Evaluation of car/bicycle traffic measures with - - PowerPoint PPT Presentation
The 18 th summer course of Behavior Modeling Final presentation Evaluation of car/bicycle traffic measures with a
The 18th summer course of Behavior Modeling Final presentation
University of Tokyo team A Takuya Iizuka (M1) Kenta Ishii (M1) Shoma Dehara (M1) Miho Yonezawa (M1)
Population: 512479 (2018.1.1.) Area: 429.06 m2
http://udcm.jp/project/
Motorcycle 11%
Train 2% Bus 1% Taxi 0%
Walk 15%
Other 1%
Car Motorcycle Train Bus Taxi Bicycle ※経路情報が得られたトリップを抽出
(2007 Feb.19 – Mar.23)
→By providing bicycle lanes, traffic accidents can be suppressed !!
◆ Traffic Volume in the center of Matsuyama Car Trip Bicycle Trip
Center Station City Hall Dogo Onsen JR Station Center Station City Hall Dogo Onsen JR Station
the car & bicycle trips are separated.
The smaller the traffic of the car, the more traffic of the bicycle. On links with heavy car traffic, sidewalks are maintained, increasing bicycle traffic.
→ We adopt Link Base Route Choice Model for analysis.
Different Estimation method Behavior model; RL model Inverse Reinforcement Learning (IRL) compare parameter
Link based route choice model link length lanes right turn dummy
◆ Sequential Route Choice Model: Recursive Logit model (RL) (Fosgerau et al., 2013)
𝐵: set of links 𝜉: set of nodes absorbing state
𝑜 𝑒(𝑏)
At each current state 𝑙, a traveler chooses an action 𝑏 (next link). 𝜁𝑜 𝑏 : error term (i.i.d. Gumbel distribution) 𝜈: scale parameter 𝛾: discount rate
from the selected state 𝑏 to the destination link 𝑒
𝑊
𝑜 𝑒 𝑙 = 𝐹
max
𝑏∈𝐵(𝑙) 𝑤𝑜 𝑏 𝑙 + 𝜈𝜁𝑜 𝑏 + 𝛾𝑊 𝑜 𝑒(𝑏)
∀𝑙 ∈ 𝐵
The value function is defined by the Bellman equation (Bellman, 1957);
𝑜 𝑒 𝑏 𝑙 =
1 𝜈(𝑤𝑜 𝑏 𝑙 +𝛾𝑊
𝑜 𝑒(𝑏))
1 𝜈(𝑤𝑜 𝑏′ 𝑙 +𝛾𝑊
𝑜 𝑒(𝑏′))
Link choice probability
𝑡𝑢 𝑡𝑢+1
𝑏~𝜌(𝑡, 𝑏) 𝒬
𝑡𝑡′ 𝑏
= Pr{𝑡𝑢+1 = s′|𝑡𝑢 = s, 𝑏𝑢 = 𝑏} 𝑠
𝑢+1
𝑊𝜌 𝑡 = 𝐹𝜌
𝑙=0 ∞
𝛿𝑙𝑠𝑢+𝑙+1|𝑡𝑢 = 𝑡 = 𝐹𝜌 𝑠𝑢+1 + 𝛿
𝑙=0 ∞
𝛿𝑙𝑠𝑢+𝑙+2|𝑡𝑢 = 𝑡 =
𝑏
𝜌 𝑡, 𝑏
𝑡′
𝒬
𝑡𝑡′ 𝑏
ℛ𝑡𝑡′
𝑏 + 𝛿𝐹𝜌 𝑙=0 ∞
𝛿𝑙𝑠𝑢+𝑙+2|𝑡𝑢+1 = 𝑡′ =
𝑏
𝜌 𝑡, 𝑏
𝑡′
𝒬
𝑡𝑡′ 𝑏 ℛ𝑡𝑡′ 𝑏 + 𝛿𝑊𝜌 𝑡′
𝛿: discount rate (0 < 𝛿 ≤ 1)
ℛ𝑡𝑡′
𝑏 : expected reward
(= 𝐹{𝑠𝑢+1|𝑡𝑢 = s, 𝑏𝑢 = 𝑏, 𝑡𝑢+1 = s′}) ◆ Bellman equation
Transition state
◆ The estimation method : Recursive Logit model (RL) -NPL Parameter 𝜾 Value function Choice probability Likelihood Convergence test estimated Parameter 𝜾∗ Yes No
𝑢 = 𝜾𝑼𝒀
The algorithm for calculating fixed point of value function 𝑊 Con
est
𝑢
𝑊
𝑢 𝜾∗ − 𝑊 𝑢(𝜾) + 𝑢
𝜾𝑼 − 𝜾 < 𝜀
◆ The estimation method : Max entropy - Inversed Reinforced Learning (IRL) Parameter 𝜾 Reward 𝑠𝑢 Policy (𝑅 value) Likelihood 𝑀𝑀 estimated Parameter 𝜾∗ No Reinforced Learning
𝑢 = 𝜾𝑼𝒀
𝜾
𝑢 + 𝛿𝑅𝑢+1
Convergence test Yes
◆ IRL estimation (car)
Variables Parameters t-Value Link Length
Right-Turn
Lanes
L(0)
LL
Rho-Square 0.46 Adjusted Rho-Square 0.46
𝛾 = 0.47 (given) ◆ RL estimation (car)
Variables Parameters t-Value Link Length
Right-Turn
Lanes 0.37 2.76** L(0)
LL
Rho-Square 0.03 Adjusted Rho-Square 0.02
𝛾 = 0.47 (given)
◆ Recursive Logit estimation (bicycle)
Variables Parameters t-Value Link Length
Right-Turn
Car Traffic
β 0.00 15.15** L(0)
LL
Rho-Square 0.06 Adjusted Rho-Square 0.06
Car traffic Car Assignment
Bicycle Assignment
Network Policy 𝐻 = 𝑚𝑗𝑜𝑙, 𝑜𝑝𝑒𝑓, 𝑚𝑏𝑜𝑓
←Bicycle traffic
To decide the policy by calculating the fixed point of demand of cars and bicycles Policy change Demand change Variables is changed Consumer surplus
Upper Problem: traffic network
(pedestrian/bicycle only) traffic volume
Lower Problem: route choice behavior Car Bicycle
Assign each OD volume
network Different Estimation method Behavior model; RL model Inverse Reinforcement Learning (IRL) compare parameter
Link based route choice model link length lanes right turn dummy