Scholarly article on topic 'Agent-based Simultaneous Optimization of Congestion and Air Pollution: A Real-World Case Study'

Agent-based Simultaneous Optimization of Congestion and Air Pollution: A Real-World Case Study Academic research paper on "Economics and business"

CC BY-NC-ND
0
0
Share paper
Academic journal
Procedia Computer Science
OECD Field of science
Keywords
{"Air Pollution" / Congestion / "Vehicle Emissions" / "Road Pricing" / "Simultaneous Pricing" / Optimization / "Agent-based Modeling"}

Abstract of research paper on Economics and business, author of scientific article — Amit Agarwal, Benjamin Kickhöfer

Abstract The exclusion of external costs from the behavioral decision making process of individuals yields travel demand beyond the system optimum which implies inefficiencies in the transport system. The present study investigates the effect of congestion optimization on emissions levels and vice versa while considering heterogeneity in individual attributes and choice behavior. In consequence, the resulting correction terms (tolls) are highly differentiated. Furthermore, and going beyond existing literature, the present study proposes a joint optimization of vehicular congestion and emissions. The proposed model uses a microscopic agent-based simulation framework which is applied to a real-world scenario of the Munich metropolitan area in Germany. The combined pricing scheme accounts for both external effects and in an iterative process, agents learn how to adapt their route and mode choice decisions in presence of this combined toll. The results indicate that the combined pricing strategy moves the car transport system towards the optimum, measured by a strong decrease of congestion and emission costs. Furthermore, it is found that pricing emissions only pushes users on routes with shorter distances, whereas pricing congestion only steers users on routes with shorter travel times, and potentially longer distances. That is, the two pricing strategies influence behavior by tendency into opposite directions.

Academic research paper on topic "Agent-based Simultaneous Optimization of Congestion and Air Pollution: A Real-World Case Study"

(8)

CrossMark

Available online at www.sciencedirect.com

ScienceDirect

Procedia Computer Science 52 (2015) 914 - 919

4th International Workshop on Agent-based Mobility, Traffic and Transportation Models, Methodologies and Applications, ABMTRANS 2015

Agent-Based Simultaneous Optimization of Congestion and Air Pollution: A Real-World Case Study

Amit Agarwal*, Benjamin Kickhofer

Technische Universität Berlin, Transport Systems Planning and Transport Telematics, Salzufer 17-19, 10587 Berlin, Germany

Abstract

The exclusion of external costs from the behavioral decision making process of individuals yields travel demand beyond the system optimum which implies inefficiencies in the transport system. The present study investigates the effect of congestion optimization on emissions levels and vice versa while considering heterogeneity in individual attributes and choice behavior. In consequence, the resulting correction terms (tolls) are highly differentiated. Furthermore, and going beyond existing literature, the present study proposes a joint optimization of vehicular congestion and emissions. The proposed model uses a microscopic agent-based simulation framework which is applied to a real-world scenario of the Munich metropolitan area in Germany. The combined pricing scheme accounts for both external effects and in an iterative process, agents learn how to adapt their route and mode choice decisions in presence of this combined toll. The results indicate that the combined pricing strategy moves the car transport system towards the optimum, measured by a strong decrease of congestion and emission costs. Furthermore, it is found that pricing emissions only pushes users on routes with shorter distances, whereas pricing congestion only steers users on routes with shorter travel times, and potentially longer distances. That is, the two pricing strategies influence behavior by tendency into opposite directions.

© 2015 The Authors. PublishedbyElsevierB.V.This is an open access article under the CC BY-NC-ND license

(http://creativecommons.Org/licenses/by-nc-nd/4.0/).

Peer-review under responsibility of the Conference Program Chairs

Keywords: Air Pollution, Congestion, Vehicle Emissions, Road Pricing, Simultaneous Pricing, Optimization, Agent-based Modeling

1. Introduction

Car traffic in urban areas has led to an increase in negative externalities such as road congestion, damage to the environment, and human health.27,18 These externalities yield efficiency losses which can sum up to a significant part of a country's GDP (Gross Domestic Product). For example, total external costs by motorized traffic in Beijing range between 7.5% and 15% of GDP.8 The total external costs in the EU-27 for year 2008 was 373 billion EUR -equivalent to 3.0% of the region's GDP.4

Congestion externalities occur, since every vehicle on the network imposes costs on other vehicles in terms of increased travel times which could be used to perform a beneficial activity. Similarly, exhaust emission externalities

* Corresponding author. Tel.: +49-30-314-23308 ; fax: +49-30-314-26269. E-mail address: amit.agarwal@campus.tu-berlin.de

1877-0509 © 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license

(http://creativecommons.Org/licenses/by-nc-nd/4.0/).

