Optimizing the Profitability and Quality of Service in Carshare...
Transcript of Optimizing the Profitability and Quality of Service in Carshare...
![Page 1: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/1.jpg)
Optimizing the Profitability and Quality ofService in Carshare Systems
Siqian Shen
joint work with Zhihao Chen (U of Michigan)and Mengshi Lu (Purdue)
Seminar in IE Department
Clemson University
September 23, 2016
Shen (U of Michigan) 1/33
![Page 2: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/2.jpg)
Outline
Introduction
Problem formulation
Solution algorithm
Computational results
Applications and extensions
Shen (U of Michigan) 2/33
![Page 3: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/3.jpg)
Outline
Introduction
Problem formulation
Solution algorithm
Computational results
Applications and extensions
Shen (U of Michigan) 3/33
![Page 4: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/4.jpg)
Carsharing
I Short term car rental
I Consumer benefitsI Private vehicleI No ownership responsibility
I Societal & environmental benefitsI Reduced congestion (6 cars replaced per shared vehicle)I Reduced fuel consumption (vehicle mileage ↘ 44% per
carshare user)I Reduced greenhouse gas emissions
I Becoming more popularI Over 1,000 cities worldwide have adopted carsharingI Over 1 million individuals sharing over 20,000 vehicles in the
U.S.
Shen (U of Michigan) 4/33
![Page 5: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/5.jpg)
Types of carshare
I Reservation-based vs. free-floatingI Contracted paid parking lots for reservation-based usersI Free-float parking permits for free float users
I One-way vs. round-trip rentalsI One-way more desirable for consumers
I Flexibility of using vehiclesI Potentially save on rental fees by splitting trips
I One-way trips problematic for companiesI Management complexitiesI Unbalanced demand requires redistribution of fleet
Shen (U of Michigan) 5/33
![Page 6: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/6.jpg)
Carshare design and optimization
I Strategic decisions to consider, e.g.I Implement one-way or not?I Offset cost with price differentiation?
(Zipcar charges $7.50-$8.50 per hour round-trip and $12 perhour one-way in Boston)
I Evaluate the impactI Field testing (Zipcar’s ONE>WAY beta program)I Mathematical modeling (focus of this talk)
I Optimize profitability and quality of service via models thatI Incorporate round-trip & one-way uncertain demandsI Understand customer behavior in response to decisionsI Optimize & evaluate strategic decisions
Shen (U of Michigan) 6/33
![Page 7: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/7.jpg)
Outline
Introduction
Problem formulation
Solution algorithm
Computational results
Applications and extensions
Shen (U of Michigan) 7/33
![Page 8: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/8.jpg)
Framing the problem
How does the proportion of one-way demand vs round-trip demandaffect profitability and QoS?
Assumptions:
I A set of service zones and a finite number of service periodsI Serve random one-way and round-trip rentals
I Vehicle movement and demand aggregated by zoneI Vehicles travel between zones at different periodsI Demand uncertainty with known distribution
I Cars can be relocated, to balance vehicle distributions
I Unsatisfied demand is immediately lost
Shen (U of Michigan) 8/33
![Page 9: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/9.jpg)
A two-stage stochastic programming framework
I Strategic decisions:I zone-based # of parking lots to buy (for reservation-based)I zone-based # of free-float permits to buy (for free float)I # of cars initially allocated in each zone (for both)
I Recourse decisions:I Movement of cars (from car users and also relocation)
I Objective: To minimizeI costs of allocating cars and purchasing parking lots/permitsI (uncertain) costs of operating and relocating carsI (uncertain) penalty of undesirable quality of service (related to
unsatisfied demand and denied trips)
Shen (U of Michigan) 9/33
![Page 10: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/10.jpg)
1st-stage master problem
minw,x
∑i∈I
(c loti wi +
(cffp + c loc
i
)xi
)+ Q(w , x)
s.t. (w , x) ∈ X =
w ∈ Z|I |+ , x ∈ Z|I |
+ :∑i∈I
xi ≤ S, xi ≤ wi , ∀i ∈ I
I I : Set of zones
I S : Maximum # cars
I c loti , ∀i ∈ I : cost for locating one car in zone i .
I cffp = 0 for reservation-based, and c loti = 0, ∀i ∈ I for free-float.
I w ∈ Z|I |: # parking lots to purchase in each zone
I x ∈ Z|I |: # cars to initially allocate to each zone
I Q(w , x) models the 2st-stage recourse problem by using spatial-temporalnetworks to model realized one-way & round-trip demands.
