Prediction of Traffic Patterns in Bike Sharing Systems: 🚲
Translated title
Prediction of Traffic Patterns in Bike Sharing Systems
Authors
Weinreich, Nicolai André ; van Diepen, Daniel Bernard
Term
4. semester
Education
Publication year
2022
Submitted on
2022-06-03
Pages
183
Abstract
Denne afhandling undersøger, hvordan folk bruger dock-baserede bycykelsystemer, for at gruppere stationer efter deres daglige aktivitet og forudsige disse mønstre ud fra forhold i stationernes nærområder. Målet er at kunne forudsige stationers gennemsnitlige daglige trafikmønstre på baggrund af rumlige faktorer i deres opland. Analysen bygger på rejsedata fra systemer i New York City, Chicago, Washington DC, Boston, London, Helsinki, Oslo og Madrid samt supplerende data uden for systemerne. Stationerne bliver grupperet efter formen på deres gennemsnitlige døgnprofiler ved hjælp af forskellige klyngealgoritmer. K-means-klyngedannelse med fem klynger gav klart adskilte typer af trafikmønstre. Disse stationstyper blev derefter sat i relation til eksterne rumlige faktorer via en logistisk regressionsmodel (en udbredt statistisk metode, der estimerer sandsynligheden for, at noget tilhører en bestemt kategori). På tværs af alle byer findes en stærk sammenhæng mellem stationstype og rumlige faktorer, og forskelle mellem byernes modeller hænger sammen med variationer i pendlerkultur. Den gennemsnitlige efterspørgsel ved hver station modelleres med en generaliseret lineær model med log-link (en standardmetode til at beskrive ikke-negative udfald). Sammen med den logistiske model gør det det muligt at forudsige gennemsnitlige trafikmønstre med rimelig præcision. Endelig vises en praktisk anvendelse i et casestudie af udvidelsen af Citi Bike i efteråret 2019, hvor efterspørgselsmodellen bruges til at optimere placeringen af nye stationer.
This thesis examines how people use dock-based bike-sharing systems to group stations by their daily activity and to predict those patterns from characteristics of the surrounding area. The goal is to predict stations’ average daily traffic patterns based on spatial factors in their service areas. The analysis uses trip data from systems in New York City, Chicago, Washington DC, Boston, London, Helsinki, Oslo, and Madrid, together with external data. Stations are clustered by the shape of their average daily traffic profiles using several algorithms. K-means clustering with five clusters produced clearly distinct types of traffic patterns. These station types are then linked to external spatial factors using a logistic regression model (a common statistical method that estimates the probability of belonging to a category). Across all cities, there is a strong relationship between station type and spatial factors, and differences between city models align with differences in commuting culture. Average demand at each station is modeled with a generalized linear model with a logarithmic link (a standard approach for non-negative outcomes). Together with the logistic model, this makes it possible to predict average traffic patterns with reasonable precision. Finally, a case study of the Citi Bike expansion in autumn 2019 demonstrates an application: the demand model is used to optimize where new stations are placed.
[This summary has been rewritten with the help of AI based on the project's original abstract]
Keywords
Bicycle ; Bike ; Cycle ; Sharing ; Bike Share ; Bike Share System ; Clustering ; Statistical Modelling ; Regression ; Prediction ; Planning ; Mobility ; SDG 11
Documents