Peer-review under responsibility of the Conference Program Chairs

doi:10.1016/j.procs.2015.05.165

occur, since vehicular traffic emits exhaust emissions such as CO, NOx, PM, SO2 which are the main components for polluting air and these in turn are responsible for adverse effects on health and living conditions. However, exclusion of congestion and/or emission externalities in people's mobility decisions deviates demand from system optimum.

In order move the transport system towards the optimum, planners and policy makers may look for measures which reduce the efficiency losses caused by the negative externalities. From the literature, it is known that pricing schemes can be used to achieve this goal of shifting the system from the user equilibrium towards the system optimum.21 This lead to many studies on road pricing which focus on finding the theoretically optimal tolls for road congestion. 17,24,25,11,1 However, there is only a limited number of contributions that aimed at finding optimal toll levels for emissions16 or for congestion and emissions simultaneously.26,22 The former solve the problem using a simulation-based optimization for large-scale scenarios considering dynamic traffic flows. The latter use an analytical approach for a small six-node network, and a large-scale scenario of Brussels (Belgium) with static traffic flows, respectively. None of these studies attempted a combined optimization for congestion and emissions in a large-scale scenario with dynamic traffic flows and activity-based demand. More practically oriented contributions investigated the impact of various pricing strategies on congestion and emission levels and find that these two externalities are positively correlated.22,3,9,5,20,2

The present study extends the existing literature by applying a combined optimization approach for automobile emissions and congestion in a large-scale scenario with dynamic traffic flows and activity-based demand. For that purpose, the congestion pricing approach by Kaddoura and Kickhofer13 and the emission pricing approach by Kickhofer and Nagel16 are simultaneously applied to a real world scenario of the Munich metropolitan area in Germany. The contribution of the combined approach is to determine individual vehicle-specific, time-dependent toll levels that include both externalities under consideration. The study calculates the effects of the correction term, which increases system efficiency, for a particular case study. The methodology can, however, be applied to any scenario worldwide.

2. Modeling

Travel demand simulator. MATSim19 provides a framework for simulating transport in large-scale scenarios. Every individual is considered as an agent who learns within an iterative process that is composed of three steps: (1) Execution: Selected plans of all agents are executed simultaneously in the physical environment. (2) Evaluation: The performance of the executed plan is calculated using a scoring function. In this study, the MATSim default scoring function7 is used with behavioral parameters from Kickhofer (2014).14 (3) Re-planning: A new plan is generated for some agents by modifying an existing plan (e.g. route or transport mode)1.

Emission pricing. The existing emission modeling tool along with MATSim framework is used for the calculation of cold and warm emissions which depend on parking duration, traveled distance, vehicle characteristics, and engine type, road category, speed of vehicles, respectively.15 Furthermore, time-dependent, vehicle-specific emission tolls are calculated using the methodology developed by Kickhofer and Nagel16 and the emission cost factors provided by Maibach et al18. In the simulation, every time an agent leaves a road segment, it is charged with a toll equivalent to the produced emissions.

Congestion pricing. Following the methodology developed by Kaddoura and Kickhofer13, dis-aggregated delays on each link for each agent are computed. It is defined as the difference of actual and minimum link travel time. A tracking of agents identifies causing and affected agents. The former can then be charged with the equivalent monetary amount of the delays. Since congestion is - in contrast to emissions - inherent to road traffic, the behavioral parameters of MATSim14 can directly be used to convert delays into toll levels. These monetary payments are then charged to the causing agent.

All monetary payments are considered in the performance-based learning cycle of MATSim, and hence, the external effect is taken into account by the agents for their route and mode choice behavior.