Shen (U of Michigan) 10/33
![Page 11: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/11.jpg)
Building spatial-temporal network
I Example:I Zones A, BI Time periods, 0, 1, 2, 3I Travel times between zones, A↔B = 2I nit : Zone i at time t
Type Volume Origin Destination Start EndOne-way 4 A B 0 –
Round-trip 2 B – 1 3
nA0
nB0
nA1
nB1
nA2
nB2
nA3
nB3
Figure: Spatial-temporal nodes
Shen (U of Michigan) 11/33
![Page 12: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/12.jpg)
Building spatial-temporal network
I Example:I Zones A, BI Time periods, 0, 1, 2, 3I Travel times between zones, A↔B = 2I nit : Zone i at time t
Type Volume Origin Destination Start EndOne-way 4 A B 0 –
Round-trip 2 B – 1 3
nA0
nB0
nA1
nB1
nA2
nB2
nA3
nB3
4
Figure: One-way arcs
Shen (U of Michigan) 11/33
![Page 13: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/13.jpg)
Building spatial-temporal network
I Example:I Zones A, BI Time periods, 0, 1, 2, 3I Travel times between zones, A↔B = 2I nit : Zone i at time t
Type Volume Origin Destination Start EndOne-way 4 A B 0 –
Round-trip 2 B – 1 3
nA0
nB0
nA1
nB1
nA2
nB2
nA3
nB3
2
Figure: Round-trip arcs
Shen (U of Michigan) 11/33
![Page 14: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/14.jpg)
Building spatial-temporal network
I Example:I Zones A, BI Time periods, 0, 1, 2, 3I Travel times between zones, A↔B = 2I nit : Zone i at time t
Type Volume Origin Destination Start EndOne-way 4 A B 0 –
Round-trip 2 B – 1 3
nA0
nB0
nA1
nB1
nA2
nB2
nA3
nB3
∞
Figure: Relocation arcs
Shen (U of Michigan) 11/33
![Page 15: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/15.jpg)
Building spatial-temporal network
I Example:I Zones A, BI Time periods, 0, 1, 2, 3I Travel times between zones, A↔B = 2I nit : Zone i at time t
Type Volume Origin Destination Start EndOne-way 4 A B 0 –
Round-trip 2 B – 1 3
nA0
nB0
nA1
nB1
nA2
nB2
nA3
nB3wB
wA
Figure: Idling arcs
Shen (U of Michigan) 11/33
![Page 16: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/16.jpg)
Building spatial-temporal network
I Example:I Zones A, BI Time periods, 0, 1, 2, 3I Travel times between zones, A↔B = 2I nit : Zone i at time t
Type Volume Origin Destination Start EndOne-way 4 A B 0 –
Round-trip 2 B – 1 3
nA0
nB0
nA1
nB1
nA2
nB2
nA3
nB3
4
2
Types of arcs:
One-way trip
Round trip
Relocation
Idle
Figure: Spatial-temporal network
Shen (U of Michigan) 11/33
![Page 17: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/17.jpg)
Building spatial-temporal network
I Example:I Zones A, BI Time periods, 0, 1, 2, 3I Travel times between zones, A↔B = 2I nit : Zone i at time t
Type Volume Origin Destination Start EndOne-way 4 A B 0 –
Round-trip 2 B – 1 3
nA0
nB0
nA1
nB1
nA2
nB2
nA3
nB3
4
2
Types of arcs:
One-way trip
Round trip
Relocation
Idle
Figure: Spatial-temporal network
Shen (U of Michigan) 11/33
![Page 18: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/18.jpg)
Spatial-temporal network
I Random demandI done
ijts : # of one-way rentals from zone i starting at period tand returned to zone j at period s.
I d twoits : # of round-trip rentals from zone i starting at period t
and returned to zone i at period s.
I Original cost parametersI rone: Revenue per one-way rental per periodI r two: Revenue per round-trip rental per periodI cmnt: Maintenance cost per car per periodI c rel: Relocation cost per car per period
I Network arc parameters
Type of arc Cost per unit flow fa Capacity uaOne-way arc (nit , njs) −(rone − cmnt)(s − t) done
ijts
Round-trip arc (nit , nis) −(r two − cmnt)(s − t) d twoits
Relocation arc (nit , nj,t+`ij ) (c rel + cmnt)`ij +∞Idle arc (nit , ni,t+1) c idle wi
Shen (U of Michigan) 12/33
![Page 19: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/19.jpg)
2nd-stage minimum cost flow problem
Q(w , x) =
miny1,...,y|K|
∑k∈K
pk∑a∈A
fayka + g(y1, . . . , y |K |)
s.t. yk ∈ Y (w , x , uk ) = {yk ∈ R|A|+ :
∑a∈δ+(nit )
yka −
∑a∈δ−(nit )
yka =
xi if t = 0
0 if t = 1, 2, ...,T − 1
−xi if t = T
∀i ∈ I
yka ≤ uka ∀a ∈ Aone ∪ Atwo
yka ≤ wi ∀i ∈ I , a = (nit , ni,t+1) ∈ Aidle} ∀k ∈ K
I K : Set of demand scenarios (sampled from known distribution)
I A: Set of spatial-temporal arcs
I y k : Flows on spatial-temporal network
I Y (w , x , uk): Flow balance & capacity constraints
I Q(w , x) is a large-scale linear program given fixed x and w .
