CN109886126A - A kind of region traffic density estimation method based on dynamic sampling mechanism and RBF neural - Google Patents
A kind of region traffic density estimation method based on dynamic sampling mechanism and RBF neural Download PDFInfo
- Publication number
- CN109886126A CN109886126A CN201910064378.0A CN201910064378A CN109886126A CN 109886126 A CN109886126 A CN 109886126A CN 201910064378 A CN201910064378 A CN 201910064378A CN 109886126 A CN109886126 A CN 109886126A
- Authority
- CN
- China
- Prior art keywords
- traffic density
- sampling
- target area
- rbf neural
- activation primitive
- 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.)
- Granted
Links
Abstract
The region traffic density estimation method based on dynamic sampling mechanism and RBF neural that the invention discloses a kind of, according to the sample information of traffic density in target area, Primary Construction traffic density database；The estimated value for comparing multiple traffic density defines one group of activation primitive and establishes the estimation model based on RBF neural using the sampled data of storage as the input variable of RBF neural；Kalman filtering algorithm is applied to based in RBF neural algorithm for estimating；According to the degree of correlation function of traffic density in the weight coefficient of each neuroid and target area, the traffic density at any point in target area is estimated；Finally by judging whether estimated result meets mission requirements, the dynamic estimation to traffic density in target area is realized.The present invention has faster estimated efficiency, lower computational load and higher estimated accuracy, can estimate the time-varying traffic density in target area effectively in real time, have a wide range of applications space and usage range.
Description
Technical field
The invention belongs to spatial information distribution technique fields, and in particular to one kind is based on dynamic sampling mechanism and RBF nerve
The region traffic density estimation method of network.
Background technique
With the high speed development of social economy, car ownership rapid growth year by year.Bring traffic jam issue therewith
Also more serious.Since traffic density information is to react the most direct index of road traffic situation, alleviate traffic jam issue
Premise be traffic density information in master goal region.Grasping in region after traffic density information, can effectively and
When the traffic in target area is managed and is dredged, alleviate congestion, avoid traffic accident generation.
Currently, people generally use in the conventional estimateds method prediction such as interative least square method or Gauss estimation target area
Traffic density.The estimating speed of these conventional estimated algorithms is relatively slow, can not effectively update the change of traffic density information
Change, and its estimated accuracy is generally lower.In face of problems, conventional method improves precision by the way that a large amount of sampled point is arranged.
But need to expend a large amount of data sampling cost and operation cost again in this way.And these algorithms can not often eliminate the shadow of noise
It rings, and then affects the accuracy of estimated result.
Further, since the traffic density information change in target area is usually all nonlinear, but in traditional estimation
In algorithm sample mode, the sampling configuration of fixed intervals is generallyd use.Such sampling configuration is slow in traffic density information change
The computing resource of estimating system can be wasted in the slow period, improve operation cost.And it is faster to traffic density information change
Situation, estimated result with follow effect poor, tend not to reflect real-time traffic behavior in time.Therefore, how to improve
The problem of efficiency of estimating system is urgent need to resolve.
Summary of the invention
In view of the above-mentioned deficiencies in the prior art, the technical problem to be solved by the present invention is that providing a kind of based on dynamic
The region traffic density estimation method of sampling mechanism and RBF neural considers sampling noise and traffic density time-varying at the same time
In the case where, efficient, the accurate estimation of traffic density in target area is effectively realized, and can reasonably distribute estimating system
Workload improves estimated efficiency.
The invention adopts the following technical scheme:
A kind of region traffic density estimation method based on dynamic sampling mechanism and RBF neural, according to target area
The sample information of interior traffic density, Primary Construction traffic density database, and initialize relevant parameter；Compare multiple traffic density
Estimated value, dynamic adjust the sampling interval, update sampled data；Using the sampled data of storage as the input of RBF neural
Variable defines one group of activation primitive and establishes the estimation model based on RBF neural；For the noise shadow in sampled data
It rings, Kalman filtering algorithm is applied to update while filtering out sampling noise based in RBF neural algorithm for estimating
The weight of RBF neural；According to the degree of correlation of traffic density in the weight coefficient of each neuroid and target area
Function estimates the traffic density at any point in target area；Finally by the error e compared between estimated value and sampled value
(tk), judge whether estimated result meets mission requirements, realizes the dynamic estimation to traffic density in target area.
Specifically, n high-definition camera is arranged in target area, and using the traffic density in target area as target, meter
Calculate the traffic density at each camera shooting point, the traffic density information at real-time collecting current time all shooting points, if
The location information of all camera shooting points is P=[p1,…,pn]T, the hits of traffic density in Primary Construction target area
According to libraryIt is as follows:
Wherein, Y (P, tk) indicate traffic density estimating system sampling set, calculate it is as follows:
Y(P,tk)=[y1(tk),…,yn(tk)]T
Wherein,Indicate the i-th camera in tkThe sampled value at moment.
