Analysing and predicting user demand in bike-sharing system
Visualitza/Obre
Estadístiques de LA Referencia / Recolecta
Inclou dades d'ús des de 2022
Cita com:
hdl:2117/192076
Tipus de documentProjecte Final de Màster Oficial
Data2020-01-29
Condicions d'accésAccés obert
Tots els drets reservats. Aquesta obra està protegida pels drets de propietat intel·lectual i
industrial corresponents. Sense perjudici de les exempcions legals existents, queda prohibida la seva
reproducció, distribució, comunicació pública o transformació sense l'autorització del titular dels drets
Abstract
Shared transport is an economical and sustainable mode of urban mobility.
It occupies very important position in the mobility market nowadays.
Given the growing market size and importance of Bike-Sharing System, we
analysed the urban mobility patterns based on 2 year real data from NYC
Citibike Static Bike-Sharing System dataset which contains users' mobility
information. We performed meteorological analysis, demographic analysis,
temporal and spatial analysis on the dataset to explore the correlation
between variables. Thus it gives us a better understanding about urban
mobility and the possibility to improve some mobility issues.
One of the issues we discussed in this thesis is the Bike Rebalancing
Problem. Given the demand-driven nature of shared transport systems,
availability of its infrastructure heavily depends on users' mobility patterns.
An unbalanced situation where the number of available vehicles cannot meet
user demand will reduce the e ciency of vehicle-sharing system. We designed
several machine learning approaches to predict the user demand using
historical trip data, weather data and geographic data.
All models were evaluated under 150 comparisons with di erent hyperparameter
settings. The Long Short-Term Memory (LSTM) with encoderdecoder
architecture which feeds with past sequential feature and future
sequential feature, outperform all other models built in this thesis. It can
successfully predict the user demand of next future month with much better
accuracy than the baseline model. By using data of last 12 months, the
LSTM can lower down 24.8% of RMSE and MAE comparing with historical
average.
Keywords
TitulacióMÀSTER UNIVERSITARI EN INNOVACIÓ I RECERCA EN INFORMÀTICA (Pla 2012)
Col·leccions
Fitxers | Descripció | Mida | Format | Visualitza |
---|---|---|---|---|
143474.pdf | 51,90Mb | Visualitza/Obre |