CN110263962A - Termination environment is marched into the arena Tendency Prediction method - Google Patents
Termination environment is marched into the arena Tendency Prediction method Download PDFInfo
- Publication number
- CN110263962A CN110263962A CN201910162007.6A CN201910162007A CN110263962A CN 110263962 A CN110263962 A CN 110263962A CN 201910162007 A CN201910162007 A CN 201910162007A CN 110263962 A CN110263962 A CN 110263962A
- Authority
- CN
- China
- Prior art keywords
- arena
- marching
- model
- termination environment
- marched
- 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
- 238000000034 method Methods 0.000 title claims abstract description 38
- 238000012360 testing method Methods 0.000 claims abstract description 21
- 230000007787 long-term memory Effects 0.000 claims abstract description 13
- 238000012549 training Methods 0.000 claims description 21
- 239000011159 matrix material Substances 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 11
- 238000000926 separation method Methods 0.000 claims description 7
- 238000010606 normalization Methods 0.000 claims description 5
- 230000008859 change Effects 0.000 claims description 4
- 230000007704 transition Effects 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 2
- 238000006243 chemical reaction Methods 0.000 claims description 2
- 238000004140 cleaning Methods 0.000 claims description 2
- 230000008569 process Effects 0.000 claims description 2
- 238000012545 processing Methods 0.000 claims description 2
- 230000009467 reduction Effects 0.000 claims description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims 1
- 238000005259 measurement Methods 0.000 abstract description 6
- 238000011156 evaluation Methods 0.000 description 11
- 238000011160 research Methods 0.000 description 7
- 238000010801 machine learning Methods 0.000 description 4
- 210000004027 cell Anatomy 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 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
- 230000006399 behavior Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 239000004615 ingredient Substances 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- 230000016776 visual perception Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Theoretical Computer Science (AREA)
- Tourism & Hospitality (AREA)
- General Physics & Mathematics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
It marches into the arena Tendency Prediction method the present invention relates to a kind of termination environment comprising: the forecasting traffic flow model of marching into the arena based on shot and long term memory network is established, according to different parametric configuration model sets;Optimal models in Selection Model set, and selected optimal models are verified on actual test collection;The parameter of optimal models is fixed, and traffic jam level is predicted.By establishing marching into the arena forecasting traffic flow model and obtain situation of marching into the arena based on shot and long term memory network, the traffic circulation situation of objective convenient for administrative staff, accurate measurement termination environment.
Description
Technical field
The present invention relates to aviation fields, and in particular to a kind of termination environment is marched into the arena Tendency Prediction method.
Background technique
In recent years, the termination environment congestion problems as caused by the increase of flight amount become increasingly conspicuous.Termination environment is in route grid
In specific position become high density, the highly complex busy airspace of typical case, the traffic of objective, accurate measurement termination environment
Operation situation analyzes its regular dynamic change characterization, is that auxiliary controlling officer takes suitable flow allocating measure necessity branch
Hold means.Track is the objective record of aircraft interaction and control intervening act, changeability with higher.Another party
Face, aircraft run the constraint by intended path and mission program, and track has certain normalization and convergence property, by end
The clustering recognition method of petiolarea traffic flow can extract difference and march into the arena traffic flow character.
Traffic situation refers to the traffic behavior and situation that all aircraft operation actions are constituted in airspace, and expression airspace is whole
Body and global concept, and the layer that situation emulation, perception and display are still rested essentially within for airspace traffic situation domestic at present
Face not yet forms effective Appraisal System & Method.In September, 2010, Li Jie et al. deliver " traffic in air traffic management system
The research and application of battle state display ", mainly discuss some problems and key technology in traffic situation display research;2011
2 months, Zhao's a word used in place name fly et al. delivered at " Civil Aviation University of China journal " " the airway traffic Situation Assessment based on fuzzy overall evaluation ", text
In discuss traffic situation index and integrated evaluating method only for air route.Unit relevant traffic situation in airspace is commented as a result,
There are biggish application demands and research space for valence.Present Research at present about airspace traffic situation is as follows:
(1) not yet by traffic situation from basic visual perception, it is abstracted as specific data information.