Specifically, enablingFor the initial samples information of traffic density in target area, pass through
The estimated value at current time and the estimated value of its previous moment are compared, the Pearson came correlation of traffic density estimated value twice is acquired
CoefficientIt is as follows:
Wherein, tkWith tk-1Respectively indicate current time and the previous moment of estimating system；For traffic density estimation
ValueIn j-th of element,For tkThe average value of moment traffic density estimated value,Codomain be [0,1].
Further, if being divided into Δ t between the double sampling of sample devicess, the upper limit is expressed as with lower limitWithc, will
Pearson came relative coefficient is divided into three sections, i.e. [0≤c1< c2≤ 1], then sampling interval Δ tsDynamic adjustment mechanism indicate such as
Under:
Wherein, k is the timing serial number in sampling period, and θ is the minimum step for increasing the sampling interval.
Specifically, the activation primitive defined in RBF neural is as follows:
Wherein, i=1 ..., m, ψi(q) i-th of activation primitive in activation primitive Vector Groups is indicated；M is activation primitive vector
Group Ψ (q)=[ψ1(q),…,ψi(q),…,ψm(q)]TDimension；CiIndicate the coordinate bit of each cluster centre in target area
It sets；σiFor activation primitive ψi(q) width vector；Q ∈ Q indicates any point in target area.
Further, the step of obtaining the related coefficient and RBF neural relevant parameter in activation primitive is as follows:
S301, the precision ξ and maximum number of iterations N that neural network is set；
S302, the position coordinates C that each cluster centre is calculated by K-MEANS algorithmi；
S303, width vector σ is calculated using KNN algorithmi；
Hidden layer is to the influence degree of input vector by σiValue determine, specifically acquired by following formula:
Wherein, ChThe center position coordinates of sample are clustered for h-th, h is cluster sum；The vector form Ψ of activation primitive
(q) it is expressed as follows:
Wherein,For m dimension vector space, then in target area traffic density sampling set Y (P, tk) be expressed as
Y(P,tk)=ΨT(P)ω(tk)
Wherein,For one group of ideal weight coefficient；Activation primitive Vector Groups exist
Activation primitive matrix Ψ (P) on sampling location is expressed as follows:
Wherein,Indicate the vector space of m × n dimension.
Specifically, the step of being iterated update to weight each in neural network according to Kalman filtering algorithm is as follows:
The initial value of correlation matrix in S401, setting Kalman filtering algorithm, including error co-variance matrix Pk(0) and noise
Covariance matrix R (0)；
S402, the activation primitive matrix Ψ (P) according to activation primitive in neural network at sampling location calculate Kalman
Gain matrix Kp(tk) it is as follows:
Kp(tk)=Pk(tk-1)ΨT(P)/(Ψ(P)Pk(tk-1)ΨT(P)+R(tk))
Wherein, Kp(tk) it is tkMoment obtained kalman gain matrix, Pk(tk-1) be -1 moment of kth evaluated error
Covariance matrix, R (tk) it is sampling noise covariance matrix；
S403, set the weight coefficient of initial time asCalculate the weight coefficient of subsequent time activation primitive
Wherein,WithThe weight coefficient of respectively subsequent time and current time is estimated
Evaluation；
S404, evaluated error covariance matrix P is updatedk(tk) it is as follows:
Pk(tk)=Pk(tk-1)-Kp(tk)Ψ(P)Pk(tk-1)。
Specifically, with current time traffic density q point estimated valueAs traffic density, calculate as follows:
Wherein, c (q, P) indicates the space correlation degree in region between arbitrary point and sampling location, and c (P, P) expression is adopted
The mutual space correlation degree in sample position.
Further, i-th of element c in c (q, P)iWith i-th, j element c in c (P, P)ijIt calculates as follows:
Wherein, σrFor the fixed gain parameter of the function, σsFor spatial sensitivity coefficient, σtFor time sensitivity coefficient, δij
For Kronecker function, tkiWith tkjRespectively indicate the sampling time of ith sample value Yu j-th of sampled value.
Specifically, judging e (tk) whether meet default precision, if error e (tk) meet default precision, then terminate estimation and calculates
Method；Otherwise sampled data and related data are updated, return reinitializes relevant parameter, compares the estimation of multiple traffic density
Value, dynamic adjust the sampling interval, update sampled data；Evaluated error e (tk) calculate it is as follows:
Wherein,For the traffic density estimated value at current time sampling location.
