Introduction of my study:
    A method for improving merging of
highway traffic by tuning each car’s behavior

        Suri-Jyokyo-no-kai (SJK seminar)
                 18th, Jun, 2012
                 Ryosuke Nishi

                                            1
Table of contents
• Background of traffic flow
  – Engineering and physics
• Direction of my study
  – Utilize each car’s behavior for improving flow
• Main topic: improving merging efficiency
• Conclusions



                                                     2
Background
   • Easing jams is very significant
       – Financial loss of traffic jam in Japan is 11 trillion
         yen/year*
   • Traffic science is composed of many fields
       –   Engineering
       –   Physics
       –   Economics
       –   Phycology and other fields


* Road Bureau, Ministry of Land, Infrastructure, Transport and Tourism,Japan.
平成18 年度達成度報告書・平成19 年度業績計画書第2 部ii施策-2 効果的な渋滞対策の推進.
http://www.mlit.go.jp/road/ir/ir-perform/h19/02.pdf                             3
Progress in traffic engineering
• Dynamics of traffic flow (at least 1930’s~)
  – Data accumulation(velocity-headway relationships)
  – A lot of models (kinematic, fluid, car-following)
• Development of strategies controlling traffic
  – Ramp metering
  – Route guidance
  – Variable speed limits


                                                    4
Traffic flow in Physics(mainly 1990’s~)

• Self-driven particles (SDPs)
   – Cars, pedestrians, animals, molecular motors
   – action≠reaction: phenomenological modeling


• Microscopic – macroscopic phenomena
   –   free-jam phase transitions
   –   meta-stable branches, synchronized flow
   –   experiment of spontaneous traffic jams
   –   a lot of models(OV model, NaSch model, IDM)
                                                     5
Direction of my study
• To ease jams with each car’s behavior with least
  devices
  – lateral interactions




      Controllers(traffic lights, speed signs)

            Road
                             Car’s behavior
       infrastructure

                                                 6
We focus on interchanges and junctions
 (major place where jams occur in Japanese three highways)
  percentage




We used the data in the following web sites (Feb. 2012)
East Nippon Expressway Company Limited, http://www.e-nexco.co.jp/activity/safety/mechanism.html,
Central Nippon Expressway Company Limited, http://www.c-nexco.co.jp/traffic/jam/cause/,
West Nippon Expressway Company Limited, http://www.w-nexco.co.jp/traffic_info/trafficjam_comment/index2.html 7
To induce zipper merging in weaving sections

• Jam caused by disorderly lane-changes
   – Impossible for cars in a road to see cars in the other road before entering
     merging area
   – High demand to change lanes




                    Disorderly lane-changes



                merging                     bifurcation

                                                                                   8
Emergent zipper merging by drawing orange lines
• Orange lines banns lane-changes
  – Car-to-car repulsive interactions achieves zipper
    configurations

                    Disorderly lane-changes




                           Orange line




               Zipper configurations                    9
Spatial change of configurations
         before lane-changes
               Orange lines




                     Orange lines
Lane1
Lane2


                                      10
Modeling of cars’ movements:
               Cellular Automata (CA)
• Parallel update: discrete time, updating simultaneously
• Movements:stochastic process
   – If the next cell is empty, car 𝑖 moves to it with
     probability 𝑣 𝑖𝑡 between time 𝑡 and 𝑡 + 1




                                                        11
• Time update of 𝑣 𝑖𝑡 :
Multiple lanes stochastic optimal velocity Model

                                                        [1], [2]
          𝑡       𝑡
   – 𝑉(⊿𝑥1,𝑖 , ⊿𝑥2,𝑖 ) : optimal velocity (OV) function
           𝑡       𝑡
      • ⊿𝑥1,𝑖 , ⊿𝑥2,𝑖 : distances of car 𝑖
   – 𝑎 (0 ≤ 𝑎 ≤ 1): sensitivity parameter
      • 𝑣 𝑖𝑡 approaches 𝑉 more quickly as 𝑎 becomes larger




                                                              12
• Interactions between two lanes
  – Four kinds of




    – Deceleration-like Interactions :




                                         13
The measurement of the alternative
            configuration : Simulation

• Example of a
  configuration


• 10 states at 𝑥 = 𝑘, 𝑘 + 1 denoted by 𝑆1 , … , 𝑆10




                                                      14
• Perfect alternative
  configuration: 𝑆3
   𝐶𝑗 (1 ≤ 𝑗 ≤ 10): The number of 𝑆 𝑗 in simulations
• 𝐺(𝑘): the degree of alternative
  configurations at 𝑥 = 𝑘
                  The number ofthe perfect state state
                      The number of the perfect
  𝐺(𝑘)
           That of the states holding
           That of the states holding vehicles at   cars at 𝑥 = 𝑘