(2) complete sector situation evaluation index set and system are not yet established, in existing index set and in a certain respect,
And index granularity refinement degree is lower.
(3) in addition to situation evaluation in air route rarely has research, to the evaluation of control sector/termination environment situation synthesis still in sky
It is white, airspace management and traffic management are lacked and supported.
Therefore marching into the arena for termination environment, accurately prediction can make up above-mentioned white space for traffic situation progress, to be
Termination environment controller makes correct decisions and provides important evidence.In general, it is ground at present about the evaluation of traffic situation with prediction
It is as follows to study carefully status:
(1) generally using the overall situation of traffic as object, evaluation is also rested on substantially to traffic flow research, it is not yet right
Traffic flow situation is effectively predicted.
(2) basic or use traditional measure of criterions for the evaluation of air traffic situation, fail effectively to combine it is microcosmic and
The traffic characteristics measurement of macroscopic view.
(3) more for ingredient subjective in the evaluation of air traffic situation, fail to become using the numerical value of traffic characteristics itself
Change feature.
Therefore, quantitative and science research is carried out to termination environment traffic flow situation, can make more accurate prediction with
Evaluation, when another evaluation from outside traffic flow situation and influence factor, from acquired traffic flow data, screening is extracted and fixed
The certain key indexes of justice and the selection and extraction that scientific method is carried out to index, can with when subsequent prediction result it is more preferable.From finger
Calibration is adopted and screens, arrives the label grade separation based on cluster again finally carries out the prediction based on deep learning, it is possible to reduce
Artificial subjective assessment, the evaluation and prediction of termination environment traffic situation are more accurate, scientific.
On the other hand, due to the preeminent accomplishment that machine learning obtains in each field, many scholars apply to machine learning
Civil aviaton field.It is marched into the arena feature of the influence factor as machine learning model of congestion by the way that termination environment will be caused, termination environment is marched into the arena
Situation grade as prediction result the mathematical model established is trained.Because policymaker often can not directly obtain
Take correlative factor to influence caused by flight operation, so the prediction technique based on machine learning has obtained some actual fortune
With, however due to the high changeability of the high density of track and aircraft, utilize the predictablity rate of the method for conventional machines study
Tend not to the termination environment control for meeting airport.
How to solve the above problems, is urgently to be resolved at present.
Summary of the invention
It marches into the arena Tendency Prediction method the object of the present invention is to provide a kind of termination environment.
It marches into the arena Tendency Prediction method in order to solve the above-mentioned technical problems, the present invention provides a kind of termination environment, comprising:
The forecasting traffic flow model of marching into the arena based on shot and long term memory network is established, according to different parametric configuration Models Sets
It closes;
Optimal models in Selection Model set, and selected optimal models are verified on actual test collection;
The parameter of optimal models is fixed, and traffic jam level is predicted.
The invention has the advantages that marching into the arena Tendency Prediction method the present invention provides a kind of termination environment comprising: it establishes
Forecasting traffic flow model of marching into the arena based on shot and long term memory network, according to different parametric configuration model sets;Selection Model
Optimal models in set, and selected optimal models are verified on actual test collection;The parameter of optimal models is fixed, and
Traffic jam level is predicted.By establishing marching into the arena forecasting traffic flow model and obtain based on shot and long term memory network
It marches into the arena situation, the traffic circulation situation of objective convenient for administrative staff, accurate measurement termination environment.
Detailed description of the invention
Present invention will be further explained below with reference to the attached drawings and examples.
Fig. 1 is that termination environment provided by the present invention is marched into the arena the flow chart of Tendency Prediction method.
Fig. 2 is additional flying distance provided by the present invention and relative flight speed scatter plot.
Fig. 3 is the track goodness of fit provided by the present invention and relative flight speed scatter plot.
Fig. 4 is LSTM unit internal structure chart provided by the present invention.
Fig. 5 is prediction model concept map provided by the present invention.
Specific embodiment
In conjunction with the accompanying drawings, the present invention is further explained in detail.These attached drawings are simplified schematic diagram, only with
Illustration illustrates basic structure of the invention, therefore it only shows the composition relevant to the invention.