Compared with prior art, the present invention at least has the advantages that
A kind of region traffic density estimation method based on dynamic sampling mechanism and RBF neural of the present invention, by adopting
The sampled value in target area is mapped to higher dimensional space by lower dimensional space with RBF neural, by script linearly inseparable
Traffic density function is converted into high-dimensional linearly separable function；The connection weight for meeting required precision is found out by successive ignition again
Value.According to the relevance function of the connection weight and the traffic density of target area, real-time estimation goes out the vehicle in target area
Density.Compared with the prior art, this method can greatly improve estimated accuracy and estimating speed, and can reflect target area in real time
The variation of interior traffic density.
Further, the dynamic sampling mechanism of design, can be by judging the variation degree of traffic density estimated value, dynamic
The sampling period of sampling network is adjusted, the computational load of estimating system is more reasonably deployed, has expanded application model of the invention
It encloses, and improves the accuracy and real-time to the estimation of traffic density information.
Further, the Pearson correlation coefficient at current time and previous moment traffic density estimated value is applied to dynamic
In sampling mechanism, the change rate of vehicle density estimation value can be reacted in real time, and effectively by it is this variation be mapped to it is limited
Codomain within the scope of (Codomain be [0,1]).The change rate of traffic density estimated value is described in this way, it can will not
It with the sampling period discrimination standard under mission requirements, is unified in [0,1] range, enhances theory analysis effect of the invention.
In addition, the Pearson came relative coefficient standard set in claim 4: [0≤c1< c2It≤1], can be according to appointing under different situations
Business demand dynamically adjusts corresponding segmentation criteria, and the traffic density for allowing the invention to be applied under different accuracy demand is estimated
Situation is counted, practicability of the invention is improved.
Further, using exponential type function as the activation primitive in RBF neural.Relative to traditional with multinomial
Formula function is the neural network algorithm for estimating of activation primitive, and using exponential type function as activation primitive, linear combination exists the present invention
There is preferable continuity and slickness in two-dimensional space, and the generation of nonlinear terms can be effectively prevented from.In addition, of the invention
The centre coordinate position C of each cluster in target area is introduced in activation primitiveiAnd width vector σi, increase activation letter
Sensibility of the number in target area, can effectively improve the accuracy of algorithm for estimating.
Further, Kalman filtering algorithm is applied in RBF neural estimation model, significantly reduces sampling
Influence of the noise to estimated result.Relative to traditional Learning Algorithm, which effectively can inhibit environment to make an uproar
Interference of the sound to estimated result, and then improve the precision estimated traffic density in target area.
Further, the estimation method of traffic density consists of two parts in target area proposed by the present invention, and first
It is divided into the linear combination of activation primitive Yu its weight coefficient, error compensation of the second part between estimated result and sampled value
?.By this two linear combinations, the vehicle that the present invention can efficiently and accurately estimate any point in target area is close
Degree.In addition, the present invention passes through the side of the spatial correlation function of any two points in setting target area in error compensation item
Formula is converted to the evaluated error of traffic density on target point, improves the present invention effectively by the evaluated error at sampling location
Accuracy.
Further, the mode for taking feedback iteration, according to the relationship between evaluated error and default precision, dynamic is adjusted
The number of iterations.Such mode is taken, effectively the traffic density in the case of time-varying can be estimated, it is ensured that through the invention
Calculated traffic density can reflect the situation of change of traffic density in target area in real time, expand of the invention answer
With range, and improve the stability of traffic density estimating system in target area.
In conclusion the present invention announce based on the region traffic density of dynamic sampling mechanism and RBF neural estimate
Method adjusts the sampling period of traffic density estimating system first by the dynamic sampling mechanism of design in real time, can be effectively
The operation efficiency of the announced algorithm of the present invention is improved, and reduces the computational load of estimating system.On this basis, with exponential function
As activation primitive, RBF neural is constructed, to improve the accuracy of algorithm for estimating.In the process, using Kalman filtering
Algorithm designs the more new algorithm of weight coefficient, significantly reduces influence of the sampling noise to estimated result.Pass through weight coefficient
Estimate any in target area with the linear combination of activation primitive in conjunction with the evaluated error of traffic density estimated value and sampled value
The traffic density of a bit.Finally, comparison evaluated error and default precision adjust estimating for traffic density in a manner of iterative cycles
Meter as a result, realize real-time, the accurate estimation to traffic density in target area in turn.Relative to traditional traffic density estimation side
Method, the present invention combine Kalman filtering algorithm with RBF neural, the traffic density estimation being able to solve under influence of noise
Problem；In view of dynamic sampling mechanism proposed by the present invention, which has faster estimated efficiency, lower computational load
And higher estimated accuracy.Further, since the present invention by the way of iterative cycles, can estimate target area effectively in real time
Time-varying traffic density in domain, has a wide range of applications space and usage range.
Below by drawings and examples, technical scheme of the present invention will be described in further detail.