1 In the present study, dropping or re-scheduling of activities is not considered.

3. Real-World Application

The initial scenario is taken from Kickhofer and Nagel16 and shortly described next:

Input. Raw data provided by the municipality of Munich23 is used to create the road network. Activity-based demand is created using different data sources: (i) 1.4 million inner urban travelers10, (ii) 0.3 million commuters and 0.2 million reverse commuters6, and, (iii) roughly 0.15 million freight trips12. For the present study, 1% of the total population is used for computational performance reasons.

Scenario set-up. Initially a base case is set up by running the simulation for 1000 iterations. It is used as the reference scenario for the different pricing strategies. Four strategies namely, Business As Usual (BAU), Emission Internalization (EI), Congestion Internalization (CI), and combined Emission Congestion Internalization (ECI) are then simulated. These strategies are differentiated based on internalization of external costs as their names depict. Each strategy is run for another 500 iterations. For 80% of the iterations, 15% agents switch route and 15% shifts travel mode between car and public transport (PT). The rest of the agents choose plans according to a multinomial logit model. After 80% of the iterations, all agents choose plans from their generated choice set following the latter strategy. In the study, travel modes other than car are assumed to run emission free and as without capacity constraints and thus, such travel modes are grouped together as non-car travel modes.

Analysis. Time-dependent and person-specific emission costs are calculated and aggregated for each strategy in order to obtain total emissions costs. Similarly, dis-aggregated delays are calculated for each affected agent and then converted into toll levels. Afterwards, these values are summed up to get the total congestion costs for each strategy.

Table 1: Comparison of pricing strategies in terms of modal split, external costs and travel times

Strategies BAU EI CI ECI

Change in emission costs - -0.57% -1.94% -2.48%

Change in congestion costs - -3.26% -42.83% -43.32%

Modal split (car : non-car) 32.47 : 67.53 32.21 : 67.64 30.90 : 69.10 30.76 : 69.12

Avg. trip travel time car (min) 60.32 59.68 47.80 47.75

Avg. trip travel time non-car (min) 16.95 17.06 18.09 18.22

Avg. trip travel time car to non-car (min) 20.71 22.72 36.19 38.24

Table 1 shows the comparison of all strategies with respect to BAU. When pricing emissions only (EI), the reduction in emission and congestion costs are 0.57% and 3.26%, respectively. Similarly, pricing congestion reduces emission and congestion costs by 1.94% and 42.83%, respectively2. Thus, in accordance to the literature, pricing one externality has a positive impact on other and consequently, they prove to be positively correlated. 3,9,22,5,20,2

However, pricing only one externality does not achieve the same increase in car system efficiency as the combined pricing scheme (ECI): The latter strategy reduces emission and congestion costs caused in the car mode by 2.48% and 43.32%, respectively. That is, the combined approach yields the highest efficiency gains of the presented strategies for the car transport system. This is supported by the decrease in average trip travel time for car3 in Table 1. For non-car users, this average trip travel time increases for the different pricing strategies which implies that especially longer

2 Compared to the changes in emission costs, the change in congestion costs seems rather high. This is, however, due to the fact that delays can actually be avoided (e.g. by shifting enough individuals to PT), whereas emission costs can only be avoided by shifting all individuals to PT, or by changing completely to zero emission vehicles which is not considered in this study.

3 The average car travel times seem rather high compared to PT. This is due to the fact that freight, commuters and reverse commuters (mainly) use car and drive long distances.

trips are shifting to the non-car mode. This indicates that the above efficiency gain in the car transport system needs to be compared with the increased travel times for individuals who switch from car to non-car mode. For them, it can observed that the average trip travel time increases for all pricing strategies. However, given that the users remaining in the car mode by far outnumber the mode switchers (see change in modal split), the pricing strategies still yield efficiency gains for the whole transport system.

(a) Absolute NO2 emissions (b) Absolute delays

Fig. 1: Absolute NO2 emissions and absolute delays for BAU

Furthermore, in order to understand the agent's behavior under different pricing strategies, spatial plots are used. Gaussian distance weighing function is used to smooth emissions and delays throughout the area of Munich and surroundings, using uniform hexagonal cells with a smoothing radius of 500 m. Fig. 1 and Fig. 2 show a variety of spatial plots. Absolute NO2 emissions for BAU are presented in Fig. 1a as a reference.4 One can observe that absolute emissions are most important on primary roads (inner and middle ring road, main arterials, and the tangential motorway in the north-west of Munich). In contrast, congestion is evident on almost all major roads inside the city area, but not as important on the tangential motorway (see Fig. 1b).

