Introduction
-
Comparison of three pro-active relocation heuristics for shared automated vehicles under parking constraints.
-
Introducing a fleet of shared automated vehicles into an agent-based model for a large-scale case study based on the city of Amsterdam.
-
Holistic impact analysis of SAV in regard to service efficiency, service provision equity, and service externalities.
Relocating shared automated vehicles
Problem description
Network
Demand for SAV
Supply of SAV
Vehicle relocation heuristics
Relocation strategy “Demand Anticipation”
Relocation strategy “Supply Anticipation”
Relocation according to “Demand–Supply Deficit Minimization”
Performance and level-of-service synthesis
Study | Applied relocation strategy | Service efficiency indicators | Service externality indicators | Service equity indicators |
---|---|---|---|---|
Azevedo et al. (2016) | Demand–Supply Balancing Relocating of vehicles between stations to balance out supply and demand | Average passenger waiting time per person-trip | – | – |
Babicheva et al. (2018) | Demand Anticipation Relocating of vehicles based on current and future demand at pick-up stations Demand–Supply Balancing Reducing vehicle surplus or deficit at pick-up stations | Average and maximal passenger time per person-trip; total vehicle run-time | – | – |
Bischoff and Maciejewski (2016) | Supply Anticipation Hourly rebalancing of vehicles at pick-up stations based on anticipated supply deficit while minimizing costs of rebalancing trip | Average passenger waiting time per person-trip; Average vehicle utilisation (in hourly time shares) | – | – |
van Engelen et al. (2018) | Demand–Supply Balancing Rebalancing of vehicles to balance out supply and demand | Average passenger waiting time per person-trip | Vehicle-Miles travelled (VMT); | Percentage of rejected requests |
Fagnant and Kockelman (2014) | Demand–Supply Balancing Simulation of 4 strategies spreading idle vehicles out either according to “block balance” or move excess idle vehicle (any more than 2 per zone) to relocate to zones unoccupied by idle vehicles | Average passenger waiting time per person-trip; Average vehicle utilisation (in VMT) | Vehicle-Miles travelled (VMT); VMT caused by induced demand; Number of warm and cold starts per person-trip and per day | – |
Hörl et al. (2019) | Demand Anticipation optimized relocation under full knowledge of future demand based on feedforward fluidic optimal rebalancing algorithm Supply & Demand Anticipation Even distribution of vehicles during off-peak hours and demand anticipatory relocation during peak hours based on adaptive uniform rebalancing algorithm | Passenger waiting time per person-trip; fleet utilisation (active time per vehicle, empty mileage per vehicle); occupancy, operating time | Average speed, trip length per passenger | – |
Sayarshad and Chow (2017) | Demand–Supply Balancing Simulation of a heuristics using queuing-based approximation on a zonal level, solved by Lagrangian decomposition | User cost and system cost measured in waiting time | – | – |
Winter et al. (2017) | Supply Anticipation Based on heuristic to balance supply on a zonal level Demand Anticipation Based on heuristic to anticipate demand on | Average passenger waiting time per person-trip; Average vehicle utilisation (in time shares) | Maximum parking demand per link; average parking duration | – |
Zhang et al. (2015) | Demand–Supply Balancing Idle vehicles cruises for a couple of minutes in area with highest demand–supply deficit before parking there | Average passenger waiting time per person-trip | Average parking demand per SAV and daily parking demand of the total fleet | Spatial distribution of parking demand |
Zhang and Pavone (2016) | Demand–Supply Balancing balance demand and supply per pick-up stations determined by k-means clustering | – | Congestion effects | – |
Zhang et al. (2016) | Cruising Simulation of 1 algorithm randomly cruising idle vehicles Demand Anticipation 5 demand-anticipatory algorithms | Average passenger waiting time per person-trip | – | – |
Case study application
Travel demand
Specification of SAV and their infrastructure needs
Behavioural model and model specifications
Mode | ASCm(q) | βtravel_time,m(q) | βtravel_cost | costq |
---|---|---|---|---|
Car | 0.0 | − 10.7 | 1 | 30 €-cent/km |
Public transport | − 8.3 | − 6.65 | 1 | 25 €-cent/km |
Cycling | − 1.0 | − 10.7 | 1 | 0 €-cent/km |