Detailed description of the invention
Fig. 1 is flow chart of the invention；
Fig. 2 is dynamic sampling mechanism flow chart；
Fig. 3 is that time-varying traffic density of the invention estimates distributed simulation figure, wherein (a) inscribes target area when being t=1s
The estimation distributed simulation figure of interior traffic density, (b) the estimation distributed simulation to inscribe traffic density in target area when t=5s
Figure, (c) the estimation distributed simulation figure to inscribe traffic density in target area when t=10s.
Specific embodiment
The region traffic density estimation method based on dynamic sampling mechanism and RBF neural that the present invention provides a kind of,
Firstly, according to the traffic density sample information of sampled point in target area, Primary Construction traffic density database, and initialize phase
Close parameter；Secondly, comparing the estimated value of multiple traffic density, dynamic adjusts the sampling interval, updates sampled data；With adopting for storage
Input variable of the sample data as RBF neural defines one group of activation primitive and designs the estimation mould based on RBF neural
Type；It is adopted using Kalman filtering algorithm as the learning algorithm of neural network filtering out for the influence of noise in sampled data
The weight of RBF neural is updated while sample noise；Then, according to the weight coefficient and target area of each neuroid
The degree of correlation function of traffic density in domain estimates the traffic density at any point in target area；Finally, by comparing estimated value
Error between sampled value, judges whether evaluated error meets mission requirements, and then realizes to traffic density in target area
Dynamic estimation.
Referring to Fig. 1, a kind of region traffic density estimation side based on dynamic sampling mechanism and RBF neural of the present invention
Method, comprising the following steps:
S1, in target area arrange n high-definition camera, using the traffic density in target area as target, pass through by
The picture shot in camera carries out the technological means such as image procossing, calculates the traffic density of each shooting point, then by wireless
The mode of communication, the traffic density information of the real-time collecting current time point.Together with the location information of sample devices, Primary Construction
The sampling database of traffic density in target area.
The sample information of traffic density is represented by Y (P, t in target areak)=[y1(tk),…,yn(tk)]T
Wherein, Y (P, tk) indicate to estimate the sampling set of network；P=[p1,…,pn]TFor the location information of sample devices,Indicate the equipment in tkThe sampled value at moment；
On this basis, the data acquisition system of traffic density algorithm for estimating is
Wherein,Indicate the sampled data set comprising sampling location information.
S2, orderFor the initial samples information of traffic density in target area.Pass through comparison
The estimated value of the estimated value at current time and its previous moment acquires the Pearson came correlation system of this traffic density estimated value twice
Number.
Specific calculation method are as follows:
Wherein,For the Pearson came relative coefficient for estimating traffic density, tkWith tk-1Respectively indicate the current of estimating system
Moment and previous moment,For traffic density estimated valueIn j-th of element,For tkMoment traffic density estimated value
Average value, byCalculation method it is found thatCodomain be [0,1].
If being divided into Δ t between the double sampling of sample devicess, the interval upper limit is expressed as with lower limitWithc, by Pierre
Inferior relative coefficient is divided into three sections, i.e. [0≤c1< c2≤ 1], then sampling interval Δ tsDynamic adjustment mechanism can be expressed as follows:
Wherein, k is Δ tsThe timing serial number in sampling period, θ are the minimum step for increasing the sampling interval.
Assuming that being divided into Δ t between the initial samples of sample devicess=5s, the interval upper limit and lower limit be expressed as 10s and
1s.Due toCodomain be that Pearson came relative coefficient according to inter-related task demand is divided into three sections, i.e., [0≤0.8 by [0,1]
< 0.9≤1].
Sampling interval Δ tsDynamic adjustment mechanism be represented by
Wherein, θ=1s is the minimum step for increasing the sampling interval.
According to the sampling interval Δ t at current times(k), the sampling instant t in an estimating system lower period is calculatedk=tk-1+Δ
ts(k), and sampled data Y (P, the t of estimating system are updatedk)。
Detailed dynamic sampling mechanism process is as shown in Figure 2.As shown in Figure 2, it is assumed that current tkThe sampling interval at moment is
5s.By comparing current time tkWith last moment tk-1Estimation traffic density, calculate current timeIfThe next sampling interval increases 1s.IfThe next sampling interval reduces 1s.If
Directly enabling the next sampling interval is its minimum value.The variation that this process passes through traffic density estimated value in comparison target area
Rate in the sampling period of dynamic adjustment estimation network, can effectively improve the computational efficiency of algorithm for estimating.