Embodiment
It marches into the arena Tendency Prediction method as shown in Figure 1, present embodiments providing a kind of termination environment.Termination environment is marched into the arena Tendency Prediction
Method is convenient for administrator by marching into the arena forecasting traffic flow model and obtain situation of marching into the arena based on establishing shot and long term memory network
The traffic circulation situation of objective, the accurate measurement termination environment of member.Include:
S110: the forecasting traffic flow model of marching into the arena based on shot and long term memory network is established, according to different parametric configurations
Model set;
S120: the optimal models in Selection Model set, and selected optimal models are verified on actual test collection;
S130: the parameter of optimal models is fixed, and traffic jam level is predicted.
Wherein, step S110 includes:
S111: traffic flow data of marching into the arena is obtained;
S112: according to marching into the arena, traffic flow data obtains the traffic flow sample set D={ x that marches into the arena1, x2..., xm}
S113: the traffic characteristic of marching into the arena of definition;
S114: congestion levels classification is calculated;
S115: cleaning data and pre-processed, according to traffic flow character and grade separation result building training set with
Test machine;
S116: traffic flow situation grade prediction model of marching into the arena is established, and prediction model set is set.
In the present embodiment, due to air station flight have height convergence, can be regarded as an entirety with airport
For the convergence flow of convergent point, step S113 includes:
Additional flight time Ad_Tk, refer to that the flight time for aviation of marching into the arena in termination environment in statistical time piece k is right beyond its
The average duration for the reference track reference time answered, it may be assumed that
Wherein m is the traffic flow quantity of marching into the arena of termination environment, NjIt marches into the arena for jth and flows track quantity, t in timeslice kijFor
It marches into the arena and flows the flight duration of i-th of track in j,For the reference track duration for flowing j of marching into the arena;
Additional flying distance Ad_Dk, refer to that the flying distance for aviation of marching into the arena in termination environment in statistical time piece k is right beyond its
The average distance for the reference track length answered, it may be assumed that
Wherein dijFor the flight duration for flowing i-th of track in j of marching into the arena,For the reference track distance for flowing j of marching into the arena;
Relative velocityRefer in statistical time piece k, termination environment is marched into the arena reference track length corresponding to aircraft
With the average level of actual flying time ratio, it may be assumed thatWhereinI-th of aviation in j is flowed to march into the arena
The relative flight speed of device;
Track goodness of fit Av_Ck, refer to statistical time piece k, all aircraft flight profiles are corresponding with anchor point with reference to boat
The average level of otherness between mark uses Euclidean distance, c ' apart from calculationijBe conversion after the goodness of fit, the track goodness of fit,
That is:
It crosses and is lined up quantity Av_Wk, refer to the aircraft for entering termination environment in statistical time piece k, Zi into termination environment to
Land the average level of preamble flight quantity in time range, it may be assumed that
wijFor the queuing number for flowing i-th of aircraft in j of marching into the arena
Average rate of decrease Av_Hk, refer in statistical time piece k, aircraft of marching into the arena enters the vertical fall off rate in termination environment
Average level, wherein hijFor the average fall off rate for flowing i-th of aircraft in j of marching into the arena, average rate of decrease, it may be assumed that
Total flight timeRefer in statistical time piece k, respectively marching into the arena, it is total to flow each aircraft
Flight time;
Total flying distanceRefer in statistical time piece k, respectively marching into the arena, it is total to flow each aircraft
Flying distance.
It is selected to march into the arena stream attribute and traffic situation of marching into the arena all has an impact, observe the traffic of marching into the arena based on measured data
Numeric distribution between feature finds the regular changes in distribution between feature and changing features and terminal by nonlinear fitting curve
The relationship of area's jam situation.Such as shown in Fig. 2, additional distance be less than -8 negative value section, relative flight speed maintain compared with
High level;It is near 0 in additional flying distance, numerical point is obviously intensive, and relative velocity rapid decrease shows that aircraft is abided by substantially
Follow reference track operation.Each stage respectively corresponds in figure shown in each dotted line framework, can characterize unimpeded, transition and busy shape respectively
State.From the point of view of the scatterplot distribution situation in Fig. 2, Fig. 3, termination environment jam situation and additional flight time, relative flight speed, boat
The mark goodness of fit has close association.Following table is the correlation data and F inspection data of each data characteristics, and showing between each feature has
Preferable relevance.