                                                                    15
Spatial change to large 𝐺 with car-to-car
                  interactions
                – simulations(dots), mean-field(lines)
zipper




                                                                              p= 1
                                                                           𝑞 = 𝑟 = 0.8
         𝐺




                                                                            d = 100
                                                                             α = 0.05
                                                                       𝛽 𝑗 = 𝑣 (𝑗 = 1,2)
                                                                      105 ≤ 𝑡 < 2 × 105
     *                                                                      #simu: 10

*   *
                                   space[cell]


                                                                                   16
R. Nishi, H. Miki, A. Tomoeda, K. Nishinari, Phys. Rev. E 79, 066119 (2009)
The measurement of the alternative
       configuration: Cluster approximation
• Calculation of the stationary state of the configurations
  on the four cells at 𝑥 = 𝑘, 𝑘 + 1
• Π 𝑡𝑘 (𝑗)(1 ≤ 𝑗 ≤ 10) : The probability of having the
  state 𝑆 𝑗 at time 𝑡 at 𝑥 = 𝑘, 𝑘 + 1
• 𝚷 𝑡𝑘 = Π 𝑡𝑘 1 , … , Π 𝑡𝑘 10 𝑇 : The state vector of the
  four cells at 𝑥 = 𝑘, 𝑘 + 1 at time 𝑡




                                                        17
• State transition
      : State transition matrix at 𝑥 = 𝑘, 𝑘 + 1

• The stationary state:


• The degree of the alternative configuration
     ∞
    𝐺𝑘



                                                  18
Estimate zipper merging with flow rate
• A simple system with two-lane lattice and 1cell

      Car-to-car interactions
         OFF:non-zipper merging
         ON:        zipper merging
𝛼
                                                1
𝛼
                        𝑑




                                                    19
Modeling with CA
• Parallel update(discrete time)
• Exclusive volume effect:impossible to move if the next
  cell is occupied
• To move stochastically with at most 1 cell in 1 time step
                              Deceleration-like
                                interactions
Hopping prob.      𝑣𝑖 = 0          𝑣 𝑖 = 𝑝(≤ 1)      𝑣𝑖 = 1



configurations     * *
  *:any states
                   *                                          20
• When two cars exist just before merging
   – One car chosen in random has to stop
        t            t+1              probability




                                                    21
Slow-to-start rule
• Delay of restarts due to inertia
• Only Just after stopping due to exclusive volume effect,
  Hopping probability is multiplied by 𝑠 0 ≤ 𝑠 ≤ 1 [3]
   – s = 0 (heaviest delay, cars must rest for 1 step)
   – s = 1(no delay)

                t                            t+1
            𝑣𝑖 = 0                   𝑣𝑖 = 𝑠 × 𝑝 , 𝑣𝑖 = 𝑠 × 1



            * *                              * *

            *                                *
                                                               22
Transitions of hopping probability 𝑣 𝑖




                                         23
No slow-to-start at merging area in zipper merging
Non-zipper merging




                     Move with prob. 1 irrespective of cars on
                     neighboring lane

                        Slow-to-start rule (inertia of restarting)
Zipper merging




                                                                                        24
                     Induce zipper configurations with 𝑝 (0< 𝑝 ≤ 1)
                                                                      No blocked cars
Flow rate 𝑄: 𝑄 𝑝 < 1 > 𝑄 𝑝 = 1 for small 𝑠 (large inertia)
• Deceleration-like interactions makes flow rate larger
   – By avoiding slow-to-starting at merging area by


                                𝑄(𝑝<1) > 𝑄(𝑝=1)
            Injection prob. 𝛼




                                                                 s = 0(heavest delay)
                                                                 s = 1(no delay)
                                   Slow-to-start effect : 𝑠
                                                                                           25
R. Nishi, H. Miki, A. Tomoeda, D. Yanagisawa, K. Nishinari, J. Stat. Mech. (2011) P05027
Flow rate difference:
𝑄 𝑝<1 − 𝑄 𝑝=1
    for various 𝑝




                        26
Flow rate 𝑄
     versus
injection rate 𝛼




                   27
Dependence of system-size 𝑑




                              28
Cluster approximations of flow rate
• Simulations (dots), Cluster approximation (lines)




                                                  29
Conclusions
• A method to improve the efficiency of
  merging
  – by induce zipper merging by car-to-car
    interactions
• Spatial change to zipper configurations before
  merging with car-to-car interactions
• Zipper configurations realizes higher flow rate
  – When slow-to-start effect (inertia) is large

                                                   30
Reference
• [1] M. Bando, K. Hasebe, A. Nakayama, A.
  Shibata, and Y. Sugiyama, PRE 51, 1035 (1995)
• [2] Masahiro Kanai, Katsuhiro Nishinari, and
  Tetsuji Tokihiro, PRE 72, 035102 (2005)
• [3] Simon C. Benjamin, Neil F Johnson, and P.
  M. Hui, J. Phys. A: Math. Gen. 29 (1996) 3119–
  3127.


