FTG in driving system
FTG in driving system
This part introduces how to design the FTG model and howto construct the storage and related data structure of the frequenttrajectory model. We obtain frequent pattern substrings of taxitrajectories according to the offline preprocessing results. Sincethe key of the paper is to solve the probability that can get ataxi ride at a query point inmminutes, we need to analyze andconstruct the factors that can affect the probability of the finalsolution.If an available taxi arrived at a designated location inmminutes, we must ensure the following conditions.1) To ensure that an available taxi can reach a location inmminutes, we need to analyze the speed and the distancefrom historical data.2) To ensure that other people would not get the car ridewhen it comes on its way, we need to analyze the triggerconditions in historical data, and calculate the PGTR byothers on its way in each direction.3) To ensure that query location is on the preferred trajecto-ries. Therefore, we construct the FTG.For the query timet0in the FTGFTG=(V,E),twosymbolsw(V)andw(Ei,j)are proposed.w(Vi)=ndenotesthe total number of available taxis in an aggregation pointViduring the moment[t0−m, t0].w(Ei,j)=(P,T)shows twotuples ofPandT.Prepresents the probability that a taxi isrunning fromPointi toPointjinmminutes.Trepresents thetime of using by a taxi fromPointitoPointj.It is not complicated to construct the FTG. However,FTGwill have|V|=N≈105nodes, and it is approximate to acomplete graph; hence,|E|≈|V|2≈1010. Therefore, it willproduce too enormous space cost to acceptable ifFTGismerely stored in classic adjacency matrix and adjacency table.For instance, the complexity of the adjacency matrix spaceisO(|V|2). The complexity of the adjacent table space isO(E). We consider the conception of effective trajectory whenquerying, so that the new data structure can be stratified by thetime. Hence, we design the new data structure shown in Fig. 4,which can reduce the space. We usento represent a 10 m×10 m hotspot statistics frequency in the new data structure,from the sparsest area to the densest areas by an ascendingorder. In fact, there exists a lot of the repeated frequency ofhotspot zones; hence, there are 105different values. We usettorepresent time from 0 to 24 by stratifying per minute. It wouldbe convenient for querying.https://srisivasakthitravels.com/
In Fig. 4(a),tipoints to the situation of all the dividedhotspots in the current time and divides the map of per minuteinto aboutN≈105hotspots. Hash value of the hotspot isuniquely identified. Each hotspot in data structure is expressedasnodei,j.nodei,jrepresents the hotspot on a timetiofHash=j. This data structure is called a time-stratified mul-tiple adjacent table (TMAT).Suppose thatnodei,jrepresents nodeVjin the time ofti.nodex,yrepresents nodeVyin the time oftx.AsshowninFig. 4(b),E(j, y)represents the edge relationship ofnodei,jandnodex,yin the graph model,w(Ej,y)denotes the two tuples(P,T), and the two values have been calculated. In addition,we need not storew(Ej,y).Talone because it can be calculatedby the data structure directly(w(Ej,y).T=tx−ti). We onlyneed to storew(Ej,y).P.Therefore, it can be mapped into a multiple adjacent table forhotspot trajectory strings obtained from the PTScan algorithm.Meanwhile, we update the edge information in the TMAT. Forexample, for an edgenodei,j→nodex,j, if the location doesnot move as the time goes, then it reflects the parking state ofthe taxis.The data structure is shown for each nodenodei,j,inFig. 4(c). The following is the explanation of correspondingdata ofnodei,j.Location:Hashjdenotes the specific hotspot location repre-sented bynodei,j. It is represented by the Hash valuecalculated earlier.Frequency:fdenotes the vehicle frequency ofnodei,jinmomentti.NextLocation:nodei,j+1denotes the next adjacent nodenodei,j+1ofnodei,jin momentti.NextEdge:nodex,y,Pdenotes the subsequent nodes in thefrequent pattern substring of the vehiclenodei,j(edgerelationship in Fig. 5) and corresponding weightPafterthe momentti.PreviousTime:nodea,b,Pdenotes the precursor nodes in thefrequent pattern substring of the vehiclenodei,j(edgerelationship in Fig. 5) and corresponding weightPbeforethe momentti.For TMAT, we calculated and stored two files (weekdayoffline model and weekend offline model), respectively. In ad-dition, in the subsequent query, we can read the correspondingdata structure file for a specified date.