Table one: the correlation data and F inspection data of each data characteristics
In the present embodiment, step S114 includes: and obtains congestion levels using clustering algorithm to classify, i.e. the spacing of clustering cluster
From the average distance that metric function d uses Hao Siduofu, it may be assumed that
Wherein, dist (x, z) indicates corresponding two sample C in two class clustersi, CjIndicate current two for calculating distance
Class cluster;
3 are set by cluster numbers k, cluster process is as follows:
Current clustering cluster number: q=m is set;
While q > k do
Find out two nearest clustering clusters of distanceWith
MergeWith
For j=j*+1, j*+2 ..., q do
C will be clusteredjIt is renumbered as Cj-1;
End for
The jth * row and jth * for deleting distance matrix M arrange;
It exports cluster and divides C={ C1, C2, C3, three kinds of traffic flow situation classifications of marching into the arena are obtained, unimpeded state, transition are respectively corresponded
State, congestion state.
In the present embodiment, step S115 includes:
Individual features and jam level are calculated according to actual data set, and jam level field is divided into three classes,
Convert thereof into corresponding class label c0, c1, c2.In order to improve prediction accuracy, also jam level label is carried out
Embedding processing, most obtains its one-hot matrix, such as the corresponding coding c of jam level2, then its one-hot matrix be [0,
0,1].
Trained and test data set D '={ (X is obtained according to classification results and one-hot matrixi, Ci), Xi=
(xi1, xi2..., xid), Ci=(c0, c1, c2);XiIndicate the input dimensional matrix of the i-th data, CiFor label matrix.The two
Merging becomes data set.{(Xi, Ci) be mathematical set representation, xi1Indicate first element value of the matrix.
To data set according to formulaIt is normalized, wherein diIt is traffic of marching into the arena certain period
Flow data, Max and Min respectively indicate each profile maxima and minimum value in data, xiIt is diNormalization result;
Data set after normalization is divided into test set D by fixed proportion as unit of timestep1With training set D2, will
Training set is divided into the equal equal portions of m, that is, D1={ T1, T2..., Ttimestep, Ttimestep+1..., T2*timestep..., Tm*timestep,
Test set is divided into the equal equal portions of n, that is, D2={ T1, T2..., Ttimestep, Ttimestep+1..., T2*timestep...,
Tn*timestep, wherein Ti=(xi1, xi2..., xid, yi)=(Xi, Ci), wherein d is characterized number.
In the present embodiment, step S116 includes:
According to the friendship of marching into the arena of traffic characteristic and the foundation of grade separation result based on shot and long term memory network (LSTM) of marching into the arena
Through-flow jam level prediction model.Prediction model can be obtained not with 15 minutes for a period by above-mentioned AGNES cluster
Congestion (unimpeded), slight congestion (transition), the data of three kinds of severe congestion (congestion) delay states.In the congestion data divided
In sequence, the congestion of preceding several periods will cause the congestion of following sessions.Model is specifically made of several LSTM cells.It is based on
The termination environment congestion data of the present period of input, the LSTM unit in model not only calculate and export the congestion etc. of present period
Grade, is transferred to next LSTM cell by network structure for data information, and information is supplied to the calculating of next period
It uses, which can calculate and export the congestion of the termination environment of following sessions according to the information of the present period termination environment of part
Grade.The internal structure chart of LSTM unit is as shown in figure 4, the concept map of prediction model is as shown in Figure 5.
Every time using the data of a time series as the input of model, the last output of model is the traffic congestion of prediction
Grade;
According to the initialization sequence length of candidate LSTM model, the LSTM number of plies, cell number, Dropout probability and study
Rate a, it is constant according to fixed other parameters, change the method setting prediction model set M={ m of a certain preset parameter1, m2...,
mk, wherein m1For the 1st candidate family, m2For the 2nd candidate family, mkFor k-th of candidate family;
To each of M model mi, initialize network weightWherein W is all connection phases
Adjacent two layers of weight, b are the bias terms of each layer, and N (0,1) is standardized normal distribution.
