CN105760958A - Vehicle track prediction method based on Internet of vehicles - Google Patents

Abstract

The invention discloses a vehicle track prediction method based on the Internet of vehicles, which belongs to the field of track prediction, especially vehicle track prediction. A historical track database is built based on track records collected by a GPS device of a vehicle, comparative retrieval is performed in the historical track database according to the current known state information of a user to work out the probabilities of all possible states, and the state with biggest probability is a predicted next state, namely, a predicted location. The predicted location is substituted into the track database as a current state for iterative calculation, and finally, the continuous track of the vehicle in the future is acquired. The method is advantaged in that the regularity of vehicle movement can be analyzed accurately, and the possible destination at a future moment and the continuous movement track of a vehicle can be predicted accurately.

Description

A kind of track of vehicle Forecasting Methodology based on car networking

Technical field

The invention belongs to the prediction in trajectory predictions field, particularly track of vehicle.

Background technology

Positional accuracy and the hidden performance of global position system GPS reception equipment gradually step up, cost and the volume of equipment also significantly decline, GPS is widely used in vehicle monitoring and dispatching, vehicle management etc., is usually used in the registration of vehicle detailed location information in certain moment.

The regular movement of vehicle node is the primary characteristic of In-vehicle networking, the ambulant research of vehicle is also obtained the attention of people gradually and achieves certain achievement in research.Three below aspect is mainly concerned with for the ambulant research of vehicle:

The excavation of track data and process: current multiple domestic and international scientific research institutions and universities and colleges have all carried out the scientific research project obtaining mobile subscriber track data, data digging system RealityMining etc. such as the multimedia transmission system (MultiMediaTransportSystem, MMTS) of ETH development, MIT.The Cross of Britain etc. propose the motion model of a kind of registration of vehicle track, monitored area is divided into less rectangular area, the state transition probability of the distribution situation and each node of different time points of analyzing regional interior nodes sets up model, in order to describe the real motion situation of vehicle node.Kim et al. then adopts the trace tracking method of a kind of coarseness to obtain the moving characteristic of track of vehicle, they find translational speed and the dead time equal Normal Distribution of vehicle node, and moving direction non-uniform Distribution, but along the trend of road, gone out the hot spot region of stoppage of vehicle also by the position of node and the relation recognition of time.

The foundation of vehicle mobility model: network in this kind of mobile ad-hoc network at car, the mobility model of node, for describing the motion modes such as the position of node, translational speed, direction, is the basis of research self-organizing network service and application.The mobility model of early stage is mainly some relatively simple random motion models, such as random mobility model RW (RandomWaypointMobilityMode), the random mobility model ECR (ExponentialCorrelatedRadomMobilityModel) of correlation of indices, random direction mobility modeling RD (RadomDirectionMobilityModel) etc., but because of vehicle nonrandom movement, it is impossible to the real motor process describing vehicle.Subsequently, traditional RW model is improved by Hsu etc., introduces the information such as link length, spaces of vehicles, and multiple destinatioies that vehicle is moved weight in addition is to represent the preference of user.Qunwei et al., based on the investigational data of DOT, adds the information such as the working time of user, vehicle density, social activity, establishes schedule model (Agenda-basedMobilityModle).

The prediction that vehicle moves: on research vehicle mobility basis, carried out the correlational study for vehicle moving projection both at home and abroad and achieved certain achievement.UniversityofReading studies and devises a kind of vehicular traffic monitoring system VIEWS, being tracked alternately and identifying pedestrian and vehicle, also can realize the mobile identification of vehicle, tracking and track estimation under having circumstance of occlusion.Hevner etc. propose a kind of method updating location of mobile users, it is possible to the position that the mobile vehicle subsequent time of prediction may arrive.In China, the mechanism such as Beijing University, Tsing-Hua University, electronics University of Science and Technology has also carried out relevant research work.

Summary of the invention

The present invention devises a kind of track of vehicle Forecasting Methodology, it is possible to the regularity exactly vehicle moved is analyzed, and destination and continuous print motion track that vehicle future time instance is likely to arrival are made and being predicted accurately.