https://goo.gl/maps/1rNVoifRFQ9snSor6
This part introduces how to design the FTG model and howto construct the storage and related data structure of the frequenttrajectory model. We obtain frequent pattern substrings of taxitrajectories according to the offline preprocessing results. Sincethe key of the paper is to solve the probability that can get ataxi ride at a query point inmminutes, we need to analyze andconstruct the factors that can affect the probability of the finalsolution.If an available taxi arrived at a designated location inmminutes, we must ensure the following conditions.1) To ensure that an available taxi can reach a location inmminutes, we need to analyze the speed and the distancefrom historical data.2) To ensure that other people would not get the car ridewhen it comes on its way, we need to analyze the triggerconditions in historical data, and calculate the PGTR byothers on its way in each direction.3) To ensure that query location is on the preferred trajecto-ries. Therefore, we construct the FTG.For the query timet0in the FTGFTG=(V,E),twosymbolsw(V)andw(Ei,j)are proposed.w(Vi)=ndenotesthe total number of available taxis in an aggregation pointViduring the moment[t0−m, t0].w(Ei,j)=(P,T)shows twotuples ofPandT.Prepresents the probability that a taxi isrunning fromPointi toPointjinmminutes.Trepresents thetime of using by a taxi fromPointitoPointj.It is not complicated to construct the FTG. However,FTGwill have|V|=N≈105nodes, and it is approximate to acomplete graph; hence,|E|≈|V|2≈1010. Therefore, it willproduce too enormous space cost to acceptable ifFTGismerely stored in classic adjacency matrix and adjacency table.For instance, the complexity of the adjacency matrix spaceisO(|V|2). The complexity of the adjacent table space isO(E). We consider the conception of effective trajectory whenquerying, so that the new data structure can be stratified by thetime. Hence, we design the new data structure shown in Fig. 4,which can reduce the space. We usento represent a 10 m×10 m hotspot statistics frequency in the new data structure,from the sparsest area to the densest areas by an ascendingorder. In fact, there exists a lot of the repeated frequency ofhotspot zones; hence, there are 105different values. We usettorepresent time from 0 to 24 by stratifying per minute. It wouldbe convenient for querying.https://srisivasakthitravels.com/
In Fig. 4(a),tipoints to the situation of all the dividedhotspots in the current time and divides the map of per minuteinto aboutN≈105hotspots. Hash value of the hotspot isuniquely identified. Each hotspot in data structure is expressedasnodei,j.nodei,jrepresents the hotspot on a timetiofHash=j. This data structure is called a time-stratified mul-tiple adjacent table (TMAT).Suppose thatnodei,jrepresents nodeVjin the time ofti.nodex,yrepresents nodeVyin the time oftx.AsshowninFig. 4(b),E(j, y)represents the edge relationship ofnodei,jandnodex,yin the graph model,w(Ej,y)denotes the two tuples(P,T), and the two values have been calculated. In addition,we need not storew(Ej,y).Talone because it can be calculatedby the data structure directly(w(Ej,y).T=tx−ti). We onlyneed to storew(Ej,y).P.Therefore, it can be mapped into a multiple adjacent table forhotspot trajectory strings obtained from the PTScan algorithm.Meanwhile, we update the edge information in the TMAT. Forexample, for an edgenodei,j→nodex,j, if the location doesnot move as the time goes, then it reflects the parking state ofthe taxis.The data structure is shown for each nodenodei,j,inFig. 4(c). The following is the explanation of correspondingdata ofnodei,j.Location:Hashjdenotes the specific hotspot location repre-sented bynodei,j. It is represented by the Hash valuecalculated earlier.Frequency:fdenotes the vehicle frequency ofnodei,jinmomentti.NextLocation:nodei,j+1denotes the next adjacent nodenodei,j+1ofnodei,jin momentti.NextEdge:nodex,y,Pdenotes the subsequent nodes in thefrequent pattern substring of the vehiclenodei,j(edgerelationship in Fig. 5) and corresponding weightPafterthe momentti.PreviousTime:nodea,b,Pdenotes the precursor nodes in thefrequent pattern substring of the vehiclenodei,j(edgerelationship in Fig. 5) and corresponding weightPbeforethe momentti.For TMAT, we calculated and stored two files (weekdayoffline model and weekend offline model), respectively. In ad-dition, in the subsequent query, we can read the correspondingdata structure file for a specified date.
https://goo.gl/maps/1rNVoifRFQ9snSor6
Comments
Post a Comment