In the present embodiment, step S120: include:
A model m is taken out from prediction model set Mj, and simultaneously initialization model m is setjNetwork weight
To the training data of training setNetwork parameter θ is successively trained using greedy algorithm(j), j=
1,2 ..., mi, i.e., first with the first layer of input data training depth noise reduction autoencoder network, generate the ginseng of the first layer network
Number W(1)And b(1);Then the input by the output of first layer as the second layer continues training and obtains the parameter W of the second layer(2)And b(2);Finally to each layer below using same strategy, i.e., the mode that the output of front layer is inputted as next layer is successively trained, it is right
In above-mentioned training method, when training each layer parameter, other each layer parameters can be fixed and remained unchanged, greedy algorithm is utilized
The network parameter of layer-by-layer training pattern adjusts network parameter by backpropagation, the parameter value after obtaining model learning.
The volume size of hypothesized model is that m (layer) multiplies n (number of every layer of neuron), and n here is exactly every layer of neuron
Number.
The loss objective function of model uses cross entropy loss function Loss=- ∑ yiln ai, wherein yiWhat is represented is true
Real value, and aiWhat is represented is the predicted value of model;
When being trained according to loss objective function, need to define the size of learning rate, learning rate a formula are as follows:Wherein θ represents the value of the weight of network, and j indicates model in trained iterative steps;
Regularization constraint is provided with to learning parameter, the objective function after regularization constraint are as follows:
Wherein θ={ W, b };W is the weight of all connection adjacent two layers;B is the bias term of each layer;λ is then used to measure number
According to the weight between reconstruct degree and regularization constraint, the concept input matrix of x representative model.L is the actual loss letter of model
Number, J are the loss function of non-regularization.
If empiric riskIt is not converged, then Model Weight is updated according to learning rate a iterationUntil empiric riskTend to restrain, and is tested the termination environment jam level predicted and accuracy rate Accuracy on test set.It will
The model that the data of test set input to obtains actual test accuracy rateEnable M=M { mi, ifThen choosing the highest model of Accuracy is optimal models, otherwise continues Selection Model training.
In the present embodiment, step S130 includes:
The parameter of optimal models in step S120 is fixed;
Certain timestep item is marched into the arena traffic flow data (Tj, Tj+1..., Tj+timestep), wherein Ti=(xi1, xi2...,
xid, yi) optimal models are inputed to, the traffic flow for obtaining the prediction of next time granularity is marched into the arena situation ci, wherein (0,1,2) i ∈.
In conclusion marching into the arena Tendency Prediction method the present invention provides a kind of termination environment comprising: establish shot and long term memory
Forecasting traffic flow model of marching into the arena based on network, according to different parametric configuration model sets;In Selection Model set most
Excellent model, and selected optimal models are verified on actual test collection;The parameter of optimal models is fixed, and to traffic congestion
Grade is predicted.By establishing marching into the arena forecasting traffic flow model and obtain situation of marching into the arena based on shot and long term memory network,
The traffic circulation situation of objective convenient for administrative staff, accurate measurement termination environment.
Taking the above-mentioned ideal embodiment according to the present invention as inspiration, through the above description, relevant staff is complete
Various changes and amendments can be carried out without departing from the scope of the technological thought of the present invention' entirely.The technology of this invention
Property range is not limited to the contents of the specification, it is necessary to which the technical scope thereof is determined according to the scope of the claim.
Claims (8)
- A kind of Tendency Prediction method 1. termination environment is marched into the arena characterized by comprisingThe forecasting traffic flow model of marching into the arena based on shot and long term memory network is established, according to different parametric configuration model sets;Optimal models in Selection Model set, and selected optimal models are verified on actual test collection;The parameter of optimal models is fixed, and traffic jam level is predicted.
- The Tendency Prediction method 2. termination environment as described in claim 1 is marched into the arena, which is characterized in that described to establish shot and long term memory net Forecasting traffic flow model of marching into the arena based on network, the method according to different parametric configuration model sets include:Acquisition is marched into the arena traffic flow data;According to marching into the arena, traffic flow data obtains the traffic flow sample set D={ x that marches into the arena1, x2..., xm}The traffic characteristic of marching into the arena of definition;Calculate congestion levels classification;Data are cleaned and are pre-processed, according to traffic flow character and grade separation result building training set and test machine;Foundation is marched into the arena traffic flow situation grade prediction model, and prediction model set is arranged.