S3, the one layer of activation primitive defined in RBF neural are as follows:
Wherein, i=1 ..., m, ψi(q) i-th of activation primitive in activation primitive Vector Groups is indicated；M is activation primitive vector
Group Ψ (q)=[ψ1(q),…,ψi(q),…,ψm(q)]TDimension；CiIndicate the coordinate bit of each cluster centre in target area
It sets；σiFor activation primitive ψi(q) width vector；Q ∈ Q indicates any point in target area.
The step of relevant parameter of related coefficient and RBF neural in activation primitive, is as follows:
S301, the precision ξ and maximum number of iterations N that neural network is set；
Wherein, ξ=0.0001, N=1000；
S302, the position coordinates C that each cluster centre is calculated by K-MEANS algorithmi；
Calculate cluster centre position Ci, it is first determined ith cluster WiNumber of samples Ni, calculated further according to following formula
Ci:
Wherein, x is the sample in cluster, and h is cluster sum.
S303, width vector σ is calculated using KNN algorithmi；
Hidden layer is to the influence degree of input vector by σiValue size determine.σiIt can be acquired by following formula:
Wherein, ChThe center position coordinates of sample are clustered for h-th.
The vector form of activation primitive is represented byWhereinIndicate m dimension
Vector space.Then the sampling set of traffic density is represented by target area
Y(P,tk)=ΨT(P)ω(tk)
Wherein,For one group of ideal weight coefficient；Ψ (P) be activation primitive to
Activation primitive matrix of the amount group on sampling location, is specifically expressed as follows:
Wherein,Indicate the vector space of m × n dimension.
S4, update is iterated to weight each in neural network according to Kalman filtering algorithm, comprising the following steps:
The initial value of correlation matrix in S401, setting Kalman filtering algorithm, including error co-variance matrix Pk(0) and noise
Covariance matrix R (0)；
Wherein, Pk(0)=I, R (0)=I；
S402, the activation primitive matrix Ψ (P) according to activation primitive in neural network at sampling location calculate Kalman
Gain matrix Kp(tk)；
Specific calculation method is as follows:
Kp(tk)=Pk(tk-1)ΨT(P)/(Ψ(P)Pk(tk-1)ΨT(P)+R(tk))
Wherein, Kp(tk) it is tkMoment obtained kalman gain matrix, Pk(tk-1) be -1 moment of kth evaluated error
Covariance matrix, R (tk) it is sampling noise covariance matrix；
S403, set the weight coefficient of initial time asCalculate the weight coefficient of subsequent time activation primitive
Wherein,WithThe weight coefficient of respectively subsequent time and current time is estimated
Evaluation.
S404, evaluated error covariance matrix is updated；
Pk(tk)=Pk(tk-1)-Kp(tk)Ψ(P)Pk(tk-1)；
S5, the circular of traffic density are as follows:
Wherein,For current time traffic density q point estimated value, c (q, P) indicate region in arbitrary point with
Space correlation degree between sampling location, c (P, P) indicate the mutual space correlation degree in sampling location.
The concrete form of two formulas is as follows:
Wherein, ciFor i-th of element in c (q, P), cijFor i-th, j element in c (P, P), tkiWith tkjRespectively indicate i-th
The sampling time of a sampled value and j-th of sampled value.
σrFor the fixed gain parameter of the function, σsFor spatial sensitivity coefficient, σtFor time sensitivity coefficient, wherein σr
=5, σs=2, σt=8.
δijFor Kronecker function, form is as follows:
S6, evaluated error e (t is calculatedk), judge e (tk) whether meet default precision, if error e (tk) meet default essence
Degree, then terminate algorithm for estimating；Otherwise sampled data and related data, return step S2 are updated.
Wherein,For the traffic density estimated value at current time sampling location.
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.The present invention being described and shown in usually here in attached drawing is real
The component for applying example can be arranged and be designed by a variety of different configurations.Therefore, below to the present invention provided in the accompanying drawings
The detailed description of embodiment be not intended to limit the range of claimed invention, but be merely representative of of the invention selected
Embodiment.Based on the embodiments of the present invention, those of ordinary skill in the art are obtained without creative efforts
The every other embodiment obtained, shall fall within the protection scope of the present invention.
Referring to Fig. 3, estimation simulation result of the traffic density in the rectangular target areas of 10km × 1km.Target area
Interior traffic density distribution is indicated by the gray scale background of the figure, wherein traffic density value is by beside it corresponding to every kind of gray scale
Gray scale amplitude corresponds to table and shows, as light areas indicates that the traffic density in the region is higher.
By the simulation result it is found that the higher region of traffic density is concentrated to the left in target area.With the time
Passage, the higher region of traffic density is gradually mobile to the right side in the region in target area.And Fig. 3 b and Fig. 3 c table simultaneously
Bright, the traffic density estimation method of the invention based on dynamic sampling mechanism and RBF neural can estimate time-varying in real time
In the case of traffic density distribution.