Walking | 0.3 | − 6.65 | 1 | 0 €-cent/km |
SAV | 0.0 | − 10.7 | 1 | 30 €-cent/km |
1. Network & Geography | |||
Number of simulated agents (represented by 20%) | 767,495 | ||
Network: number of links | 31,131 | ||
Network: number of nodes | 17,385 | ||
Area size: greater metropolitan area (in km2) | 50,888 | ||
Area size: core city (= service area of the SAV) (in km2) | 211 | ||
Number of zones covering the core city (= service area of the SAV) | 82 | ||
2. Specified behavioural model of the MATSim agents | Phase 1 | Phase 2 | Final run |
Routing algorithm | Dijkstra | ||
Coefficient for the utility of performing an activity (\( \beta_{duration} \)) | 16.25 | ||
Coefficient for arriving late is weighed \( (\beta_{late\_arrival} \)) | − 48.75 | ||
Maximum plan memory of agents | 5 | 3 | 1 |
Fraction of iterations after which plan innovation is disabled and score convergence is enabled (in %) | 85 | 100 | 100 |
Plan innovations based on utility model (in %) | 70 | 70 | 90 |
Plan innovations based on re-routing for car trips (in %) | 10 | 5 | 5 |
Plan innovations based on changed departure time (± 15 min) (in %) | 5 | 5 | 5 |
Plan innovations based on changed single trip modes (in %) | 10 | 10 | 0 |
Plan innovations based on changed sub-tour modes (in %) | 5 | 10 | 0 |
Number of simulation runs | 75 | 15 | 2 |
→Resulting modal share for SAV (in %) | 4.5 | 4.3 | 4.3 |
3. Mode options and specifications | |||
Modes simulated by teleportation | Walk, bike, public transport | ||
Beeline distance factor for teleportation | 1.3 | ||
Teleportation speed: walking (in km/h) | 5 | ||
Teleportation speed: cycling (in km/h) | 15 | ||
Teleportation speed: public transport | Freespeed car travel | ||
Number of simulated SAV (represented by 20%) | 12,500 | ||
Number of parking spots dedicated for SAV (represented by 20%) | 15,000 | ||
MATSim-specific dispatching algorithm for SAV | “Rule-based” | ||
Re-optimization time step for SAV (in seconds) | 60 | ||
Pick-up time for SAV (in seconds) | 120 | ||
Drop-off time for SAV (in seconds) | 60 | ||
4. Average simulation runtime of 1 full day on desktop PC with Intel Core i5-3470 3.2 GHz, 16 GB RAM; including replanning phase and dumping of output files | |||
Without SAV (in seconds) | 308 | ||
With SAV, no relocation (in seconds) | 648 | ||
With SAV, “Demand-Anticipation” (in seconds) | 1715 | ||
With SAV, “Supply-Anticipation” (in seconds) | 23,043 | ||
With SAV, “Demand–Supply-Balancing” (in seconds) | 21,892 |
Results
Service efficiency
Demand Anticipation | Supply Anticipation | Demand–Supply balancing | |
---|---|---|---|
Share of empty driven mileage over total driven mileage: \( \frac{{VKT_{SAV\_empty} }}{{VKT_{SAV} }} \left( {{\text{in}}\;{\text{\% }}} \right) \) | 56.1 | 57.1 | 57.1 |
Share of driven mileage for relocation over total empty driven mileage: \( \frac{{VKT_{SAV\_relocating} }}{{VKT_{SAV\_empty} }} \left( {{\text{in}}\;{\text{\% }}} \right) \) | 70.5 | 75.0 | 75.2 |
Share of time driven emptily: \( \frac{{tt_{SAV\_empty} }}{{tt_{total} }} \left( {{\text{in}}\;{\text{\% }}} \right) \) | 14.0 | 14.5 | 14.6 |
Average in-vehicle times per trip in SAV: \( ivt_{SAV } \left( { \hbox{min} } \right) \) | 18.0 | 18.4 | 18.2 |
Average and 95% percentile of passenger waiting time: \( t_{SAV\_wait} ; t_{SAV\_wait\_95\% } \left( { \hbox{min} } \right) \) | 4.6; 12.1 | 3.6; 9.4 | 3.5; 9.1 |
Waiting times
Empty driven mileage
Trip times
Service efficiency: summary
Service externalities
Demand Anticipation | Supply Anticipation | Demand–Supply balancing | |
---|---|---|---|
Average driving speed for SAV: \( v_{SAV} \left( {\frac{{\text{km}}}{{\text{h}}}} \right) \) | 39.2 | 39.2 | 39.1 |
Average driving speed of SAV with and without passengers on-board: \( v_{SAV\_IVT} ; v_{SAV\_empty} \left( {\frac{{\text{km}}}{{\text{h}}}} \right) \) | 38.9; 39.6 | 39.3; 39.1 | 39.0; 39.2 |
Total mileage of SAV: \( VKT_{SAV } \left( {{\text{in}}\;1000\;{\text{km}}} \right) \) | 3519 | 3610 | 3608 |
Congestion
Driven mileage
Parking space consumption
Service externalities: summary
Service provision equity
Demand Anticipation | Supply Anticipation | Demand–Supply balancing | |
---|---|---|---|
Gini-coefficient for passenger waiting times \( G_{wait} \) | 0.554 | 0.517 | 0.507 |
Gini-coefficient for average zonal passenger waiting times \( G_{wait, z} \) | 0.291 | 0.276 | 0.265 |
Zonal average waiting times (in min)
|