- The Tendency Prediction method 3. termination environment as claimed in claim 2 is marched into the arena, which is characterized in thatThe method of the traffic characteristic of marching into the arena of the definition includes:Additional flight time Ad_Tk, refer to that the flight time for aviation of marching into the arena in termination environment in statistical time piece k is corresponding beyond its The average duration of reference track reference time, it may be assumed thatWherein m is the traffic flow quantity of marching into the arena of termination environment, NjIt marches into the arena for jth and flows track quantity, t in timeslice kijFor stream of marching into the arena The flight duration of i-th of track in j,For the reference track duration for flowing j of marching into the arena;Additional flying distance Ad_Dk, refer to that the flying distance for aviation of marching into the arena in termination environment in statistical time piece k is corresponding beyond its The average distance of reference track length, it may be assumed thatWherein dijFor the flight duration for flowing i-th of track in j of marching into the arena,For the reference track distance for flowing j of marching into the arena;Relative velocity Av_Vk r, refer in statistical time piece k, termination environment is marched into the arena reference track length and reality corresponding to aircraft The average level of border flight time ratio, it may be assumed thatWhereinFor the phase for flowing i-th of aircraft in j of marching into the arena To flying speed;Track goodness of fit Av_Ck, refer to statistical time piece k, it is poor between all aircraft flight profiles reference track corresponding with anchor point Anisotropic average level uses Euclidean distance, c ' apart from calculationijIt is the goodness of fit after conversion, the track goodness of fit, it may be assumed thatIt crosses and is lined up quantity Av_Wk, refer to the aircraft for entering termination environment in statistical time piece k, Zi into termination environment to when landing Between in range preamble flight quantity average level, it may be assumed thatWherein, wijFor the queuing number for flowing i-th of aircraft in j of marching into the arena;Average rate of decrease Av_Hk, refer in statistical time piece k, aircraft of marching into the arena enters the average water of the vertical fall off rate in termination environment It puts down, wherein hijFor the average fall off rate for flowing i-th of aircraft in j of marching into the arena, average rate of decrease, it may be assumed thatTotal flight timeRefer in statistical time piece k, respectively marches into the arena and flow the total flight of each aircraft Time;Total flying distanceRefer in statistical time piece k, respectively marches into the arena and flow the total flight of each aircraft Distance.
- The Tendency Prediction method 4. termination environment as claimed in claim 3 is marched into the arena, which is characterized in that the calculating congestion levels classification Method include: using clustering algorithm obtain congestion levels classify, i.e.,Distance metric function d uses the average distance of Hao Siduofu between clustering cluster, it may be assumed that3 are set by cluster numbers k, cluster process is as follows:Current clustering cluster number: q=m is set;While q > k doFind out two nearest clustering cluster c of distancei*And Cj*;Merge Ci*With Cj*: Ci*=Ci*∪Cj*;For j=j*+1, j*+2 ..., q doC will be clusteredjIt is renumbered as Cj-1;End forThe jth * row and jth * for deleting distance matrix M arrange;It exports cluster and divides C={ C1, C2, C3, obtain three kinds of traffic flow situation classifications of marching into the arena, respectively correspond unimpeded state, transition state, Congestion state.
- The Tendency Prediction method 5. termination environment as claimed in claim 4 is marched into the arena, which is characterized in that it is described to data carry out cleaning with Pretreatment, the method for constructing training set and test machine with grade separation result according to traffic flow character include:Embedding processing is carried out to traffic flow data of marching into the arena according to grade separation result, obtains one-hot matrix;Trained and test data set D '={ (X is obtained according to classification results and one-hot matrixi, Ci), Xi=(xi1, xi2..., xid), Ci=(c0, c1, c2);To data set according to formulaIt is normalized, wherein diIt is traffic fluxion of marching into the arena certain period According to Max and Min respectively indicate each profile maxima and minimum value in data, xiIt is diNormalization result;Data set after normalization is divided into test set D by fixed proportion as unit of timestep1With training set D2, will train Collection is divided into the equal equal portions of m, that is, D1={ T1, T2..., Ttimestep, Ttimestep+1..., T2*timestep..., Tm*timestep, it will survey Examination collection is divided into the equal equal portions of n, that is, D2={ T1, T2..., Ttimestep, Ttimestep+1..., T2*timestep..., Tn*timestep, Middle Ti=(xi1, xi2..., xid, yi), wherein d is characterized number.