Therefore, traffic density distribution estimation method of the invention can be effectively realized to given area traffic density information
The dynamic estimation of distribution.
The above content is merely illustrative of the invention's technical idea, and this does not limit the scope of protection of the present invention, all to press
According to technical idea proposed by the present invention, any changes made on the basis of the technical scheme each falls within claims of the present invention
Protection scope within.
Claims (10)
1. a kind of region traffic density estimation method based on dynamic sampling mechanism and RBF neural, which is characterized in that according to
The sample information of traffic density in target area, Primary Construction traffic density database, and initialize relevant parameter；Comparison is multiple
The estimated value of traffic density, dynamic adjust the sampling interval, update sampled data；Using the sampled data of storage as RBF nerve net
The input variable of network defines one group of activation primitive and establishes the estimation model based on RBF neural；For in sampled data
Kalman filtering algorithm is applied to based in RBF neural algorithm for estimating, while filtering out sampling noise by influence of noise
Update the weight of RBF neural；According to the phase of traffic density in the weight coefficient of each neuroid and target area
Pass degree function estimates the traffic density at any point in target area；Finally by the mistake compared between estimated value and sampled value
Poor e (tk), judge whether estimated result meets mission requirements, realizes the dynamic estimation to traffic density in target area.
2. the region traffic density estimation method according to claim 1 based on dynamic sampling mechanism and RBF neural,
It is characterized in that, arranging n high-definition camera in target area, using the traffic density in target area as target, calculate every
Traffic density at a camera shooting point, the traffic density information at real-time collecting current time all shooting points, if all
The location information of camera shooting point is P=[p1,…,pn]T, the sampling database of traffic density in Primary Construction target areaIt is as follows:
Wherein, Y (P, tk) indicate traffic density estimating system sampling set, calculate it is as follows:
Y(P,tk)=[y1(tk),…,yn(tk)]T
Wherein,Indicate the i-th camera in tkThe sampled value at moment.
3. the region traffic density estimation method according to claim 1 based on dynamic sampling mechanism and RBF neural,
It is characterized in that, enablingFor the initial samples information of traffic density in target area, pass through comparison
The estimated value at current time and the estimated value of its previous moment, acquire the Pearson came relative coefficient of traffic density estimated value twiceIt is as follows:
Wherein, tkWith tk-1Respectively indicate current time and the previous moment of estimating system；For traffic density estimated value
In j-th of element,For tkThe average value of moment traffic density estimated value,Codomain be [0,1].
4. the region traffic density estimation method according to claim 3 based on dynamic sampling mechanism and RBF neural,
It is characterized in that, being divided into Δ t between setting the double sampling of sample devicess, the upper limit is expressed as with lower limitWithc, by Pearson came
Relative coefficient is divided into three sections, i.e. [0≤c1< c2≤ 1], then sampling interval Δ tsDynamic adjustment mechanism be expressed as follows:
Wherein, k is the timing serial number in sampling period, and θ is the minimum step for increasing the sampling interval.
5. the region traffic density estimation method according to claim 1 based on dynamic sampling mechanism and RBF neural,
It is characterized in that, the activation primitive defined in RBF neural is as follows:
Wherein, i=1 ..., m, ψi(q) i-th of activation primitive in activation primitive Vector Groups is indicated；M is activation primitive Vector Groups Ψ
(q)=[ψ1(q),…,ψi(q),…,ψm(q)]TDimension；CiIndicate the coordinate position of each cluster centre in target area；σi
For activation primitive ψi(q) width vector；Q ∈ Q indicates any point in target area.
6. the region traffic density estimation method according to claim 5 based on dynamic sampling mechanism and RBF neural,
It is characterized in that, the step of obtaining the related coefficient and RBF neural relevant parameter in activation primitive is as follows:
S301, the precision ξ and maximum number of iterations N that neural network is set；
S302, the position coordinates C that each cluster centre is calculated by K-MEANS algorithmi；
S303, width vector σ is calculated using KNN algorithmi；
Hidden layer is to the influence degree of input vector by σiValue determine, specifically acquired by following formula:
Wherein, ChThe center position coordinates of sample are clustered for h-th, h is cluster sum；Vector form Ψ (q) table of activation primitive
Show as follows:
Wherein,For m dimension vector space, then in target area traffic density sampling set Y (P, tk) be expressed as
Y(P,tk)=ΨT(P)ω(tk)
Wherein,For one group of ideal weight coefficient；Activation primitive Vector Groups are sampling
Activation primitive matrix Ψ (P) on position is expressed as follows:
Wherein,Indicate the vector space of m × n dimension.