Fig. 2a to 2c present the change in NO2 emissions and the change in delays by color scale and opacity, respectively. Yellow to red colors indicate an increase, green colors a decrease in NO2 emissions. Transparent colors denote a decrease in delays whereas opaque colors exhibit an increase in delays.

When pricing emissions only, re-routing steers most of the agents to shorter distance routes (see Fig. 2a) and some of the agents to shift to non-car travel modes (see Table 1). As a result, an increase in NO2 emissions (orange and yellow hexagons) and an increase in congestion (opaque) can be identified mainly in the north of the inner city, whereas a decrease in emissions and delays is observed on the longer distance routes. From Fig. 2b it can be noticed that pricing congestion only reduces NO2 emissions and congestion in urban areas of Munich significantly (transparent green hexagons). On the other hand, this strategy increases emission (red hexagons) on the tangential motorway in the north-west of Munich where the intensity of NO2 emissions was already high in the BAU case (see Fig. 1a). Clearly, congestion pricing redirects agents to less congested routes (see Fig. 2b) and to non-car travel modes (see Table 1). The effects of simultaneous pricing are shown in Fig. 2c. Since congestion costs in BAU are by a factor of 3.4 higher than emission costs, the agent's behavior is dominated by the congestion toll. Fig. 2c therefore looks rather similar to Fig. 2b. However, as shown in Fig. 2c, the orange and red hexagons are a lot less than in congestion pricing. This again shows the need for investigating the effects of combined pricing approaches that incorporate more than one externality.

4 All important pollutants are considered for pricing. For illustration purposes, the emission plots only show NO2 values.

(a) Change in NO2 emissions for EI

(b) Change in NO2 emissions for CI (c) Change in NO2 emissions for ECI

Fig. 2: Change in NO2 emissions for the different pricing strategies with respect to BAU. Additionally, transparent colors denote a reduction in congestion whereas opaque colors show an increase in congestion.

4. Conclusion and Outlook

This study examined the impact of congestion pricing on emission levels and vice versa. Both externalities are found to be positively correlated since introducing a correction term in the form of a toll for one externality also reduced the other externality. Furthermore, it was demonstrated that a combined pricing strategy increases the car system efficiency most importantly yielding lowest congestion and emission costs. Additionally, agents' behavior for both pricing strategies was investigated with help of spatial plots. It was found that pricing emissions pushes users on routes with shorter distances, whereas pricing congestion steers users on routes with shorter travel times, and potentially longer distance. That is, for congested areas, the two regimes influence route choice behavior by tendency into opposite directions.

Finally, it can be concluded that with the methodology developed in this paper, efficient prices for negative externalities in large-scale urban areas can be derived and can be used as benchmarks when evaluating the effects of real-world policies. Beside the external costs considered here, the approach can also be applied to other external costs such as noise or accidents. In future studies, the authors aim at evaluating the proposed strategies economically on a more dis-aggregated level, and incorporate emission exposure to gain most out of the underlying activity-based demand.

Acknowledgements

The support given by DAAD (German Academic Exchange Service) to A. Agarwal for his PhD studies at Technische Universität Berlin is greatly acknowledged. The authors also wish to thank Kai Nagel (Technische Univser-sitat Berlin) for his helpful comments as well as H. Schwandt and N. Paschedag at the Department of Mathematics (Technische Universitat Berlin), for maintaining our computing clusters. Important data was provided by the Municipality of Munich, more precisely by Kreisverwaltungsreferat Munchen and Referat fur Stadtplanung und Bauordnung München.

References

1. Arnött, R., de Palma, A., and Lindsey, R. A structural model of peak-period congestion: A traffic bottleneck with elastic demand. The

American Economic Review 83, 1 (1993), 161-179.

2. Barth, M., and Böriböönsömsin, K. Traffic congestion and greenhouse gases. ACCESS Magazine, University of California Transportation Center 35 (2009).

3. Beamön, B. M., and Griefin, P. M. A simulation-based methodology for analyzing congestion and emissions on a transportation network. Simulation 72, 2 (1999), 105-114.

4. Becker, U. J., Becker, T., and Gerlach, J. Externe Autokosten in der EU-27: überblick über existierende Studien, 2012.

