determining estimated time of arrival |
CN106981198B (en) * | 2017-05-24 | 2020-11-03 | 北京航空航天大学 | Deep learning network model for travel time prediction and establishing method thereof |
WO2019028763A1 (en) | 2017-08-10 | 2019-02-14 | Beijing Didi Infinity Technology And Development Co., Ltd. | System and method for estimating travel time and distance |
CN108417029B (en) * | 2018-02-11 | 2020-11-06 | 东南大学 | Urban road network travel time estimation method based on self-adaptive multitask deep learning |
CN110517494A (en) * | 2019-09-03 | 2019-11-29 | 中国科学院自动化研究所 | Forecasting traffic flow model, prediction technique, system, device based on integrated study |
CN111210633B (en) * | 2020-02-09 | 2021-01-05 | 北京工业大学 | Short-term traffic flow prediction method based on deep learning |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1434946A (en) * | 2000-07-21 | 2003-08-06 | 交通预测.Com公司 | Method of providing travel time predictions |
CN101925940A (en) * | 2007-12-20 | 2010-12-22 | 意大利电信股份公司 | Method and system for forecasting travel times on roads |
CN104091081A (en) * | 2014-07-15 | 2014-10-08 | 中国科学院自动化研究所 | Traffic data make-up method |
CN104134351A (en) * | 2014-08-14 | 2014-11-05 | 中国科学院自动化研究所 | Short-term traffic flow predicting method |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP3932383B2 (en) * | 2002-09-27 | 2007-06-20 | 財団法人生産技術研究奨励会 | Traveling time prediction device |
JP2006195907A (en) * | 2005-01-17 | 2006-07-27 | Denso Corp | Onboard navigation apparatus and predicted driving time acquiring system |
-
2015
- 2015-06-19 CN CN201510345977.1A patent/CN104900063B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1434946A (en) * | 2000-07-21 | 2003-08-06 | 交通预测.Com公司 | Method of providing travel time predictions |
CN101925940A (en) * | 2007-12-20 | 2010-12-22 | 意大利电信股份公司 | Method and system for forecasting travel times on roads |
CN104091081A (en) * | 2014-07-15 | 2014-10-08 | 中国科学院自动化研究所 | Traffic data make-up method |
CN104134351A (en) * | 2014-08-14 | 2014-11-05 | 中国科学院自动化研究所 | Short-term traffic flow predicting method |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109887290A (en) * | 2019-03-30 | 2019-06-14 | 西安电子科技大学 | Traffic flow forecasting method based on equilibrium index exponential smoothing and stack self-encoding encoder |
CN109887290B (en) * | 2019-03-30 | 2021-03-23 | 西安电子科技大学 | Traffic flow prediction method based on balance index smoothing method and stack type self-encoder |
Also Published As
Publication number | Publication date |
---|---|
CN104900063A (en) | 2015-09-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104900063B (en) | A kind of short distance running time Forecasting Methodology | |
CN110223517B (en) | Short-term traffic flow prediction method based on space-time correlation | |
Shao et al. | Traffic flow prediction with long short-term memory networks (LSTMs) | |
CN102867407B (en) | Multistep prediction method for effective parking space occupation rate of parking lot | |
CN113053115B (en) | Traffic prediction method based on multi-scale graph convolution network model | |
CN104134351B (en) | A kind of Short-time Traffic Flow Forecasting Methods | |
CN103700255B (en) | A kind of traffic flow forecasting method based on spacetime correlation data mining | |
CN103226741B (en) | Public supply mains tube explosion prediction method | |
Jin et al. | Neural network multitask learning for traffic flow forecasting | |
CN102622418B (en) | Prediction device and equipment based on BP (Back Propagation) nerve network | |
CN109886444A (en) | A kind of traffic passenger flow forecasting, device, equipment and storage medium in short-term | |
CN107464419A (en) | A kind of Short-time Traffic Flow Forecasting Methods for considering space-time characterisation | |
CN107180530A (en) | A kind of road network trend prediction method based on depth space-time convolution loop network | |
CN105389980A (en) | Short-time traffic flow prediction method based on long-time and short-time memory recurrent neural network | |
CN104217258B (en) | A kind of electric load sigma-t Forecasting Methodology | |
CN105160866A (en) | Traffic flow prediction method based on deep learning nerve network structure | |
CN106570597A (en) | Content popularity prediction method based on depth learning under SDN architecture | |
CN110070715A (en) | A kind of road traffic flow prediction method based on Conv1D-NLSTMs neural network structure | |
CN106529818A (en) | Water quality evaluation prediction method based on fuzzy wavelet neural network | |
CN101893674A (en) | Pollution flashover index forecasting method for regional power grid | |
CN107705556A (en) | A kind of traffic flow forecasting method combined based on SVMs and BP neural network | |
CN103886395A (en) | Reservoir optimal operation method based on neural network model | |
CN106933649A (en) | Virtual machine load predicting method and system based on rolling average and neutral net | |
CN111141879B (en) | Deep learning air quality monitoring method, device and equipment | |
CN114565187A (en) | Traffic network data prediction method based on graph space-time self-coding network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |