Mathematical Modelling of Pedestrian Route Choices in Urban Areas Using Revealed Preference GPS Data Eka Hintaran ATKINS European Transport Conference 2017 4 – 6 October
Methodology and Modelling approach Case study: Zürich Mathematical Modelling of Pedestrian Route Choices in Urban Areas Using Revealed Preference GPS Data Observed trips Non-chosen trips Route attributes Model Methodology and Modelling approach Case study: Zürich Estimation results Conclusion Work in practice
Introduction to Route Choice Modelling “Representing route choice behaviour consists in modelling the choice of a certain route within a set of alternative routes. A route choice model predicts the probability that any given path between Origin and Destination is selected to perform a trip, given a transportation network and an OD-pair “ (Bierlaire & Frejinger, 2008) Observed trips Non-chosen trips Route attributes Model
Aim of this route choice model: Development of a pedestrian route choice model estimated from revealed preference GPS data Observed trips Non-chosen trips Route attributes Model Aim of this route choice model: To understand how pedestrians choose their routes within an urban area and which quantitative environmental street factors influence their route choice behaviour To use this knowledge to predict future route choices of pedestrians in urban areas To find out if pedestrian behaviour could be captured in a route choice model
Observed trips Non-chosen trips Route attributes Model
Methodology and Modelling approach (1) Modelling approach within the Random Utility Maximization (RUM) framework, described in Discrete Choice Models (DCM) DCMs predict choices between a set of finite distinct alternatives, based on utility maximization, assuming that pedestrians make a subjective rational choice between a finite number of choice options. Observed trips Non-chosen trips Route attributes Model Discrete Choice Models: Estimates the probability P for each alternative i of being chosen by individual n from a choice set when the individual maximizes his utility
Methodology and Modelling approach (2) Model Structure: Path-size Logit Model (Ben-Akiva & Bierlaire, 1999) Case study: Zürich Revealed preference experiment Collecting data Processing of GPS data Model development Calculation of non-chosen alternatives Calculation of route attributes Calculation of overlap between routes Model estimation Final results Observed trips Non-chosen trips Route attributes Model
Observed trips Non-chosen trips Route attributes Description of overlap
Observed routes Non-chosen routes Route attributes Calculation of overlap Observed trips Non-chosen trips Route attributes Description of overlap
Observed routes Trips: GPS data collected by trackers Only trips in Zürich area 51 participants 3053 stages Network: Data extracted from OpenStreetMap Elevation model Observed routes Non-chosen routes Route attributes Calculation of overlap
Observed routes Non-chosen routes Route attributes Description of overlap Observed trips Non-chose trips Route attributes Model
Observed routes Non-chosen routes Route attributes Description of overlap Observed trips Non-chose trips Route attributes Model
GPS data processing Filtering (errors) Smoothing (position errors caused by satellite issues) Calculating speed and acceleration Detection of stop-points and stages Map-matching (OpenStreetMap) Only walking trips: filtering by mode (speed) 51 participants and 580 stages Observed routes Non-chosen routes Route attributes Calculation of overlap Observed trips Non-chose trips Route attributes Model
Observed routes Non-chosen routes Route attributes Description of overlap Observed trips Non-chose trips Route attributes Model
Calculation of non-chosen routes Input: Observed routes (Origin – Destination) Network Choice set generation algorithm developed based on: Breadth-First-Search on Link Elimination method Observed routes Non-chosen routes Route attributes Calculation of overlap Observed routes Non-chosen routes Route attributes Description of overlap Observed trips Non-chose trips Route attributes Model
Choice sets of 6 alternatives (1 is the chosen) Observed routes Non-chosen routes Route attributes Calculation of overlap Observed trips Non-chose trips Route attributes Model Observed route Output: Choice sets of 6 alternatives (1 is the chosen) Choice sets of 7 alternatives (chosen is not generated) Alternative routes
Calculation of Route Attributes Input: Choice sets (Observed + Non chosen Routes) Network (including characteristics) Elevation model Output: Choice set (routes) with calculated route attributes Distance [km] Rise and fall characteristics [m/100m] Road type fraction [0-1] Observed routes Non-chosen routes Route attributes Calculation of overlap Observed trips Non-chose trips Route attributes Model
Calculation of Overlap Input: Choice sets (Observed + Non chosen Routes) Network Output: Path-Size attribute [0-1] for every route Observed routes Non-chosen routes Route attributes Calculation of overlap Observed trips Non-chose trips Route attributes Model 1 : 2 : 3 1/3 : 1/3 : 1/3 (1 + 2) : 3 (1/2) : 1/2
Based on Utility Maximization theory Model Development Based on Utility Maximization theory Path-Size Logit model Accounts for overlap Already applied to active modes (walk, cycle) Route attributes: Trip length (Distance) Maximum Rise (RiseMax) Road Type (WalkOnly, WalkSafe, WalkAll) Path-Size attributes (overlap: PS1, PS2 and PSC) Observed routes Non-chosen routes Route attributes Description of overlap Observed trips Non-chose trips Route attributes Model
Model estimation/calculation Observed trips Non-chose trips Route attributes Model Model: Path-Size Logit Rho-square 0.297 Adjusted rho-square 0.285 Estimated in BIOGEME (Bierlaire, 2003) Main findings: Maximum rise largest influence Overlap (PS) second Road type Walk & Bike third Trip length is not consistent Parameters Value Rob. St err t-test BETA_ACLASS -0.749 0.314 -2.38 BETA_BCLASS 1.26 0.339 3.73 BETA_CCLASS 0.909 0.367 2.48 BETA_DCLASS -1.42 0.327 -4.35 BETA_RISEMAX -30.6 7.81 -3.91 BETA_WALKONLY - 0.844 -0.53 BETA_WALKSAFE 1.82 0.811 2.24 BETA_WALKALL BETA_Log(PS2DIST) -2.37 0.829 -2.86
Conclusion Aim: To understand how pedestrians choose their routes within an urban area and which quantitative environmental street factors influence their route choice behaviour To use this knowledge to predict future route choices of pedestrians in urban areas To find out if pedestrian behaviour could be captured in a route choice model In this case study: Maximum Rise dominant, Overlap second, Road type (Walk & Bike) third Trip length has influence, but result is not consistent General Conclusion Pedestrian behaviour proves to be rational and can be described using Discrete Choice Models within Random Utility Maximization framework Therefore this methodology can be used in other case studies as well Observed trips Non-chose trips Route attributes Model
Added value of theories and results in practice As support in policy-making and design process Plan and design new infrastructures and pedestrian facilities Inform designers and area operators Determine optimal pedestrian environments new walkability guidelines Plan, organise and manage large events Observed trips Non-chose trips Route attributes Model
Observed trips Non-chose trips Route attributes Model
Ideas for Oxford Circus and other Public Realm projects Methodology to predict future route choices of pedestrians in the area Observed trips Non-chose trips Route attributes Model Data processing (Wifi data) Extract routes from Wifi data and calculate alternative routes based on these routes Use these data to predict the probability that a given path is chosen between Origin and Destination Apply split to Oxford Circus Station totals, Crossrail totals, Forecasted totals Demand as input for static assessment and dynamic modelling (Legion) Knowledge about pedestrian’s route choices Determine most influencing factors to inform designers (wayfinding, public realm, street) and operators (TfL)
Questions Observed trips Non-chose trips Route attributes Model