                                               31

Sj kseminar ryosukenishi_for_slideshare

  • 1.
    Introduction of mystudy: A method for improving merging of highway traffic by tuning each car’s behavior Suri-Jyokyo-no-kai (SJK seminar) 18th, Jun, 2012 Ryosuke Nishi 1
  • 2.
    Table of contents •Background of traffic flow – Engineering and physics • Direction of my study – Utilize each car’s behavior for improving flow • Main topic: improving merging efficiency • Conclusions 2
  • 3.
    Background • Easing jams is very significant – Financial loss of traffic jam in Japan is 11 trillion yen/year* • Traffic science is composed of many fields – Engineering – Physics – Economics – Phycology and other fields * Road Bureau, Ministry of Land, Infrastructure, Transport and Tourism,Japan. 平成18 年度達成度報告書・平成19 年度業績計画書第2 部ii施策-2 効果的な渋滞対策の推進. http://www.mlit.go.jp/road/ir/ir-perform/h19/02.pdf 3
  • 4.
    Progress in trafficengineering • Dynamics of traffic flow (at least 1930’s~) – Data accumulation(velocity-headway relationships) – A lot of models (kinematic, fluid, car-following) • Development of strategies controlling traffic – Ramp metering – Route guidance – Variable speed limits 4
  • 5.
    Traffic flow inPhysics(mainly 1990’s~) • Self-driven particles (SDPs) – Cars, pedestrians, animals, molecular motors – action≠reaction: phenomenological modeling • Microscopic – macroscopic phenomena – free-jam phase transitions – meta-stable branches, synchronized flow – experiment of spontaneous traffic jams – a lot of models(OV model, NaSch model, IDM) 5
  • 6.
    Direction of mystudy • To ease jams with each car’s behavior with least devices – lateral interactions Controllers(traffic lights, speed signs) Road Car’s behavior infrastructure 6
  • 7.
    We focus oninterchanges and junctions (major place where jams occur in Japanese three highways) percentage We used the data in the following web sites (Feb. 2012) East Nippon Expressway Company Limited, http://www.e-nexco.co.jp/activity/safety/mechanism.html, Central Nippon Expressway Company Limited, http://www.c-nexco.co.jp/traffic/jam/cause/, West Nippon Expressway Company Limited, http://www.w-nexco.co.jp/traffic_info/trafficjam_comment/index2.html 7
  • 8.
    To induce zippermerging in weaving sections • Jam caused by disorderly lane-changes – Impossible for cars in a road to see cars in the other road before entering merging area – High demand to change lanes Disorderly lane-changes merging bifurcation 8
  • 9.
    Emergent zipper mergingby drawing orange lines • Orange lines banns lane-changes – Car-to-car repulsive interactions achieves zipper configurations Disorderly lane-changes Orange line Zipper configurations 9
  • 10.
    Spatial change ofconfigurations before lane-changes Orange lines Orange lines Lane1 Lane2 10
  • 11.
    Modeling of cars’movements: Cellular Automata (CA) • Parallel update: discrete time, updating simultaneously • Movements:stochastic process – If the next cell is empty, car 𝑖 moves to it with probability 𝑣 𝑖𝑡 between time 𝑡 and 𝑡 + 1 11
  • 12.
    • Time updateof 𝑣 𝑖𝑡 : Multiple lanes stochastic optimal velocity Model [1], [2] 𝑡 𝑡 – 𝑉(⊿𝑥1,𝑖 , ⊿𝑥2,𝑖 ) : optimal velocity (OV) function 𝑡 𝑡 • ⊿𝑥1,𝑖 , ⊿𝑥2,𝑖 : distances of car 𝑖 – 𝑎 (0 ≤ 𝑎 ≤ 1): sensitivity parameter • 𝑣 𝑖𝑡 approaches 𝑉 more quickly as 𝑎 becomes larger 12
  • 13.
    • Interactions betweentwo lanes – Four kinds of – Deceleration-like Interactions : 13
  • 14.
    The measurement ofthe alternative configuration : Simulation • Example of a configuration • 10 states at 𝑥 = 𝑘, 𝑘 + 1 denoted by 𝑆1 , … , 𝑆10 14