Shen (U of Michigan) 13/33
![Page 20: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/20.jpg)
QoS penalty g(x , y 1, . . . , yK )
I Penalize # of unserved customersI i.e., unused capacity on rental arcs (uka − yk
a ), ∀a ∈ Aone ∪Atwo
I Risk-neutral modelI g(x , y1, . . . , yK ) =
∑k∈K pk
∑a∈Aone∪Atwo Ga(uka − yk
a )
I Risk-averse modelI Ha(w , x): denotes the unused capacity on arc a given w and xI g(x , y1, . . . , yK ) = G0CVaR1−ε
(∑a∈Aone∪Atwo Ha(w , x)
)
I G0 penalizes the expected value of the worst 100ε% scenarios.
Shen (U of Michigan) 14/33
![Page 21: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/21.jpg)
2nd-stage linear program when penalizing CVaR
Qc(w , x) = miny ,z,v≥0
∑k∈K
pk∑a∈A
fayka + G0
(v +
1
ε
∑k∈K
pkzk
)s.t. yk ∈ Y (w , x , uk), zk ≥ 0 ∀k ∈ K
zk ≥∑
a∈Aone∪Atwo
(uka − yka )− v ∀k ∈ K ,
where
I v : VaR1−ε(∑
a∈Aone∪Atwo Ha(w , x))
I zk = max{∑
a∈Aone∪Atwo(uka − yka )− v , 0}
Shen (U of Michigan) 15/33
![Page 22: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/22.jpg)
Outline
Introduction
Problem formulation
Solution algorithm
Computational results
Applications and extensions
Shen (U of Michigan) 16/33
![Page 23: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/23.jpg)
Branch-and-cut and parallelization
I There can be many 2nd-stage subproblems⇒ many spatial-temporal networks with huge sizes due to
I Fine division into zonesI Fine granularity of time
I Branch-and-cut algorithmI Branch on integer variables in the 1st stageI Use Benders decomposition to generate Benders cutsI Lift Benders cuts with mixed-integer rounding (MIR)
I Follow similar ideas in Bodur and Luedtke (2014) for atwo-stage stochastic integer program for call-center staffing
I Use parallel computing to speed up subproblem computationI Master-Worker scheme by OpenMPI 1.6I UM Flux HPC cluster, 20 cores each with 48GB RAM.
Shen (U of Michigan) 17/33
![Page 24: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/24.jpg)
Branch-and-cut and parallelization
I There can be many 2nd-stage subproblems⇒ many spatial-temporal networks with huge sizes due to
I Fine division into zonesI Fine granularity of time
I Branch-and-cut algorithmI Branch on integer variables in the 1st stageI Use Benders decomposition to generate Benders cutsI Lift Benders cuts with mixed-integer rounding (MIR)
I Follow similar ideas in Bodur and Luedtke (2014) for atwo-stage stochastic integer program for call-center staffing
I Use parallel computing to speed up subproblem computationI Master-Worker scheme by OpenMPI 1.6I UM Flux HPC cluster, 20 cores each with 48GB RAM.
Shen (U of Michigan) 17/33
![Page 25: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/25.jpg)
Branch-and-cut and parallelization
I There can be many 2nd-stage subproblems⇒ many spatial-temporal networks with huge sizes due to
I Fine division into zonesI Fine granularity of time
I Branch-and-cut algorithmI Branch on integer variables in the 1st stageI Use Benders decomposition to generate Benders cutsI Lift Benders cuts with mixed-integer rounding (MIR)
I Follow similar ideas in Bodur and Luedtke (2014) for atwo-stage stochastic integer program for call-center staffing
I Use parallel computing to speed up subproblem computationI Master-Worker scheme by OpenMPI 1.6I UM Flux HPC cluster, 20 cores each with 48GB RAM.