The track record that the present invention gathers by utilizing vehicle GPS equipment builds historical trajectory data storehouse, in historical trajectory data storehouse, contrast retrieval is carried out according to the status information that user is currently known, then obtain its all shape probability of states being likely to occur, wherein the state of maximum probability is exactly the NextState of prediction, is predicted position.Bring, as current state, the position predicted in track database iterative computation again, finally give the continuous path that vehicle is following, required by being.

Therefore a kind of track of vehicle Forecasting Methodology based on car networking of the present invention, the method method includes:

Step 1: obtain the historical track record of this vehicle, set up historical trajectory data storehouse.

In this data base, extracting street data by cartographic information, the street overall area being observed is divided into N section, and numbers, each street segments length is less than 3 kilometers；The time period being observed is divided into L segment, and during each time segment of observational record, which street segments this vehicle location is mostly in；

Step 2: obtain the current location information of this vehicle, in conjunction with historical trajectory data storehouse, it was predicted that the positional information of this vehicle subsequent time；

Step 2.1: when this vehicle of vehicle occurs in s1 position (street segments) of observation area, the historical trajectory data storehouse obtained by step 1 adopts formula u_s1,s2/ (L-1) calculate next one moment when this vehicle current time is in s1 position occur in other the probability of likely street locations, s2 represents a kind of situation in other all possible positions, wherein u_s1,s2Represent that in historical trajectory data storehouse, continuous two these vehicle locations of moment lay respectively at the number of times of s1, s2；

Step 2.2: find out the position of maximum probability from other all possible positions, it is determined that appear in this position for subsequent time, if the identical position of probability of occurrence is followed the following step and is selected predicted position:

2.2.1: judged by Markov Chain state-transition matrix, select predicted position, if still probability of occurrence same position, step 2.2.2 is proceeded to；

2.2.2: first come, first served principle: consider the sequencing that each location status occurs, select the state occurred the earliest during search, be considered as predicting the outcome；

Step 3: after doping next result, in conjunction with the position of prediction, with the historical trajectory data storehouse in step 1, the method for employing step 2 carries out the prediction of lower step；

Step 4: adopt the mode of step 3 to carry out subsequent prediction, it is thus achieved that complete vehicle prediction motion track.

Wherein, described step 3 can also adopt u_s1,s2,s3When/(L-2) calculates this vehicle continuously across s1, s2 position the next one moment occur in other the probability of likely street locations, s3 represents a kind of situation in other all possible positions, wherein u_s1,s2,s3Represent that in historical trajectory data storehouse, continuous three these vehicle locations of moment lay respectively at the number of times of s1, s2, s3；

Described step 4 adopts u_s1,s2,s3,s4When/(L-3) calculates this vehicle continuously across s1, s2, s3 position the next one moment occur in other the probability of likely street locations, s4 represents a kind of situation in other all possible positions, wherein u_s1,s2,s3,s4Represent that in historical trajectory data storehouse, continuous three these vehicle locations of moment lay respectively at the number of times of s1, s2, s3, s4；

According to step 2, step 3, step 4 method analogize, successively obtain subsequent vehicle predicted position；After moment number continuous in prediction process reaches m, number is not further added by, and the size of m sets according to street environment.

Wherein, described step 2.1 can also adopt formulaCalculate next one moment when this vehicle current time is in s1 position occur in other the probability of likely street locations, wherein vehicle in moving process through location status s₁Number of times be N (s₁), have passed through location status s₁After, NextState is s₂Number of times be N (s₂,s₁)。

Described step 3 can also adopt formulaCalculate this vehicle continuously across position s₁, s₂Time the next one moment occur in other the probability of likely street locations, in historical trajectory data storehouse, vehicle is through location status s₁, s₂Number of times be N (s₂,s₁), have passed through location status s₁, s₂After, NextState is s₃Number of times be N (s₃,s₂,s₁)；

Described step 4 adopts formulaCalculate this vehicle continuously across position s₁, s₂, s₃Time the next one moment occur in other the probability of likely street locations, in historical trajectory data storehouse, vehicle is through location status s₁, s₂, s₃Number of times be N (s₃,s₂,s₁), have passed through location status s₁, s₂, s₃After, NextState is s₄Number of times be N (s₄,s₃,s₂,s₁)；

According to step 2, step 3, step 4 method analogize, successively obtain subsequent vehicle predicted position；After moment number continuous in prediction process reaches m, number is not further added by, and the size of m sets according to street environment.

It is an advantage of the current invention that the regularity that can exactly vehicle be moved is analyzed, destination and continuous print motion track that vehicle future time instance is likely to arrival are made and being predicted accurately.

Accompanying drawing explanation

Fig. 1 is trajectory predictions method flow diagram.

Detailed description of the invention

The foundation of 1 track of vehicle basic model

Set all of vehicle and be equipped with integrating the car networked devices of radio communication and navigator fix GPS.Considering that Wireless Telecom Equipment all has certain coverage, be set as 3 kilometers-6 kilometers m communication distances, GPS also only exists the position error of several meters, it is possible to a range of position be can be considered same place.

We are divided into N number of part by being observed total region, street, Huo Yiduan street, each of which street is a unit, the street setting each unit is long less than 3 kilometers, and the street more than 3 kilometers will be not more than several parts of 3 kilometers and numbers respectively being divided into single hop.Therefore, whole tested region being regarded as a set Street, is divided into N number of street unit, namely each street unit is an element in set:

Street={s₀,s₁,...,s_N-1|s_i∩s_j=Φ }

Same carries out abstract to the time, can also regard a set Time whole observing time as, and element therein is continuous L timeslice:

Time={t₀,t₁,...,t_L-1|t_i∩t_j=Φ }

According to such division, for a certain single unit vehicle, the street cell position s at its some time point t place in observation time is a stochastic variable, and the street cell position set of continuous L timeslice in whole observation time of this vehicle just constitutes the motion track C of this this time period of vehicle_i, it is expressed as:

C_i=< s (t₁),s(t₂),...,s(t_L)>,s(t_k)∈Street,t_k∈Time

2 based on markovian trajectory predictions algorithm

2.1 build state-transition matrix

If stochastic process X{ (t), t ∈ T} satisfies condition:

(1) time set takes set of nonnegative integer T={n=0, and 1,2 ..., for each moment, state space is discrete set, is denoted as E={n=0,1,2 ....Namely X (t) is time discrete state discrete；

(2) to arbitrary positive integer l, m, k, and arbitrary nonnegative integer j₁>j₂>...>j_l(m>j_l), with corresponding stateFollowing formula is had to set up:

P{X (m+k) }=i_m+k{ X (m)=i_m,X(j_l)=j_l,...,X(j₂)=j₂,X(j₁)=j₁}

=P{X (m+k)=i_m+k| X (m)=i_m}

Claiming X{ (t), t ∈ T} is Markov Chain.

Utilize the track record street unit setting in conjunction with this patent of vehicle GPS equipment collection, build historical trajectory data storehouse, in historical trajectory data storehouse, retrieval is contrasted according to the status information that vehicle is currently known, obtain its all shape probability of states being likely to occur, wherein the state of maximum probability is exactly the NextState of prediction, is predicted position.Bring, as current state, the position predicted in track database iterative computation again, finally give the continuous path that vehicle is following.

The probability of vehicle-state constitutes state-transition matrix, and the core of algorithm sets up state-transition matrix exactly.

In Markov Chain, event has multiple possible state, and state transition probability represents the probability size transferring to other states from a state, is expressed as:

P{X(n_m+ k)=j | X (n_m)=i_m},k≥1

Wherein, n_mFor starting the moment of observation, state is i_m, and State Transferring is j after k moment.It is called a step transition probability as k=1, uses P_ij(1) represent；As k > 1 time be called K walk transition probability, represent that this event is from some state, after k timeslice arrival state j probability, be designated as P_ij(k)。

If the state space of an observed events has N number of state E being likely to occur₁,E₂,…,E_n, state-transition matrix can be built according to the transfer case between any two states:

$P^{\left(k\right)}=\left[\begin{array}{cccc}{p}_{11}^{\left(k\right)}& p_{12}^{\left(k\right)}& ...& p_{1n}^{\left(k\right)}\\ p_{21}^{\left(k\right)}& p_{22}^{\left(k\right)}& ...& p_{2n}^{\left(k\right)}\\ \begin{array}{c}.\\ .\\ .\end{array}& \begin{array}{c}.\\ .\\ .\end{array}& \begin{array}{c}\\ \\ \end{array}& \begin{array}{c}.\\ .\\ .\end{array}\\ p_{n1}^{\left(k\right)}& p_{n2}^{\left(k\right)}& ...& p_{nn}^{\left(k\right)}\end{array}\right]$

1 rank transfer matrix it is called as k=1；As k > 1 time be called K rank transfer matrix.Each elementFor from state E_iArrival state E is walked by k_jTransition probability, and when the current state of event is E_mTime, in k the moment behind, transfer to E₁,E₂,…,E_nIn any one state be all possible, and being also only the possibility in the state of k this event of moment is E₁,E₂,…,E_nIn one, thereforeSatisfy condition:

$\left\{\begin{array}{c}0\&le;p_{ij}^{\left(k\right)}\&le;1\\ \underset{j=0}{\overset{n}{\mathrm{\&Sigma;}}}{p}_{ij}^{\left(k\right)}=1\end{array}\right.,(i,j=1,2,...,n)$

The matrix of general satisfaction above formula all can be described as probability transfer matrix, and namely solve is obtain all possible transition probability of this event

2.2 prediction tracks of vehicle

Method according to 2.1 joints sets up the state-transition matrix of track of vehicle, as follows:

$MS=\left[\begin{array}{cccc}{m}_{c^1,e^1}& m_{c^1,e^2}& ...& m_{c^1,e^N}\\ m_{c^2,e^1}& m_{c^2,e^2}& ...& m_{c^2,e^N}\\ \begin{array}{c}.\\ .\\ .\end{array}& \begin{array}{c}.\\ .\\ .\end{array}& \begin{array}{c}\\ \\ \end{array}& \begin{array}{c}.\\ .\\ .\end{array}\\ m_{c^{N\&times;N},e^1}& m_{c^{N\&times;N},e^2}& ...& m_{c^{N\&times;N},e^N}\end{array}\right]$

In state-transition matrix MS, each element representation vehicle is from a location status cⁱ, transfer to another location status e by q timeslice^jTransition probability, can be calculated as follows according to its definition:

$m_{c^i,e^j}=P\{X(n+1)=e^j|X(n-q+1,n)=c^i\}$

Wherein, wherein cⁱIt is the N number of street region < s divided in basic model₀,s₁,...s_N-1> in the sequence pair of 2 positions of arbitrary continuation.

Obtain each transition probability m in matrix_c,e, owing to the track record of vehicle is made up of substantial amounts of historical trajectory data, approximate calculation can be carried out by the probability that vehicle occurs at correspondence position according to the law of large numbers.

Assume vehicle in moving process through location status cⁱNumber of times be N (cⁱ), have passed through location status cⁱAfter, NextState is e^jNumber of times be N (e^j,cⁱ), then from state cⁱTo state e^jTwo step transition probabilities can be approximately:

$m_{c^i,e^j}=P\{X(n+1)=e^j|c^i\}=\frac{N\left(e^j,c^i\right)}{N\left(c^i\right)}$

State probability transfer matrix MS is a N²The matrix of × N, its line number is N²The currently known track that individual historical position sequence pair is constituted, and arrange number N number of possible next position of expression.

For each car, after building the 2 above state-transition matrixes in rank according to its historical trajectory data, just can pass through the track predicting its future.

According to known current state cⁱNavigating in matrix MS corresponding line number, search again for all column elements of this row and compare, wherein maximum in probability is exactly predicted position e_prid=arg_X(n+1)Max{P (X (n+1)=e^j|cⁱ)}。

In the process determining predicted position, due to the randomness that vehicle moves, in fact it could happen that the situation that the transition probability of multiple location status is equal, then consider the sequencing that each location status occurs, select the state occurred the earliest during search, be considered as predicting the outcome.

The position doped is substituted in transition probability matrix again.

Claims (4)

1., based on a track of vehicle Forecasting Methodology for car networking, the method method includes:

Step 1: obtain the historical track record of this vehicle, set up historical trajectory data storehouse.