  • 15.
    • Perfect alternative configuration: 𝑆3 𝐶𝑗 (1 ≤ 𝑗 ≤ 10): The number of 𝑆 𝑗 in simulations • 𝐺(𝑘): the degree of alternative configurations at 𝑥 = 𝑘 The number ofthe perfect state state The number of the perfect 𝐺(𝑘) That of the states holding That of the states holding vehicles at cars at 𝑥 = 𝑘 15
  • 16.
    Spatial change tolarge 𝐺 with car-to-car interactions – simulations(dots), mean-field(lines) zipper p= 1 𝑞 = 𝑟 = 0.8 𝐺 d = 100 α = 0.05 𝛽 𝑗 = 𝑣 (𝑗 = 1,2) 105 ≤ 𝑡 < 2 × 105 * #simu: 10 * * space[cell] 16 R. Nishi, H. Miki, A. Tomoeda, K. Nishinari, Phys. Rev. E 79, 066119 (2009)
  • 17.
    The measurement ofthe alternative configuration: Cluster approximation • Calculation of the stationary state of the configurations on the four cells at 𝑥 = 𝑘, 𝑘 + 1 • Π 𝑡𝑘 (𝑗)(1 ≤ 𝑗 ≤ 10) : The probability of having the state 𝑆 𝑗 at time 𝑡 at 𝑥 = 𝑘, 𝑘 + 1 • 𝚷 𝑡𝑘 = Π 𝑡𝑘 1 , … , Π 𝑡𝑘 10 𝑇 : The state vector of the four cells at 𝑥 = 𝑘, 𝑘 + 1 at time 𝑡 17
  • 18.
    • State transition : State transition matrix at 𝑥 = 𝑘, 𝑘 + 1 • The stationary state: • The degree of the alternative configuration ∞ 𝐺𝑘 18
  • 19.
    Estimate zipper mergingwith flow rate • A simple system with two-lane lattice and 1cell Car-to-car interactions OFF:non-zipper merging ON: zipper merging 𝛼 1 𝛼 𝑑 19
  • 20.
    Modeling with CA •Parallel update(discrete time) • Exclusive volume effect:impossible to move if the next cell is occupied • To move stochastically with at most 1 cell in 1 time step Deceleration-like interactions Hopping prob. 𝑣𝑖 = 0 𝑣 𝑖 = 𝑝(≤ 1) 𝑣𝑖 = 1 configurations * * *:any states * 20
  • 21.
    • When twocars exist just before merging – One car chosen in random has to stop t t+1 probability 21
  • 22.
    Slow-to-start rule • Delayof restarts due to inertia • Only Just after stopping due to exclusive volume effect, Hopping probability is multiplied by 𝑠 0 ≤ 𝑠 ≤ 1 [3] – s = 0 (heaviest delay, cars must rest for 1 step) – s = 1(no delay) t t+1 𝑣𝑖 = 0 𝑣𝑖 = 𝑠 × 𝑝 , 𝑣𝑖 = 𝑠 × 1 * * * * * * 22
  • 23.
    Transitions of hoppingprobability 𝑣 𝑖 23
  • 24.
    No slow-to-start atmerging area in zipper merging Non-zipper merging Move with prob. 1 irrespective of cars on neighboring lane Slow-to-start rule (inertia of restarting) Zipper merging 24 Induce zipper configurations with 𝑝 (0< 𝑝 ≤ 1) No blocked cars
  • 25.
    Flow rate 𝑄:𝑄 𝑝 < 1 > 𝑄 𝑝 = 1 for small 𝑠 (large inertia) • Deceleration-like interactions makes flow rate larger – By avoiding slow-to-starting at merging area by 𝑄(𝑝<1) > 𝑄(𝑝=1) Injection prob. 𝛼 s = 0(heavest delay) s = 1(no delay) Slow-to-start effect : 𝑠 25 R. Nishi, H. Miki, A. Tomoeda, D. Yanagisawa, K. Nishinari, J. Stat. Mech. (2011) P05027
  • 26.
    Flow rate difference: 𝑄𝑝<1 − 𝑄 𝑝=1 for various 𝑝 26
  • 27.
    Flow rate 𝑄 versus injection rate 𝛼 27
  • 28.
  • 29.
    Cluster approximations offlow rate • Simulations (dots), Cluster approximation (lines) 29
  • 30.
    Conclusions • A methodto improve the efficiency of merging – by induce zipper merging by car-to-car interactions • Spatial change to zipper configurations before merging with car-to-car interactions • Zipper configurations realizes higher flow rate – When slow-to-start effect (inertia) is large 30
  • 31.
    Reference • [1] M.Bando, K. Hasebe, A. Nakayama, A. Shibata, and Y. Sugiyama, PRE 51, 1035 (1995) • [2] Masahiro Kanai, Katsuhiro Nishinari, and Tetsuji Tokihiro, PRE 72, 035102 (2005) • [3] Simon C. Benjamin, Neil F Johnson, and P. M. Hui, J. Phys. A: Math. Gen. 29 (1996) 3119– 3127. 31