Shen (U of Michigan) 17/33
![Page 26: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/26.jpg)
Branch-and-cut
Master problem
Subproblems
Solution
Master problem
Subproblems
Solution Valid cut
Valid cut
Shen (U of Michigan) 18/33
![Page 27: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/27.jpg)
Branch-and-cut with MIR (Bodur & Luedtke 2014)
Master problem
Subproblems
Solution
Master problem
Subproblems
Solution
MIR cut
Cut listMIR cut
Valid cut
Valid cut
Shen (U of Michigan) 19/33
![Page 28: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/28.jpg)
Outline
Introduction
Problem formulation
Solution algorithm
Computational results
Applications and extensions
Shen (U of Michigan) 20/33
![Page 29: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/29.jpg)
Data description
I Boston-Cambridge Zipcar; Data from Oct 1 to Dec 1, 2014
I Zipcar cost parameters; c rel = $22 or $10 per hour for car relocation
I 1-hour periods, over 24 hours
I ε = 0.1 and 0.05 for the CVaR risk-averse model
I Use Java + Gurobi 6.0.3; Intel dual-core CPU with 8GB RAM.
Demand data follows Gamma distributions
Shen (U of Michigan) 21/33
![Page 30: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/30.jpg)
Computational efficiency
I Tests run for |K | = 100, 200, 500, 1000I Across all sets of results
I Branch-and-cut slower if subproblems solved in series, faster ifsolved in parallel
I # iterations fairly constant with |K |I Using MIR can improve parallel solve time
Model Series Parallel # iterations
solve time solve time
Stoch 236,207 236,207 -
Stoch-Branch 291,389 34,430 39
Stoch-MIR 160,272 27,175 37
CVaR 190,072 190,072 -
CVaR-Branch 254,038 35,301 30
CVaR-MIR 343,550 34,544 39
Table: Computational time (in second) for |K | = 1000
Shen (U of Michigan) 22/33
![Page 31: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/31.jpg)
Computational results
Question: How does the proportion of one-way demand vsround-trip demand affect profitability and QoS?
I Vary proportion of one-way tripsI 0%, 20%, 40%, 60%, 80%, and 100%
I Vary carshare systems:I Reservation based: c lot
i = $9.6 per hour in zones i = 1, 2, 5, 6, 9and c lot
i = $7.4 per hour in other zones; cffp = 0
I Free-float: cffp = $9.6 per hour; c loti = 0, ∀i = 1, . . . , 9
I Reduce relocation cost from $22 per hour to $10I U.S. average wage in 2014 ⇒ minimum hour-pay requirement
Shen (U of Michigan) 23/33
![Page 32: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/32.jpg)
Some general remarks of the results
I Similar risk neutral and CVaR penalty model results
I Similar reservation-based and free-floating system results
I Because similar costs of parking lot and free-float permit per year
I Vehicle allocation cannot be directly inferred from demand concentration
(a) Demand heatmap (b) Vehicle allocation
Figure: Demand concentration (by starting location) vs vehicle allocation
Shen (U of Michigan) 24/33
![Page 33: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/33.jpg)
Recommending Changes
I We tried to increase the revenue of one-way to improveprofitability and quality of service (QoS)
I Does not help when one-way proportion is high
I Effective approach: Decrease unit cost for car relocationc rel = $22 ⇒ c rel = $10
I Effects of changeI Slight increase in profitabilityI Major improvement in QoSI Major reduction in trips denied
I More willing to relocate cars to meet demand
Shen (U of Michigan) 25/33
![Page 34: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/34.jpg)
Change in profitability
One-way Setup One-way trip Round-trip Relocation Profit
proportion cost ($) revenue ($) revenue ($) cost ($) ($)
60% 1,770 4,804 7,475 333 10,176
80% 1,682 6,625 3,572 454 8,061
100% 1,496 8,359 0 549 6,315
Table: Profitability before change (c rel = $22)
One-way Setup One-way trip Round-trip Relocation Profit
proportion cost ($) revenue ($) revenue ($) cost ($) ($)
60% 1,749 6,761 7,491 1,986 10,517
80% 1,635 9,282 3,570 2,696 8,521
100% 1,418 11,700 0 3,389 6,894
Table: Profitability after change (c rel = $10)
Shen (U of Michigan) 26/33
![Page 35: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/35.jpg)
Change in QoS (unsatisfied demand)In general, we observe expected unserved customers ↗ as one-wayproportion ↗ for both c rel = $22 and c rel = $10.
One-way Proportion of Idle
proportion trips unfulfilled vehicle-hours
60% 18.50% 3,209
80% 23.40% 3,433
100% 29.10% 3,264
Table: QoS before change (c rel = $22)
One-way Proportion of Idle
proportion trips unfulfilled vehicle-hours
60% 1.20% 2,871
80% 0.20% 2,981
100% 0.00% 2,711
Table: QoS after change (c rel = $10)
Shen (U of Michigan) 27/33
![Page 36: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/36.jpg)
Change in QoS (denied trips)Definition: Trips disallowed despite there being a car to servedemand. Expected denied trips ↗ as one-way proportion ↗.