7. the region traffic density estimation method according to claim 1 based on dynamic sampling mechanism and RBF neural,
It is characterized in that, the step of being iterated update to weight each in neural network according to Kalman filtering algorithm is as follows:
The initial value of correlation matrix in S401, setting Kalman filtering algorithm, including error co-variance matrix Pk(0) with noise association side
Poor matrix R (0)；
S402, the activation primitive matrix Ψ (P) according to activation primitive in neural network at sampling location calculate kalman gain
Matrix Kp(tk) it is as follows:
Kp(tk)=Pk(tk-1)ΨT(P)/(Ψ(P)Pk(tk-1)ΨT(P)+R(tk))
Wherein, Kp(tk) it is tkMoment obtained kalman gain matrix, Pk(tk-1) be -1 moment of kth evaluated error association side
Poor matrix, R (tk) it is sampling noise covariance matrix；
S403, set the weight coefficient of initial time asCalculate the weight coefficient of subsequent time activation primitive
Wherein,WithThe respectively weight coefficient estimated value of subsequent time and current time；
S404, evaluated error covariance matrix P is updatedk(tk) it is as follows:
Pk(tk)=Pk(tk-1)-Kp(tk)Ψ(P)Pk(tk-1)。
8. the region traffic density estimation method according to claim 1 based on dynamic sampling mechanism and RBF neural,
It is characterized in that, with current time traffic density q point estimated valueAs traffic density, calculate as follows:
Wherein, c (q, P) indicates the space correlation degree in region between arbitrary point and sampling location, and c (P, P) indicates sample bits
Set mutual space correlation degree.
9. the region traffic density estimation method according to claim 8 based on dynamic sampling mechanism and RBF neural,
It is characterized in that, i-th of element c in c (q, P)iWith i-th, j element c in c (P, P)ijIt calculates as follows:
Wherein, σrFor the fixed gain parameter of the function, σsFor spatial sensitivity coefficient, σtFor time sensitivity coefficient, δijFor gram
Luo Neike function, tkiWith tkjRespectively indicate the sampling time of ith sample value Yu j-th of sampled value.
10. the region traffic density estimation side according to claim 1 based on dynamic sampling mechanism and RBF neural
Method, which is characterized in that judge e (tk) whether meet default precision, if error e (tk) meet default precision, then terminate estimation and calculates
Method；Otherwise sampled data and related data are updated, return reinitializes relevant parameter, compares the estimation of multiple traffic density
Value, dynamic adjust the sampling interval, update sampled data；Evaluated error e (tk) calculate it is as follows:
Wherein,For the traffic density estimated value at current time sampling location.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910064378.0A CN109886126B (en) | 2019-01-23 | 2019-01-23 | Regional vehicle density estimation method based on dynamic sampling mechanism and RBF neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910064378.0A CN109886126B (en) | 2019-01-23 | 2019-01-23 | Regional vehicle density estimation method based on dynamic sampling mechanism and RBF neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109886126A true CN109886126A (en) | 2019-06-14 |
| CN109886126B CN109886126B (en) | 2020-10-23 |
Family
ID=66926681
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910064378.0A Active CN109886126B (en) | 2019-01-23 | 2019-01-23 | Regional vehicle density estimation method based on dynamic sampling mechanism and RBF neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109886126B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111598460A (en) * | 2020-05-18 | 2020-08-28 | 武汉轻工大学 | Method, device and equipment for monitoring heavy metal content in soil and storage medium |
| CN113392817A (en) * | 2021-08-16 | 2021-09-14 | 成都信息工程大学 | Vehicle density estimation method and device based on multi-row convolutional neural network |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102819954A (en) * | 2012-08-28 | 2012-12-12 | 南京大学 | Traffic region dynamic map monitoring and predicating system |
| CN103729688A (en) * | 2013-12-18 | 2014-04-16 | 北京交通大学 | Section traffic neural network prediction method based on EMD |
| CN103956057A (en) * | 2014-01-13 | 2014-07-30 | 广西师范学院 | Traffic flow data intelligent processing system used for security and protection |
| CN104036328A (en) * | 2013-03-04 | 2014-09-10 | 横河电机株式会社 | Self-adaptive wind power prediction system and prediction method |
| CN104484997A (en) * | 2014-12-24 | 2015-04-01 | 福州大学 | Regional moving vehicle flow detection method based on video image technology |
| KR101647370B1 (en) * | 2014-11-26 | 2016-08-10 | 휴앤에스(주) | road traffic information management system for g using camera and radar |
| CN106781509A (en) * | 2017-03-06 | 2017-05-31 | 长安大学 | A kind of collaborative urban road congestion detection method based on V2V |
| CN108804721A (en) * | 2017-04-26 | 2018-11-13 | 重庆科技学院 | Based on the Fault Diagnoses of Oil Pump method adaptively without mark Kalman filter and RBF neural |