- The Tendency Prediction method 6. termination environment as claimed in claim 5 is marched into the arena, which is characterized in that described to establish traffic fluidised form of marching into the arena Gesture grade forecast model, and the method that prediction model set is arranged includes:According to the traffic flow of marching into the arena of traffic characteristic and the foundation of grade separation result based on shot and long term memory network (LSTM) of marching into the arena Jam level prediction model, every time using the data of a time series as the input of model, the last output of model is prediction Traffic jam level;According to the initialization sequence length of candidate LSTM model, the LSTM number of plies, cell number, Dropout probability and learning rate a, It is constant according to fixed other parameters, change the method setting prediction model set M={ m of a certain preset parameter1, m2..., mk, Wherein, m1For the 1st candidate family, m2For the 2nd candidate family, mkFor k-th of candidate family;To each of M model mi, initialize network weightWherein W is all connections adjacent two The weight of layer, b are the bias terms of each layer, and N (0,1) is standardized normal distribution.
- The Tendency Prediction method 7. termination environment as claimed in claim 6 is marched into the arena, which is characterized in that in the Selection Model set Optimal models, and include: by the method that selected optimal models are verified on actual test collectionA model m is taken out from prediction model set Mj, and simultaneously initialization model m is setjNetwork weightTo training dataNetwork parameter θ is successively trained using greedy algorithm(j), j=1,2 ..., mi, i.e., First with the first layer of input data training depth noise reduction autoencoder network, the parameter W of the first layer network is generated(1)And b(1);So Input by the output of first layer as the second layer afterwards continues training and obtains the parameter W of the second layer(2)And b(2);Finally to below Each layer is successively trained the mode that the output of front layer is inputted as next layer, using same strategy for above-mentioned training side Formula can fix other each layer parameters and remain unchanged, utilize the layer-by-layer training pattern of greedy algorithm when training each layer parameter Network parameter, network parameter, parameter value after obtaining model learning are adjusted by backpropagation;The loss objective function of model uses cross entropy loss function Loss=- ∑ yilnai, wherein yiWhat is represented is true value, And aiWhat is represented is the predicted value of model;When being trained according to loss objective function, need to define the size of learning rate, learning rate a formula are as follows:Wherein θ represents the value of the weight of network, and j indicates model in trained iterative steps;Regularization constraint is provided with to learning parameter, the objective function after regularization constraint are as follows:Wherein θ={ W, b };W is the weight of all connection adjacent two layers;B is the bias term of each layer;λ is then used for metric data weight Weight between structure degree and regularization constraint, the concept input matrix of x representative model;If empiric riskIt is not converged, then Model Weight is updated according to learning rate a iterationUntil empiric riskBecome In convergence, and the termination environment jam level predicted and accuracy rate Accuracy are tested on test set.By test set The model that inputs to of data, obtain actual test accuracy rateEnable M=M { mi, ifThen select Taking the highest model of Accuracy is optimal models, otherwise continues Selection Model training.
