CN107919014A - Taxi towards more carrying kilometres takes in efficiency optimization method - Google Patents
Taxi towards more carrying kilometres takes in efficiency optimization method Download PDFInfo
- Publication number
- CN107919014A CN107919014A CN201711103910.2A CN201711103910A CN107919014A CN 107919014 A CN107919014 A CN 107919014A CN 201711103910 A CN201711103910 A CN 201711103910A CN 107919014 A CN107919014 A CN 107919014A
- Authority
- CN
- China
- Prior art keywords
- taxi
- mrow
- passenger
- msub
- time
- 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
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/20—Monitoring the location of vehicles belonging to a group, e.g. fleet of vehicles, countable or determined number of vehicles
- G08G1/202—Dispatching vehicles on the basis of a location, e.g. taxi dispatching
-
- 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"
- G06Q10/047—Optimisation of routes or paths, e.g. travelling salesman problem
Landscapes
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Quality & Reliability (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Entrepreneurship & Innovation (AREA)
- Tourism & Hospitality (AREA)
- Development Economics (AREA)
- General Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Theoretical Computer Science (AREA)
- Traffic Control Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Navigation (AREA)
Abstract
It is several grids the invention discloses a kind of taxi income efficiency optimization method towards more carrying kilometres, including by taxi operation region division, while obtains the history operation data in operational area；Position the current position of taxi and operation state；The next step running route of taxi is optimized, so as to complete to lease the income efficiency optimization of car.The present invention is up to target with the income of separate unit taxi, consider the factors such as historical data, statistics probability, costs and benefits, and dynamic optimization is carried out to the driving path of taxi using markov decision process method, so as to ensure that the revenus maximization of separate unit taxi, and the method for the present invention combines historical data and many-sided influence factor, the process of modeling and parsing more comprehensively and optimizes, and method is simple and reliable.
Description
Technical field
Present invention relates particularly to a kind of taxi towards more carrying kilometres to take in efficiency optimization method.
Background technology
With the development and the improvement of people's living standards of national economy technology, taxi trade has also welcome larger hair
Exhibition.Tax services play important role in the public transportation system in city, are indispensable in urban public transport
Pith.
But the phenomenon of taxi trade generally existing is：Sub-fraction taxi, which is in, the state that passenger uses, and one
Fraction taxi driver carries out the objective state such as stop in large-scale hub region (such as railway station, bus station etc.), and most of
Taxi driver then blindness travels on road, to the passenger for having needs of hiring a car can be run into.Therefore, present taxi
Garage's industry, common problem can not exactly be quickly found traveller, so as to reduce the rate of empty ride of taxi, improve taxi
Utilization rate, and then improve taxi driver income and reduce taxi invalid energy consumption.
At present, although having there is part to be optimized for the path of taxi, mainly in next lookup
The route recommendation of journey, improves the business efficiency of taxi；The existing factor often considered of studying is not comprehensive enough, lacks to continuous
To in a period of time in multiple mileages taxi revenus maximization method, actual application value is high.
The content of the invention
It is more comprehensive it is an object of the invention to provide a kind of Consideration, and be designed for individual taxi
Taxi towards more carrying kilometres takes in efficiency optimization method.
This taxi towards more carrying kilometres provided by the invention takes in efficiency optimization method, includes the following steps：
S1. the operational area of taxi is divided into several grids, while obtains the operation number of the history in operational area
According to；
S2. the current position of taxi and operation state are positioned；
S3. according to the current position of the taxi of step S2 acquisitions and operation state, road is run to the next step of taxi
Line optimizes, so as to complete to lease the income efficiency optimization of car.
The operational area to taxi described in step S1 carries out mesh generation, specially carries out grid using following principle
Division：The road section information being sized in the operational area with taxi of the net region of division is matched.
History operation data in operational area described in step S1, including being averaged from grid i to grid j during moment t
Time, transports passenger to the average income of grid j from grid i during moment t, and when moment t has cruised time of grid i, moment t
When find the taxis quantity of passenger in grid i, by the taxis quantity of grid i during moment t, during moment t using region j as
Passengers quantity of destination etc., wherein t are arithmetic number, and i and j are natural number.
The next step running route to taxi described in step S3 optimizes, specially using following steps to hiring out
The next step running route of car optimizes：
A. the operation state of current taxi is obtained；
B. according to the operation state of the step A taxis obtained, using markov decision process method to taxi
Next step running route optimizes.
The next step running route to taxi described in step B optimizes, and is specially carried out using following rule excellent
Change：
If R1. taxi is in the state of transport passenger, using existing ripe city path optimization algorithm to going out
The next step running route hired a car optimizes；
If R2. taxi is in non-passenger carrying status and the duration of non-passenger carrying status is greater than or equal to the threshold value set,
Then assert that taxi is in non-operational regime, stops optimizing the next step running route of taxi；
If R3. taxi is in non-passenger carrying status and the duration of non-passenger carrying status is less than the threshold value set, assert
Taxi, which is in, finds passenger status, and at this time using such as minor function as object function, the next step running route of taxi is carried out
Optimization：
Max is maximizing function in formula, V*(s, a) for taxi under current slot, the action a in state s
The expected revenus that can be brought；Pfind(l) probability of passenger, V (l can be found in grid l for taxia,t+tseek
(la), 10-a) it is taxi and has taken t+tseek(la) time, and in current grid region laIn do not look for
In the case of passenger, with the recommendation desired value that (10-a) is travel direction； V(k,t+tseek(j)+tdrive(j, k), 0) represent
The desired value of passenger is found in j regions, wherein, t+tseek(j)+tdrive(j, k) represents to have found k regions to be gone in j regions
Passenger, the hunting time taken at this time and the short time consumption from j sections to k sections that will be spent；T is current
Time, tseek(la) it is the target gridding region l that cruisesaRequired time, Pdest(j, k) is from net region j to grid regions
The possibility of domain k, r (j, k) are taxi from net region j to the expected revenus of net region k, tdrive(j, k) is taxi
From net region j to net region k the time it takes；A is to vacate taken action of hiring a car, and a ∈ A, A hire a car to vacate
The action collection that can be taken；A values are 1~9 natural number, and the Next Action of 1 expression taxi is from current location
Southwester direction running, the Next Action of 2 expression taxis are to represent taxis from current location to due south direction running, 3
Next Action be from current location southeastward direction running, 4 represent taxis Next Action be from current location to
Due west direction running, 5 represent that the Next Action of taxi represents the Next Action of taxis to be stopped in current location, 6
For from current location to a just direction running, 7 represent taxis Next Action be from current location northwestwards direction running,
8 represent taxis Next Action be from current location to direct north travel, 9 represent taxis Next Action be from
Current location direction running northeastward.
The next step running route to taxi described in regular R3 optimizes, specially using dynamic programming method pair
Function is solved at present, so as to be optimized to the next step running route of taxi.
This taxi towards more carrying kilometres provided by the invention takes in efficiency optimization method, with separate unit taxi
Multiple mileages income is up to target in a period of time, consider historical data, statistics probability, costs and benefits etc. because
Element, and dynamic optimization is carried out to the driving path of taxi using markov decision process method, so as to ensure that separate unit goes out
The revenus maximization hired a car, and the method for the present invention combines the mistake of historical data and many-sided influence factor, modeling and parsing
Cheng Gengjia is comprehensive and optimizes, and method is simple and reliable.
Brief description of the drawings
Fig. 1 is the method flow diagram of the method for the present invention..
Fig. 2 be the method for the present invention embodiment in different periods taxi boarding position hot spot schematic diagram.
Fig. 3 be the method for the present invention embodiment in before ranking after 10% driver and ranking 10% driver contrast column
Figure.
Fig. 4 is the schematic diagram for the action collection that the taxi of the method for the present invention can be taken.
Fig. 5 is the simulation comparison schematic diagram of the method for the present invention and historical data.
Fig. 6 is the method for the present invention and the simulation comparison schematic diagram of historical data during the day.
Embodiment
It is as shown in Figure 1 flow chart of the method for the present invention：This taxi towards more carrying kilometres provided by the invention
Efficiency optimization method is taken in, is included the following steps：
S1. the operational area of taxi is divided into several grids, while obtains the operation number of the history in operational area
According to including from grid i to the average time of grid j during moment t, passenger is transported being averaged to grid j from grid i during moment t
Income, when moment t, have cruised time of grid i, and when moment t finds the taxis quantity of passenger in grid i, and when moment t passes through
Cross the taxis quantity of grid i, using region j as passengers quantity of destination etc. during moment t, wherein t is arithmetic number, and i and j are
Natural number；
S2. the current position of taxi and operation state are positioned；
S3. according to the current position of the taxi of step S2 acquisitions and operation state, road is run to the next step of taxi
Line optimizes, so as to complete to lease the income efficiency optimization of car；Specially the next step of taxi is transported using following steps
Walking along the street line optimizes：
A. the operation state of current taxi is obtained；
B. according to the operation state of the step A taxis obtained, using markov decision process method to taxi
Next step running route optimizes；Specially optimized using following rule：
If R1. taxi is in the state of transport passenger, using existing ripe city path optimization algorithm to going out
The next step running route hired a car optimizes；
If R2. taxi is in non-passenger carrying status and the duration of non-passenger carrying status is greater than or equal to the threshold value set,
Then assert that taxi is in non-operational regime, stops optimizing the next step running route of taxi；
If R3. taxi is in non-passenger carrying status and the duration of non-passenger carrying status is less than the threshold value set, assert
Taxi, which is in, finds passenger status, and at this time using such as minor function as object function, the next step running route of taxi is carried out
Optimization：
Max is maximizing function in formula, V*(s, a) for taxi under current slot, the action a in state s
The expected revenus that can be brought；Pfind(l) probability of passenger, V (l can be found in grid l for taxia,t+tseek
(la), 10-a) it is taxi and has taken t+tseek(la) time, and in current grid region laIn do not look for
In the case of passenger, with the recommendation desired value that (10-a) is travel direction； V(k,t+tseek(j)+tdrive(j, k), 0) represent
The desired value of passenger is found in j regions, wherein, t+tseek(j)+tdrive(j, k) represents to have found k regions to be gone in j regions
Passenger, the hunting time taken at this time and the short time consumption from j sections to k sections that will be spent；T is current
Time, tseek(la) it is the target gridding region l that cruisesaRequired time, Pdest(j, k) is from net region j to grid regions
The possibility of domain k, r (j, k) are taxi from net region j to the expected revenus of net region k, tdrive(j, k) is taxi
From net region j to net region k the time it takes；A is to vacate taken action of hiring a car, and a ∈ A, A hire a car to vacate
The action collection that can be taken；A values are 1~9 natural number, and the Next Action of 1 expression taxi is from current location
Southwester direction running, the Next Action of 2 expression taxis are to represent taxis from current location to due south direction running, 3
Next Action be from current location southeastward direction running, 4 represent taxis Next Action be from current location to
Due west direction running, 5 represent that the Next Action of taxi represents the Next Action of taxis to be stopped in current location, 6
For from current location to a just direction running, 7 represent taxis Next Action be from current location northwestwards direction running,
8 represent taxis Next Action be from current location to direct north travel, 9 represent taxis Next Action be from
Current location direction running northeastward；
Meanwhile current function can be solved using dynamic programming method, so as to be run to the next step of taxi
Route optimizes.
The method of the present invention is further described below in conjunction with a specific embodiment：
In taxi position data, the taxi operation information in whole H cities region is contained, survey region is set as
City scope, will be filtered more than the data of this scope；Then city scope is divided to the grid regions for 128 × 128
Domain, the size in each region is 200 meters × 200 meters；The setting of net region will be more favorable for analysis efficiency to data；200
The net region of meter great little still can match road section information；
Then the history operation data in operational area is obtained, including from grid i to the average time of grid j during moment t,
Passenger is transported to the average income of grid j from grid i during moment t, when moment t, has cruised time of grid i, during moment t
Found in grid i passenger taxis quantity (as shown in Figure 2 be different periods taxi boarding position hot spot schematic diagram,
Fig. 2 (a) is the taxi boarding position hot spot schematic diagram of 6~7 periods, and Fig. 2 (b) is the taxi of 9~10 periods
Boarding position hot spot schematic diagram), by the taxis quantity of grid i during moment t, passenger during moment t using region j as destination
Quantity etc., wherein t are arithmetic number, and i and j are natural number；Basic data for subsequent analysis.
Service times of all taxis is calculated first to estimate the lease time of taxi driver：The operation of taxi
Time (uses TbusTo represent) include two parts, the total carrying time (uses TdriveTo represent) and total seek the objective time and (use Tseek
To represent), it is shown below：
Tbus=Tdrive+Tseek
Amount to the summation that the carrying time is the carrying time of a taxi All Activity.Total seeks the objective time equal to all
The sum of time between chain transaction.But the calculating for seeking the objective time is stringenter, because between continuous transaction twice
Time difference is change, and some drivers may select bait between merchandising twice.According to historical data into
Row statistical analysis, can know that 90% objective time of seeking is below 25 minutes, therefore herein using 25 minutes as a time threshold
Value；Higher than the data of this threshold value, will not count into total lookup time, because this period, taxi driver may select
Rest, refuel and exchange class etc..
Take in efficiency：The taxi operation time is longer, and total income also will be higher.But the daily service time is always solid
Fixed (when small not over 24), so how in given time restriction, the income for maximizing the unit interval just seems more
Add significant.Accordingly, definition income efficiency concept (uses ErevRepresent)：The income of the unit service time of taxi driver, such as
Shown in following formula：
Wherein M represents the total income of taxi driver, and the formula for taking in efficiency can be applied to one hour, one day even one
Year.
Success and the definition of unsuccessful taxi driver：The taxi driver of day shift is subjected to ranking according to income efficiency,
Then taxi driver in the top and rearward is found；According to knowable to being analyzed historical data, about 80% taxi
The income efficiency of driver is between 0.72 to 0.80.Therefore before income efficiency ranking 10% taxi driver is defined as into
The driver of work(, unsuccessful taxi driver is defined as by after income efficiency ranking 10% taxi driver；Meanwhile also divide
Period compares the income efficiency of 10% taxi driver after 10% taxi driver and ranking before ranking (such as Fig. 3 (a) institutes
Show), driving efficiency (such as Fig. 3 (b) shown in) and average search time (such as Fig. 3 (c) is shown) compare；It is light in Fig. 3
Column column represents the driver of 10% taxi before ranking, and dark column column represents 10% taxi driver after ranking.
Taxi total service time includes the carrying time and seeks the objective time, and taxi only just has in the case of carrying
Do not taken in when taking in, and seek visitor.Therefore the income efficiency of a taxi driver depends on the 1, energy in the case of carrying
How much (carrying efficiency driving efficiency) enough earned；2nd, can taxi quickly find next multiply in empty driving
Objective (seeking objective efficiency seeking efficiency).It has selected a traveling when taxi driver's carrying that some are lacked experience
Compare slow route or their carrying efficiency and income efficiency are often reduced when entering traffic congestion section.First
Define following carrying efficiency Edrive:
The mode that best mode goes to improve income efficiency is exactly to improve carrying efficiency EdriveWhen reducing total lookup at the same time
Between.Fig. 3 (b) compare income ranking before 10% taxi driver and ranking after 10% taxi driver carrying efficiency, he
Between about there are 10% to 13% difference.It also compares 10% taxi driver and row before each period ranking at the same time
The average search time of 10% taxi driver after name, as shown in Fig. 3 (c), the results showed that 10% taxi department before ranking
Taxi driver of the machine than after ranking 10% can save for 25% to 35% lookup time.
, it is necessary to judge its current state for a taxi, so as to could be the optimization process of its next step
Foundation is provided：
If R1. taxi is in the state of transport passenger, using existing ripe city path optimization algorithm to going out
The next step running route hired a car optimizes；
If R2. taxi is in non-passenger carrying status and the duration of non-passenger carrying status is greater than or equal to the threshold value set,
Then assert that taxi is in non-operational regime, stops optimizing the next step running route of taxi；
If R3. taxi is in non-passenger carrying status and the duration of non-passenger carrying status is less than the threshold value set, assert
Taxi, which is in, finds passenger status, at this time then using the markov decision process method proposed in the present invention to taxi
Next step operating status optimizes.
Seek objective strategy clearly for optimal for the taxi income for improving each period to depend on taxi current
State and remaining time window size.In markov decision process model, the state of taxi depends on 3 ginsengs
Number, current position (i.e. the numbering in region, from 1,2 ... to 128 × 128), (which in some period of current time
Minute), taxi reaches current region from which adjacent area, is represented here using arrival direction D,
For each state, record arrival direction be in order to prevent taxi be absorbed in several regions circulation seek visitor or
In the case of the unconfined original place of person waits.8 adjacent areas are included around each region, so each region corresponding one
The direction of a arrival, therefore have the direction of 8 arrival current regions.At the same time also define a short side to
Come represent taxi just under current region passenger so direction without any arrival.Represented with 10 numerals (0-9)
, as shown in Fig. 4 (a) and Fig. 4 (b), for any one grid 5, there are 8 grids 1,2,3,4,6,7,8 in these directions in theory
5 can be reached with 9；And for any one grid 5, there are 8 grids 1,2,3,4,6,7,8 and 9 to be used as in theory next
The purpose grid at a moment.
In a model, the empty wagons of each state has 9 action that can be taken.Each action refer to taxi from work as
Forefoot area is moved to one of region in 8 adjacent regions or stops and waits in situ.It is by this procedural representation：a
∈ A,Direction is represented using this 9 numerals of 1-9, such as
Shown in Fig. 4 (b).The direction of arrow correspond to the parameter d in state.Although expression digital in d and a direction is identical be not
The same.Such as：If taxi reaches current region from left side net region, then arrival direction d is exactly 4, if continued
Turn right and seek visitor, then the route taken is exactly 6 with regard to a, has equation d=a -10 for same direction.In addition it is not reaching to
Direction (taxi gets off in current region and then starts to seek the situation of visitor) represented with 0.
In order to ensure MDP models accuracy, it is necessary to make following limitation：1st, taxi cannot enter an invalid net
Lattice region (such as：High mountain, river etc.)；2nd, taxi purpose adjacent area must have road network to be connected with current region.For every
A net region numbering i, has counted all and has sought objective record from net region i to its 8 adjacent region herein.For each
The region j adjacent with i, calculates the probability left for j from i and seek visitor, if this probability is less than some threshold value, then speculate history
Taxi seldom goes to the region to seek visitor in data, therefore algorithm will not provide in net region i and go region j to seek objective recommendation.
It is 0 that the probability of No. 1 region (upper left) and No. 3 regions (upper right) is removed in the region, therefore the region and adjacent No. 1 region and No. 3
Exist between region and seek objective probability possibility very little, since algorithm does not recommend taxi original place to stop, so No. 5 areas are gone in the region
The probability in domain (one's respective area) is also 0.
Finally also to avoid taxi driver from being circulated between several net regions and seek visitor or unrestrictedly in some area
Domain waits, and as shown in Fig. 4 (c), can so reduce taxi driver seeks objective efficiency.Therefore the method for the present invention requires taxi department
The action that machine is taken can only be that a subset of A is known as Aallowed(s), but if taxi driver is just under current region
Car, then arrival direction is 0, and the action that this when, taxi driver can take can be any direction.When a taxi
When car seeks visitor in some region and do not find passenger, he must come out current region.Taxi leaves one at the same time
The direction in region cannot form zig zag with the direction reached.Especially, if a taxi comes in t moment from d directions
To current grid region l, then he allows the action A takenAllowed (l, t, d)Should with the angle of arrival direction at -90 ° and
Between 90 °, as shown in Fig. 4 (d).Such as：One taxi comes current region from left side d directions (direction 4), then he can
The action a taken is (2,3,6,8,9).
Assuming that the state of current taxi is s=(i, t, d).The action a of taxi can be such that taxi is moved from net region i
J is moved, then taxi will be cruised in region j to find next passenger, and hypothesis has been cruised the time of whole region j
It is tseek(j) minute.Following two results so will be just produced here：
(1) taxi has been cruised tseek(j) minute has been successfully found next passenger in region j, and then taxi is just
A passenger designated destination (it is assumed that k) can be reached, the probability from j to k is Pdest(j, k).Once taxi reaches mesh
Ground k, current transaction just completes, then using tdrive(j, k) is used as the time that current transaction needs to spend, and specifically hands over
Easy income is assumed to r (j, k), and r (j, k) is exactly taxi desired income from j to k.So taxi meeting again since k
Continue start next time seek visitor.The state of taxi is transformed to s '=(k, t+tseek(j)+tdrive(j, k), 0).
(2) if taxi has been cruised t in region jseek(j) passenger is not found after minute, then taxi must be from
Open current region and go to next region searching passenger.Assuming that the behavior a=6 (→) that taxi is taken, then taxi is cruised
After state be exactly s '=(j, t+tseek(j), 4) (because taxi comes from the left side).
To sum up, an empty wagons is in any state s0=(j, t, d) (s0∈ S) there are the certainly possible a ∈ that take action
Aallowed(s0) come the adjacent area j that cruises.And there is probability 1-Pfind(j) from state s0It is converted into state s1=(j, t+tseek
(j), 10-a), this State Transferring is no any income.Also have probability P at the same timefind(j)×Pdest(j, k), (k=
1,2 ... | L |) make taxi from state s0It is converted into state s2=(k, t+tseek(j)+tdrive(j, k), 0), and this shape
State changes the income for having r (j, k) members.
The target of model is exactly the income for the taxi driver for maximizing current slot.Model has a series of termination shape
State, that is, the end of time is up some period, i.e., when the 60th minute.Once reach this time, then go out
Hire a car to take any action again.Start within first minute in next period, and continue to carry out seeking visitor according to recommendation.
For each action a, with V* (s, a) represents current slot, in state s, phase that action a can be brought
Hope income.V (s) represents the maximum return in the state s remaining times that can be obtained.This mistake is stated using equation below
Journey：
Wherein s=(l, t, d) is a state, and a is taxi from l to laThe action taken；Record above-mentioned formula into
Vertical all path process, you can for optimizing the income of taxi.
Max is maximizing function in formula, V*(s, a) for taxi under current slot, the action a in state s
The expected revenus that can be brought；Pfind(l) probability of passenger, V (l can be found in grid l for taxia,t+tseek
(la), 10-a) it is taxi and has taken t+tseek(la) time, and in current grid region laIn do not look for
In the case of passenger, with the recommendation desired value that (10-a) is travel direction；V(k,t+tseek(j)+tdrive(j, k), 0) represent
The desired value of passenger is found in j regions, wherein, t+tseek(j)+tdrive(j, k) represents to have found k regions to be gone in j regions
Passenger, the hunting time taken at this time and the short time consumption from j sections to k sections that will be spent；T is current
Time, tseek(la) it is the target gridding region l that cruisesaRequired time, Pdest(j, k) is from net region j to grid regions
The possibility of domain k, r (j, k) are taxi from net region j to the expected revenus of net region k, tdrive(j, k) is taxi
From net region j to net region k the time it takes；A is to vacate taken action of hiring a car, and a ∈ A, A hire a car to vacate
The action collection that can be taken；A values are 1~9 natural number.
The probability P of passenger can be found in grid i for taxifind(i)：In region, i finds the taxi of passenger
Quantity (including finds the taxis quantity n of passenger than on by the total taxis quantity of region ifindWith not finding passenger
Taxis quantity npassThe sum of) it is exactly the probability that taxi can find passenger in region i, such as：Certain section passes by 100
Car, wherein 50 cars have found passenger, then the probability that passenger is found in the section is exactly 50%.Due to only having in the data of use
The data got on or off the bus in region, it is therefore desirable to estimation under taxi visitor to the path passed through carrying next time.For every
Once seek visitor, calculated using existing algorithm from starting point (taxi load zones) to terminal (region of carrying next time)
Between all paths for being passed through.Then these paths are all mapped in net region, thus obtain regional each
The taxis quantity that a period passes through.The probability tables of lookup passenger, which reaches, to be shown below：
For the possibility P from net region i to net region jdest(i,j)：In order to estimate all areas of each period
Domain destination probability parameter, calculated from historical data each net region to other all areas quantity 16384 ×
16484 matrix W, parameter W in matrixijThe corresponding period is represented from net region i to the taxis quantity of region j.By WijThan
On region i total quantity of getting on the bus i.e. from region i to the probability P of region jdest。
For expected revenus r (j, k) of the taxi from net region j to net region k：Appointed by counting each period
Average income between two regions of meaning (reaching load zones from boarding area) is as the expectation income between region.
For taxi from net region j to net region k the time it takes tdrive(j,k)：Assuming that from a region
The income that can be obtained for 10% taxi driver after 10% taxi driver before ranking and ranking to other regions is
It is the same, then consider further that the average driving efficiency of 10% taxi driver after 10% taxi driver and ranking before ranking, point
The ratio for not calculating this two groups of taxi driver's total incomes and total driving time is their driving efficiency.By using driving
Strategy, just can estimate that transaction every time needs the time i.e. t spent by merchandising incomedrive(i, j)=r (i, j)/Edrive。
Finally, current function is solved using dynamic programming method, so that the next step running route to taxi
Optimize.
Finally, the present invention verifies the effect of optimization of the method for the present invention by a simulation calculating process：
For in history taxi position data, 10% taxi driver in the top and what is ranked behind 10% hire out
Under car driver after visitor, recommend most preferably to seek objective region to taxi driver based on the method for the present invention, and simulate and seek objective process, that is, go out
Hire a car under driver after visitor, according to the lower objective time, place of getting off, algorithm can recommend next region to taxi driver to seek visitor,
After taxi driver, which reaches, recommends region, using the lookup probability of the regional obtained above, to judge taxi driver
Passenger can be found in current region.If it can not find passenger, then continue to recommend taxi driver to go to next region to find
Passenger, if having found passenger, then according to the probability distribution of the destination of the passenger of car on the area in historical data, selection
One destination region is to taxi driver, while taxi driver obtains corresponding income, continues to give out after lower visitor again
The driver that hires a car recommends, and the ending of a period is reached until the time.
As described above, assuming that taxi driver has different driving efficiencies, therefore in simulation process, use ranking
Preceding 10% and ranking after 10% taxi driving efficiency, to determine that the dealing of taxi driver between the two regions needs to spend
The time T takendrive.Below figure 5 is obtained by being simulated 6000 times to 10% taxi driver after before ranking 10% and ranking respectively
Result.Fig. 5 (a)-Fig. 5 (b) illustrate before ranking the income efficiency distribution figure of 10% taxi driver and based on ranking before
The income efficiency distribution figure of the analog result of 10% driving efficiency.For before ranking 10% taxi driver, theirs is averaged
Income efficiency from 0.828 yuan it is per minute brought up to 0.89 yuan it is per minute, increase rate is 7.6%.Fig. 5 (c)-Fig. 5 (d) is illustrated
The income efficiency distribution figure of 10% taxi driver and the income based on the analog result of 10% driving efficiency after ranking after ranking
Efficiency distribution figure.Their average income efficiency from 0.758 yuan it is per minute brought up to 0.82 yuan it is per minute, increase rate is
9.8%.
All day time periods have all been done using same method and have simulated and compare result.Fig. 6 (a) compares ranking
Preceding 10% taxi driver and income efficiency based on the driving efficiency of 10% taxi driver before ranking compare.The present invention proposes
The taxi driver of method (MDP is denoted as in figure) than before ranking 10% improve 3.6% to 12%.Fig. 6 (b) is compared
10% taxi driver and the income efficiency based on the driving efficiency of 10% taxi driver after ranking compare after ranking.The present invention
Taxi driver of the algorithm of proposition than before ranking 10% improves 4.2% to 15.5%.As seen from the figure, algorithm the morning and under
The recommendation effect of some peak times at noon has more significant raising, because there is more reference datas this when, therefore studies
Obtained parameter is also just more accurate, and it is also more accurate to recommend.Fig. 6 (c) compares 10% taxi driver before ranking and is based on
The income efficiency of the driving efficiency of 10% taxi driver compares afterwards.By relatively, using the method for the present invention recommendation
Afterwards, the income efficiency of 10% taxi driver can also reach the income level of 10% taxi driver before ranking after ranking, this
More demonstrate the validity for recommending method herein.
Finally, the recommendation results of the method for the present invention and similar Maximum Net Profit (MNP) algorithm are carried out
Contrast.In MNP methods, preset time section and out-of-the-car position, can recommend 5 sections of following expected revenus maximum to taxi
Car driver goes to seek visitor, if taxi driver does not find passenger, then MNP can continue to recommend 5 in last end section
As a result taxi driver is given.
Contrasted herein using grid instead of section, under same parameter setting, in each period, all together
Sample simulates two methods 6000 times, and Fig. 6 (d) analog results show recommendation method of the recommendation method based on MDP compared with MNP each
A period is all significantly increased, and maximum increase rate is 8.4%.
Patent of the present invention obtains Science and Technology Department of Hunan Province emphasis research and development plan financial aid (project number： 2017GK2272).
Claims (6)
1. a kind of taxi towards more carrying kilometres takes in efficiency optimization method, include the following steps：
S1. the operational area of taxi is divided into several grids, while obtains the history operation data in operational area；
S2. the current position of taxi and operation state are positioned；
S3. according to the step S2 current positions of taxi obtained and operation state, to the next step running route of taxi into
Row optimization, so as to complete to lease the income efficiency optimization of car.
2. the taxi according to claim 1 towards more carrying kilometres takes in efficiency optimization method, it is characterised in that step
The operational area to taxi described in rapid S1 carries out mesh generation, specially carries out mesh generation using following principle：Division
The road section information being sized in the operational area with taxi of net region matched.
3. the taxi according to claim 2 towards more carrying kilometres takes in efficiency optimization method, it is characterised in that step
The history operation data in operational area described in rapid S1, including from grid i to the average time of grid j, moment t during moment t
When passenger is transported to the average income of grid j from grid i, when moment t, has cruised time of grid i, in grid i during moment t
In find the taxis quantity of passenger, by the taxis quantity of grid i during moment t, using region j as destination during moment t
Passengers quantity, wherein t are arithmetic number, and i and j are natural number.
4. the taxi according to claim 3 towards more carrying kilometres takes in efficiency optimization method, it is characterised in that step
The next step running route to taxi described in rapid S3 optimizes, the next step specially using following steps to taxi
Running route optimizes：
A. the operation state of current taxi is obtained；
B. according to the operation state of the step A taxis obtained, using markov decision process method to the next of taxi
Step running route optimizes.
5. the taxi according to claim 4 towards more carrying kilometres takes in efficiency optimization method, it is characterised in that step
The next step running route to taxi described in rapid B optimizes, and is specially optimized using following rule：
R1. if taxi is in the state of transport passenger, the next step running route using path optimization's algorithm to taxi
Optimize；
If R2. taxi is in non-passenger carrying status and the duration of non-passenger carrying status is greater than or equal to the threshold value set, recognize
Order a taxi and be in non-operational regime, stop optimizing the next step running route of taxi；
If R3. taxi is in non-passenger carrying status and the duration of non-passenger carrying status is less than the threshold value set, assert and hire out
Car, which is in, finds passenger status, and at this time using such as minor function as object function, the next step running route of taxi is optimized：
<mfenced open = "" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<mi>max</mi>
<mi> </mi>
<msup>
<mi>V</mi>
<mo>*</mo>
</msup>
<mrow>
<mo>(</mo>
<mi>s</mi>
<mo>,</mo>
<mi>a</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>-</mo>
<msub>
<mi>P</mi>
<mrow>
<mi>f</mi>
<mi>i</mi>
<mi>n</mi>
<mi>d</mi>
</mrow>
</msub>
<mo>(</mo>
<mi>l</mi>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mo>&times;</mo>
<mi>V</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>l</mi>
<mi>a</mi>
</msub>
<mo>,</mo>
<mi>t</mi>
<mo>+</mo>
<msub>
<mi>t</mi>
<mrow>
<mi>s</mi>
<mi>e</mi>
<mi>e</mi>
<mi>k</mi>
</mrow>
</msub>
<mo>(</mo>
<msub>
<mi>l</mi>
<mi>a</mi>
</msub>
<mo>)</mo>
<mo>,</mo>
<mn>10</mn>
<mo>-</mo>
<mi>a</mi>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mo>+</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>k</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mrow>
<mo>|</mo>
<mi>L</mi>
<mo>|</mo>
</mrow>
</munderover>
<msub>
<mi>P</mi>
<mrow>
<mi>f</mi>
<mi>i</mi>
<mi>n</mi>
<mi>d</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
<mo>&times;</mo>
<msub>
<mi>P</mi>
<mrow>
<mi>d</mi>
<mi>e</mi>
<mi>s</mi>
<mi>t</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>j</mi>
<mo>,</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
<mo>&times;</mo>
<mrow>
<mo>(</mo>
<mi>r</mi>
<mo>(</mo>
<mrow>
<mi>j</mi>
<mo>,</mo>
<mi>k</mi>
</mrow>
<mo>)</mo>
<mo>+</mo>
<mi>V</mi>
<mo>(</mo>
<mrow>
<mi>k</mi>
<mo>,</mo>
<mi>t</mi>
<mo>+</mo>
<msub>
<mi>t</mi>
<mrow>
<mi>s</mi>
<mi>e</mi>
<mi>e</mi>
<mi>k</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>j</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
<msub>
<mi>t</mi>
<mrow>
<mi>d</mi>
<mi>r</mi>
<mi>i</mi>
<mi>v</mi>
<mi>e</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mrow>
<mi>j</mi>
<mo>,</mo>
<mi>k</mi>
</mrow>
<mo>)</mo>
</mrow>
<mo>,</mo>
<mn>0</mn>
</mrow>
<mo>)</mo>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
Max is maximizing function in formula, V*(s, a) for taxi under current slot, the action a in state s being capable of band
The expected revenus come；Pfind(l) probability of passenger, V (l can be found in grid l for taxia,t+tseek(la),10-a)
For taxi and t+t is takenseek(la) time, and in current grid region laIn do not find the feelings of passenger
Under condition, with the recommendation desired value that (10-a) is travel direction；V(k,t+tseek(j)+tdrive(j, k), 0) represent to find in j regions
The desired value of passenger, wherein, t+tseek(j)+tdrive(j, k) represents to have found the passenger in k regions to be gone in j regions, at this time
The hunting time taken and the short time consumption from j sections to k sections that will be spent；T is the current time, tseek
(la) it is the target gridding region l that cruisesaRequired time, Pdest(j, k) is from net region j to the possibility of net region k
Property, r (j, k) is taxi from net region j to the expected revenus of net region k, tdrive(j, k) is taxi from net region
J is to net region k the time it takes；A is to vacate taken action of hiring a car, and a ∈ A, and A, which is vacated, to hire a car and can take
Action collection；A values are 1~9 natural number, and 1 represents that the Next Action of taxi is from current location southwester direction
Traveling, the Next Action of 2 expression taxis are to represent next row of taxis from current location to due south direction running, 3
Move as from current location southeastward direction running, 4 represent that the Next Actions of taxis is from current location to due west direction row
Sail, 5 represent the Next Action of taxi to be stopped in current location, and the Next Action of 6 expression taxis is from present bit
Put to a just direction running, 7 represent the Next Action of taxi for northwestwards direction running, 8 expressions are hired out from current location
The Next Action of car is to be travelled from current location to direct north, and the Next Action of 9 expression taxis is from current location
Direction running northeastward.
6. the taxi according to claim 5 towards more carrying kilometres takes in efficiency optimization method, it is characterised in that rule
Then the next step running route to taxi described in R3 optimizes, specially using dynamic programming method to current function into
Row solves, so as to be optimized to the next step running route of taxi.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711103910.2A CN107919014B (en) | 2017-11-10 | 2017-11-10 | Taxi running route optimization method for multiple passenger mileage |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711103910.2A CN107919014B (en) | 2017-11-10 | 2017-11-10 | Taxi running route optimization method for multiple passenger mileage |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107919014A true CN107919014A (en) | 2018-04-17 |
| CN107919014B CN107919014B (en) | 2020-10-30 |
Family
ID=61895319
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711103910.2A Active CN107919014B (en) | 2017-11-10 | 2017-11-10 | Taxi running route optimization method for multiple passenger mileage |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107919014B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109017885A (en) * | 2018-08-17 | 2018-12-18 | 湖南大学 | A kind of dynamic auto recombination car team method and system |
| CN109034455A (en) * | 2018-06-28 | 2018-12-18 | 清华大学 | With vehicle dispatching method, system, server and computer readable storage medium |
| CN109034456A (en) * | 2018-06-28 | 2018-12-18 | 清华大学 | With vehicle dispatching method, system, server and computer readable storage medium |
| CN109670703A (en) * | 2018-12-17 | 2019-04-23 | 湖南大学 | A kind of taxis optimization methods such as the passenger based on markov decision process |
| CN110332942A (en) * | 2019-04-07 | 2019-10-15 | 武汉理工大学 | A kind of zero load taxi driving path optimization method |
| CN111047159A (en) * | 2019-11-26 | 2020-04-21 | 淮阴工学院 | Passenger carrying allocation method for taxi in airport |
| CN111079008A (en) * | 2019-12-11 | 2020-04-28 | 山东师范大学 | Method and system for recommending scheme of taxi driver staying in storage pool |
| WO2020086409A1 (en) * | 2018-10-23 | 2020-04-30 | Uber Technologies, Inc. | Prediction engine for a network-based service |
| CN113962752A (en) * | 2021-12-21 | 2022-01-21 | 华南理工大学 | Electric taxi individual behavior analysis decision-making system based on multivariate information interaction |