-
2019
- 2019-01-23 CN CN201910064378.0A patent/CN109886126B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102819954A (en) * | 2012-08-28 | 2012-12-12 | 南京大学 | Traffic region dynamic map monitoring and predicating system |
| CN104036328A (en) * | 2013-03-04 | 2014-09-10 | 横河电机株式会社 | Self-adaptive wind power prediction system and prediction method |
| CN103729688A (en) * | 2013-12-18 | 2014-04-16 | 北京交通大学 | Section traffic neural network prediction method based on EMD |
| CN103956057A (en) * | 2014-01-13 | 2014-07-30 | 广西师范学院 | Traffic flow data intelligent processing system used for security and protection |
| KR101647370B1 (en) * | 2014-11-26 | 2016-08-10 | 휴앤에스(주) | road traffic information management system for g using camera and radar |
| CN104484997A (en) * | 2014-12-24 | 2015-04-01 | 福州大学 | Regional moving vehicle flow detection method based on video image technology |
| CN106781509A (en) * | 2017-03-06 | 2017-05-31 | 长安大学 | A kind of collaborative urban road congestion detection method based on V2V |
| CN108804721A (en) * | 2017-04-26 | 2018-11-13 | 重庆科技学院 | Based on the Fault Diagnoses of Oil Pump method adaptively without mark Kalman filter and RBF neural |
Non-Patent Citations (1)
| Title |
|---|
| 张建: ""基于智能交通系统的城市路网短时交通预测"", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111598460A (en) * | 2020-05-18 | 2020-08-28 | 武汉轻工大学 | Method, device and equipment for monitoring heavy metal content in soil and storage medium |
| CN111598460B (en) * | 2020-05-18 | 2023-09-29 | 武汉轻工大学 | Method, device, equipment and storage medium for monitoring heavy metal content of soil |
| CN113392817A (en) * | 2021-08-16 | 2021-09-14 | 成都信息工程大学 | Vehicle density estimation method and device based on multi-row convolutional neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109886126B (en) | 2020-10-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104156524B (en) | The Aggregation Query method and system of transport data stream | |
| CN110059385B (en) | Grid dynamics scenario simulation method and terminal equipment coupled with different-speed growth | |
| Chandra et al. | Small area prediction of counts under a non-stationary spatial model | |
| CN107622085A (en) | A kind of adaptive method of trajectory clustering based on grid and density | |
| Zhang et al. | Time series prediction using Lyapunov exponents in embedding phase space | |
| Liang-Qun et al. | Intuitionistic fuzzy joint probabilistic data association filter and its application to multitarget tracking | |
| CN110428614B (en) | Traffic jam heat degree space-time prediction method based on non-negative tensor decomposition | |
| CN103425994B (en) | A kind of feature selection approach for pattern classification | |
| CN109886126A (en) | A kind of region traffic density estimation method based on dynamic sampling mechanism and RBF neural | |
| CN113033654A (en) | Indoor intrusion detection method and system based on WiFi channel state information | |
| CN106056577B (en) | SAR image change detection based on MDS-SRM Mixed cascading | |
| CN108053424A (en) | Method for tracking target, device, electronic equipment and storage medium | |
| CN110932807A (en) | Spectrum sensing method of MIMO (multiple input multiple output) system under non-Gaussian noise | |
| Franco-Villoria et al. | Bootstrap based uncertainty bands for prediction in functional kriging | |
| CN115829063A (en) | Wireless positioning differential privacy federation learning method based on dynamic privacy budget | |
| Pei et al. | The clustering algorithm based on particle swarm optimization algorithm | |
| CN112261609B (en) | Millimeter wave internet of vehicles resource allocation method and device, electronic equipment and storage medium | |
| Etiabi et al. | Federated distillation based indoor localization for IoT networks | |
| CN114882713B (en) | Signal control method, system, equipment and storage medium based on multiple scenes | |
| CN115578425A (en) | Dynamic tracking method applied to fry counter and based on unscented Kalman filtering | |
| Remm | An online calculator for spatial data and its applications | |
| Liu et al. | Evaluation of nonlinear filtering for radar data tracking | |
| CN103268586A (en) | Window fusion method based on heat diffusion theory | |
| Wang et al. | Urban traffic signal system control structural optimization based on network analysis | |
| CN113808015A (en) | Spatial interpolation method, device and equipment for meteorological parameters of power transmission line region |
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 | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20210113 Address after: 1-1-505-1, Xindu building, southwest of the intersection of Weijin road and wandezhuang street, Nankai District, Tianjin Patentee after: HUIJIAWANG (TIANJIN) TECHNOLOGY Co.,Ltd. Address before: 710064 middle section of South Second Ring Road, Beilin District, Xi'an City, Shaanxi Province Patentee before: CHANG'AN University |