In this data base, extracting street data by cartographic information, the street overall area being observed is divided into N section, and numbers, each street segments length is less than 3 kilometers；The time period being observed is divided into L segment, and during each time segment of observational record, which street segments this vehicle location is mostly in；

Step 2: obtain the current location information of this vehicle, in conjunction with historical trajectory data storehouse, it was predicted that the positional information of this vehicle subsequent time；

Step 2.1: when this vehicle of vehicle occurs in s1 position (street segments) of observation area, the historical trajectory data storehouse obtained by step 1 adopts formula u_s1,s2/ (L-1) calculate next one moment when this vehicle current time is in s1 position occur in other the probability of likely street locations, s2 represents a kind of situation in other all possible positions, wherein u_s1,s2Represent that in historical trajectory data storehouse, continuous two these vehicle locations of moment lay respectively at the number of times of s1, s2；

Step 2.2: find out the position of maximum probability from other all possible positions, it is determined that appear in this position for subsequent time, if the identical position of probability of occurrence is followed the following step and is selected predicted position:

2.2.1: judged by Markov Chain state-transition matrix, select predicted position, if still probability of occurrence same position, step 2.2.2 is proceeded to；

2.2.2: first come, first served principle: consider the sequencing that each location status occurs, select the state occurred the earliest during search, be considered as predicting the outcome；

Step 3: after doping next result, in conjunction with the position of prediction, with the historical trajectory data storehouse in step 1, the method for employing step 2 carries out the prediction of lower step；

Step 4: adopt the mode of step 3 to carry out subsequent prediction, it is thus achieved that complete vehicle prediction motion track.

2. as claimed in claim 1 a kind of based on car networking track of vehicle Forecasting Methodology, it is characterised in that described step 3 can also adopt u_s1,s2,s3When/(L-2) calculates this vehicle continuously across s1, s2 position the next one moment occur in other the probability of likely street locations, s3 represents a kind of situation in other all possible positions, wherein u_s1,s2,s3Represent that in historical trajectory data storehouse, continuous three these vehicle locations of moment lay respectively at the number of times of s1, s2, s3；

Described step 4 adopts u_s1,s2,s3,s4When/(L-3) calculates this vehicle continuously across s1, s2, s3 position the next one moment occur in other the probability of likely street locations, s4 represents a kind of situation in other all possible positions, wherein u_s1,s2,s3,s4Represent that in historical trajectory data storehouse, continuous three these vehicle locations of moment lay respectively at the number of times of s1, s2, s3, s4；

According to step 2, step 3, step 4 method analogize, successively obtain subsequent vehicle predicted position；After moment number continuous in prediction process reaches m, number is not further added by, and the size of m sets according to street environment.

3. as claimed in claim 1 a kind of based on car networking track of vehicle Forecasting Methodology, it is characterised in that described step 2.1 can also adopt formulaCalculate next one moment when this vehicle current time is in s1 position occur in other the probability of likely street locations, wherein vehicle in moving process through location status s₁Number of times be N (s₁), have passed through location status s₁After, NextState is s₂Number of times be N (s₂,s₁)。

4. as claimed in claim 3 a kind of based on car networking track of vehicle Forecasting Methodology, it is characterised in that described step 3 can also adopt formulaCalculate this vehicle continuously across position s₁, s₂Time the next one moment occur in other the probability of likely street locations, in historical trajectory data storehouse, vehicle is through location status s₁, s₂Number of times be N (_s2,s₁), have passed through location status s₁, s₂After, NextState is s₃Number of times be N (s₃,s₂,s₁)；

Wherein said step 4 adopts formulaCalculate this vehicle continuously across position s₁, s₂, s₃Time the next one moment occur in other the probability of likely street locations, in historical trajectory data storehouse, vehicle is through location status s₁, s₂, s₃Number of times be N (s₃,s₂,s₁), have passed through location status s₁, s₂, s₃After, NextState is s₄Number of times be N (s₄,s₃,s₂,s₁)；

According to step 2, step 3, step 4 method analogize, successively obtain subsequent vehicle predicted position；After moment number continuous in prediction process reaches m, number is not further added by, and the size of m sets according to street environment.