| CN115526387A (en) * | 2022-09-16 | 2022-12-27 | 暨南大学 | Visitor-seeking path recommendation method based on reinforcement learning and dynamic pricing mechanism |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102799179A (en) * | 2012-07-06 | 2012-11-28 | 山东大学 | Mobile robot path planning algorithm based on single-chain sequential backtracking Q-learning |
| WO2014173849A1 (en) * | 2013-04-22 | 2014-10-30 | Ineo | Method and device for dynamic management of urban mobility |
| CN107038886A (en) * | 2017-05-11 | 2017-08-11 | 厦门大学 | A kind of taxi based on track data cruise path recommend method and system |
-
2017
- 2017-11-10 CN CN201711103910.2A patent/CN107919014B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102799179A (en) * | 2012-07-06 | 2012-11-28 | 山东大学 | Mobile robot path planning algorithm based on single-chain sequential backtracking Q-learning |
| WO2014173849A1 (en) * | 2013-04-22 | 2014-10-30 | Ineo | Method and device for dynamic management of urban mobility |
| CN107038886A (en) * | 2017-05-11 | 2017-08-11 | 厦门大学 | A kind of taxi based on track data cruise path recommend method and system |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109034455A (en) * | 2018-06-28 | 2018-12-18 | 清华大学 | With vehicle dispatching method, system, server and computer readable storage medium |
| CN109034456A (en) * | 2018-06-28 | 2018-12-18 | 清华大学 | With vehicle dispatching method, system, server and computer readable storage medium |
| CN109034455B (en) * | 2018-06-28 | 2024-06-07 | 清华大学 | Method, system, server and computer readable storage medium for scheduling vehicles |
| CN109017885A (en) * | 2018-08-17 | 2018-12-18 | 湖南大学 | A kind of dynamic auto recombination car team method and system |
| WO2020086409A1 (en) * | 2018-10-23 | 2020-04-30 | Uber Technologies, Inc. | Prediction engine for a network-based service |
| US11082529B2 (en) * | 2018-10-23 | 2021-08-03 | Uber Technologies, Inc. | Prediction engine for a network-based service |
| CN109670703B (en) * | 2018-12-17 | 2021-10-26 | 湖南大学 | Passenger taxi and other taxi optimization method based on Markov decision process |
| CN109670703A (en) * | 2018-12-17 | 2019-04-23 | 湖南大学 | A kind of taxis optimization methods such as the passenger based on markov decision process |
| CN110332942A (en) * | 2019-04-07 | 2019-10-15 | 武汉理工大学 | A kind of zero load taxi driving path optimization method |
| CN111047159A (en) * | 2019-11-26 | 2020-04-21 | 淮阴工学院 | Passenger carrying allocation method for taxi in airport |
| CN111047159B (en) * | 2019-11-26 | 2023-12-05 | 淮阴工学院 | Airport taxi passenger carrying distribution method |
| CN111079008A (en) * | 2019-12-11 | 2020-04-28 | 山东师范大学 | Method and system for recommending scheme of taxi driver staying in storage pool |
| CN113962752A (en) * | 2021-12-21 | 2022-01-21 | 华南理工大学 | Electric taxi individual behavior analysis decision-making system based on multivariate information interaction |
| CN115526387A (en) * | 2022-09-16 | 2022-12-27 | 暨南大学 | Visitor-seeking path recommendation method based on reinforcement learning and dynamic pricing mechanism |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107919014B (en) | 2020-10-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107919014A (en) | Taxi towards more carrying kilometres takes in efficiency optimization method | |
| WO2021212866A1 (en) | Vehicle travel volume prediction model construction method, and prediction method and system | |
| CN102044149B (en) | City bus operation coordinating method and device based on time variant passenger flows | |
| CN108133302B (en) | Public bicycle potential demand prediction method based on big data | |
| CN106485359A (en) | A kind of urban track traffic section passenger flow estimation method based on train schedule | |
| Zong et al. | Understanding parking decisions with structural equation modeling | |
| Choudhury et al. | Modelling preferences for smart modes and services: a case study in Lisbon | |
| CN103984994B (en) | Method for predicting urban rail transit passenger flow peak duration | |
| Sharma et al. | Park-and-ride lot choice model using random utility maximization and random regret minimization | |
| CN110836675B (en) | Decision tree-based automatic driving search decision method | |
| CN106127357A (en) | A kind of customization public transport automatic routing system based on reservation data and method | |
| CN107563540A (en) | A kind of public transport in short-term based on random forest is got on the bus the Forecasting Methodology of the volume of the flow of passengers | |
| CN112784000B (en) | Passenger searching method based on taxi track data | |
| CN112309119B (en) | Urban traffic system capacity analysis optimization method | |
| CN106022514A (en) | Public electric bicycle leasing point address-selecting method based on trip chain | |
| Shao et al. | Logistic regression models for the nearest train station choice: A comparison of captive and non-captive stations | |
| CN111291321A (en) | Urban traffic model construction method based on big data | |
| CN106021902A (en) | Grading method used for urban rail transit passenger flow congestion index | |
| CN102708283A (en) | Modeling and simulating method for travelling times of residents based on travelling chain | |
| Xueyu et al. | Research on the Bi-level programming model for ticket fare pricing of urban rail transit based on particle swarm optimization algorithm | |
| CN116129651A (en) | Traffic capacity calculation method based on resident trip behavior selection | |
| Li et al. | Exploring interactive and nonlinear effects of key factors on intercity travel mode choice using XGBoost | |
| Lei et al. | Research on multi-objective bus route planning model based on taxi GPS data | |
| Wu et al. | Optimized two-directional phased development of a rail transit line | |
| Altaf | Investigating the factors influencing the use of public bike sharing schemes for the last mile travel in Belgium |
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 |