One-way Mean 99th
proportion percentile
60% 11.64% 15.47%
80% 15.30% 20.40%
100% 19.71% 25.03%
Table: Denied trips before change (c rel = $22)
One-way Mean 99th
proportion percentile
60% 0.08% 0.82%
80% 0.00% 0.21%
100% 0.00% 0.21%
Table: Denied trips after change (c rel = $10)
Shen (U of Michigan) 28/33
![Page 37: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/37.jpg)
Outline
Introduction
Problem formulation
Solution algorithm
Computational results
Applications and extensions
Shen (U of Michigan) 29/33
![Page 38: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/38.jpg)
Extensions of for-profit carshare
I Not-for-profit car sharing
I City Carshare (Carma): nonprofit carsharing program in BayArea since 2001
I Function: provide last-miles solutions for public transitI Mission: Reduce cars; improve traffic; reduce emissions
I > 80% of Carma’s vehicles are electric powered ⇒I Project 1: Composition of shared car fleet design for meeting
not-for-profit goals and encouraging carsharing
I Project 2: Locating charging stations for shared electricvehicles (EVs)
I Project 3: Joint management of shared EVs in coupled powerand transportation networks
Shen (U of Michigan) 30/33
![Page 39: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/39.jpg)
Extensions of for-profit carshare
I Not-for-profit car sharing
I City Carshare (Carma): nonprofit carsharing program in BayArea since 2001
I Function: provide last-miles solutions for public transitI Mission: Reduce cars; improve traffic; reduce emissions
I > 80% of Carma’s vehicles are electric powered ⇒I Project 1: Composition of shared car fleet design for meeting
not-for-profit goals and encouraging carsharing
I Project 2: Locating charging stations for shared electricvehicles (EVs)
I Project 3: Joint management of shared EVs in coupled powerand transportation networks
Shen (U of Michigan) 30/33
![Page 40: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/40.jpg)
Extensions of for-profit carshare
I Not-for-profit car sharing
I City Carshare (Carma): nonprofit carsharing program in BayArea since 2001
I Function: provide last-miles solutions for public transitI Mission: Reduce cars; improve traffic; reduce emissions
I > 80% of Carma’s vehicles are electric powered ⇒I Project 1: Composition of shared car fleet design for meeting
not-for-profit goals and encouraging carsharing
I Project 2: Locating charging stations for shared electricvehicles (EVs)
I Project 3: Joint management of shared EVs in coupled powerand transportation networks
Shen (U of Michigan) 30/33
![Page 41: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/41.jpg)
Carshare for underserved communities
Underserved communities: Disabled, elderly, and low-income
I limited access to personal vehicles
I limited access to high techs used for booking vehicles
I limited access to services other than transportation need
A vehicle-and-service-sharing system (V3S):
I allocate vehicles to communities shared by households
I match vehicle sharing with service sharing needs
I encourage service sharing by waiving/decreasing drivers’vehicle usage fee
I periodically relocates vehicles across multiple communities bylearning service sharing behavior
Shen (U of Michigan) 31/33
![Page 42: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/42.jpg)
Carshare for underserved communities
Underserved communities: Disabled, elderly, and low-income
I limited access to personal vehicles
I limited access to high techs used for booking vehicles
I limited access to services other than transportation need
A vehicle-and-service-sharing system (V3S):
I allocate vehicles to communities shared by households
I match vehicle sharing with service sharing needs
I encourage service sharing by waiving/decreasing drivers’vehicle usage fee
I periodically relocates vehicles across multiple communities bylearning service sharing behavior
Shen (U of Michigan) 31/33
![Page 43: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/43.jpg)
Take a bigger step?
Next: Autonomous vehicle (AV) sharing?
I Obtain data about AV speed and connectivity from M-city.
I Design AV-based carshare system and V3S.
I Develop stochastic dynamic program for AV control.
Shen (U of Michigan) 32/33
![Page 44: Optimizing the Profitability and Quality of Service in Carshare ...siqian/docs/presentation/2016...Optimizing the Pro tability and Quality of Service in Carshare Systems Siqian Shen](https://reader034.fdocuments.us/reader034/viewer/2022051903/5ff45f0c7e4b391fc37fff4b/html5/thumbnails/44.jpg)
Thank you!
Questions?
Shen (U of Michigan) 33/33