- The Tendency Prediction method 8. termination environment as claimed in claim 7 is marched into the arena, which is characterized in that the parameter by optimal models It is fixed, and to the method that traffic jam level is predicted, that is,The parameter of optimal models is fixed;Certain timestep item is marched into the arena traffic flow data (Tj, Tj+1..., Tj+timestep), wherein Ti=(xi1, xi2..., xid, yi) optimal models are inputed to, the traffic flow for obtaining the prediction of next time granularity is marched into the arena situation ci, wherein (0,1,2) i ∈.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910162007.6A CN110263962A (en) | 2019-03-04 | 2019-03-04 | Termination environment is marched into the arena Tendency Prediction method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910162007.6A CN110263962A (en) | 2019-03-04 | 2019-03-04 | Termination environment is marched into the arena Tendency Prediction method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110263962A true CN110263962A (en) | 2019-09-20 |
Family
ID=67911737
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910162007.6A Pending CN110263962A (en) | 2019-03-04 | 2019-03-04 | Termination environment is marched into the arena Tendency Prediction method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110263962A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113242226A (en) * | 2021-05-05 | 2021-08-10 | 航天云网云制造科技(浙江)有限公司 | Big data-based intelligent network security situation prediction method |
CN113255342A (en) * | 2021-06-11 | 2021-08-13 | 云南大学 | Method and system for identifying product name of 5G mobile service |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105389980A (en) * | 2015-11-09 | 2016-03-09 | 上海交通大学 | Short-time traffic flow prediction method based on long-time and short-time memory recurrent neural network |
CN109242140A (en) * | 2018-07-24 | 2019-01-18 | 浙江工业大学 | A kind of traffic flow forecasting method based on LSTM_Attention network |
-
2019
- 2019-03-04 CN CN201910162007.6A patent/CN110263962A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105389980A (en) * | 2015-11-09 | 2016-03-09 | 上海交通大学 | Short-time traffic flow prediction method based on long-time and short-time memory recurrent neural network |
CN109242140A (en) * | 2018-07-24 | 2019-01-18 | 浙江工业大学 | A kind of traffic flow forecasting method based on LSTM_Attention network |
Non-Patent Citations (1)
Title |
---|
袁立罡: "终端区动态交通特征与运行态势研究", 《中国博士学位论文全文数据库 (工程科技Ⅱ辑)》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113242226A (en) * | 2021-05-05 | 2021-08-10 | 航天云网云制造科技(浙江)有限公司 | Big data-based intelligent network security situation prediction method |
CN113255342A (en) * | 2021-06-11 | 2021-08-13 | 云南大学 | Method and system for identifying product name of 5G mobile service |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108492555A (en) | A kind of city road net traffic state evaluation method and device | |
CN109508812A (en) | A kind of aircraft Trajectory Prediction method based on profound memory network | |
CN111414991B (en) | Meteorological frontal surface automatic identification method based on multiple regression | |
CN108062561A (en) | A kind of short time data stream Forecasting Methodology based on long memory network model in short-term | |
CN108375808A (en) | Dense fog forecasting procedures of the NRIET based on machine learning | |
CN109255790A (en) | A kind of automatic image marking method of Weakly supervised semantic segmentation | |
CN110176141B (en) | Traffic cell division method and system based on POI and traffic characteristics | |
CN108647470A (en) | A kind of localization method at the beginning of based on the leakage loss with depth belief network is clustered | |
CN110503833A (en) | A kind of Entrance ramp inter-linked controlling method based on depth residual error network model | |
CN106781489A (en) | A kind of road network trend prediction method based on recurrent neural network | |
CN106504535B (en) | A kind of trip distribution modeling method of combination Gravity Models and Fratar models | |
CN106372748A (en) | Hard-rock tunnel boring machine boring efficiency prediction method | |
CN110164129B (en) | Single-intersection multi-lane traffic flow prediction method based on GERNN | |
CN103455578A (en) | Association rule and bi-clustering-based airline customer data mining method | |
CN110263962A (en) | Termination environment is marched into the arena Tendency Prediction method | |
CN107991876A (en) | Aero-engine condition monitoring data creation method based on production confrontation network | |
CN111553394A (en) | Reservoir water level prediction method based on cyclic neural network and attention mechanism | |
CN109993225A (en) | A kind of airspace complexity classification method and device based on unsupervised learning | |
CN101908213A (en) | SAR image change detection method based on quantum-inspired immune clone | |
CN110210648A (en) | Control zone strategy method for predicting based on grey shot and long term memory network | |
CN109376763A (en) | Sample classification method, system and medium based on multisample reasoning neural network | |
Du et al. | Traffic events oriented dynamic traffic assignment model for expressway network: a network flow approach | |
CN113076686A (en) | Aircraft trajectory prediction method based on social long-term and short-term memory network | |
CN110287995A (en) | The multiple features learning network model method of whole day overhead transport jam situation classification | |
CN107092988A (en) | Method for predicting station-parking time of bus on special lane |
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: 20190920 |