5. Beevers, S. D., and Carslaw, D. C. The impact of congestion charging on vehicle emissions in London. Atmospheric Environment 39 (2005), 1-5.

6. Böhme, S., and Eigenmüller, L. Pendlerbericht Bayern. Tech. rep., IAB, 2006.

7. Charypar, D., and Nagel, K. Generating complete all-day activity plans with genetic algorithms. Transportation 32, 4 (2005), 369-397.

8. Creützig, F., and He, D. Climate change mitigation and co-benefits of feasible transport demand policies in Beijing. Transportation Research Part D: Transport and Environment 14, 2 (2009), 120-131.

9. Daniel, J. I., and Bekka, K. The environmental impact of highway congestion pricing. Journal of Urban Economics 47 (2000), 180-215.

10. Föllmer, R., Künert, U., Klöas, J., and Küheeld, H. Mobilitat in Deutschland - Ergebnisbericht. Tech. rep., infas/DIW, 2004.

11. Henderson, J. V. Road congestion: A reconsideration of pricing theory. Journal of Urban Economics 1 (1974), 354-365.

12. ITP, and BVU. Prognose der deutschlandweiten Verkehrsverflechtungen 2025. Tech. rep., Intraplan Consult GmbH, Beratergruppe Verkehr+Umwelt GmbH, 2007.

13. Kaddöüra, I., and Kickhöfer, B. Optimal road pricing: Towards an agent-based marginal social cost approach. VSP Working Paper 14-1, TU Berlin, Transport Systems Planning and Transport Telematics, 2014. See http://www.vsp.tu-berlin.de/publications.

14. Kickhöfer, B. Economic Policy Appraisal and Heterogeneous Users. PhD thesis, Technische Universitat Berlin, 2014. Also VSP WP 14-11, see http://www.vsp.tu-berlin.de/publications.

15. Kickhöfer, B., Hülsmann, F., Gerike, R., and Nagel, K. Rising car user costs: comparing aggregated and geo-spatial impacts on travel demand and air pollutant emissions. In ,Smart Transport Networks: Decision Making, Sustainability and Market structure, T. Vanoutrive and A. Verhetsel, Eds., NECTAR Series on Transportation and Communications Networks Research. Edward Elgar Publishing Ltd, 2013, pp.180-207.

16. Kickhöfer, B., and Nagel, K. Towards high-resolution first-best air pollution tolls. Networks and SSpatial Economics (2013), 1-24.

17. Lindsey, R., and Verhöee, E. T. Traffic congestion and congestion pricing. In Handbook of Transport Systems and Traffic Control, D. Hensher and K. Button, Eds., vol. 3 of Handbooks in Transport. 2001, pp. 77-105.

18. Maibach, M., Schreyer, D., Sutter, D., van Essen, H., Boon, B., Smökers, R., Schroten, A., Doll, C., Pawlöwska, B., and Bak, M. Handbook on estimation of external costs in the transport sector. Tech. rep., CE Delft, 2008. Internalisation Measures and Policies for All external Cost of Transport (IMPACT).

19. MATSim. MultiAgent Transport SIMulation webpage, accessed 2014.

20. Namdeo, A., and Mitchell, G. An empirical study of estimating vehicle emissions under cordon and distance based road user charging in Leeds, UK, 2008.

21. Pigou, A. The Economics of Welfare. MacMillan, New York, 1920.

22. Proost, S., and van Dender, K. The welfare impacts of alternative policies to address atmospheric pollution in urban road transport. Regional Science and Urban Economics 31, 4 (2001), 383-411.

23. RSB. Municipality of Munich: Referat fur Stadtplanung und Bauordnung, 2005.

24. Small, K. A., and Verhöee, E. T. The economics of urban transportation. Routledge, 2007.

25. Vickrey, W. Congestion theory and transport investment. The American Economic Review 59, 2 (1969), 251-260.

26. Wang, J., Chi, L., Hü, X., and Zhou, H. Urban traffic congestion pricing model with the consideration of carbon emissions cost. Sustainability 6(2) (2014), 676-691.

27. Weinreich, S., Rennings, K., Schlomann, B., Gegner, C., and Engel, T. External costs of road, rail and air transport - A bottom-up approach.

ZEW Discussion Papers 98-06 (1998).