Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE credits
Over the past few years, many cities have witnessed the increasing popularity of cycling,
especially among ordinary commuters. Accordingly, there has also been a fast growing
demand for the knowledge of cycling performance as well as cyclist behavior, which can be
valuable for both traffic planners and policy makers when it comes to the bicycle-related
issues. The aim of this study, hence, is to investigate the cycling performance in detail
and to further develop proper models which can be implemented in the microscopic cycling
The study was initiated with data collection in the summer of 2013 in Stockholm. A
number of commuter cyclists were recruited and then provided with GPS devices to record
their daily cycling trips. The GPS devices were portable but qualified enough to measure
cyclists’ position, speed and altitude with a time interval of one second. Before the winter,
around 100 natural cycling trips made in the urban area of Stockholm were collected and
a database was later established to manage the raw data. Prior to the data analysis, measurement
noise cancellation and profile smoothing were performed by implementing multiple
processing approaches, including the robust locally weight regression and the Kalman filtering.
A cycling regime which separates the cyclist behavior into three different kinds
(acceleration, deceleration and cruising) was constructed based on the data observation.
According to this regime, a normal cyclist should always endeavor to achieve and maintain
a desired speed which varies depending on a number of factors, such as the cyclist’s own
demographics and the road grade. If a cyclist’s present speed was not corresponding to
her present desired speed, she would accelerate or decelerate immediately. Based on this
assumption, the GPS data were classified into three parts, including dedicated datasets for
acceleration profiles, deceleration profiles and cruising profiles. The profiles were analyzed
statistically and some significant cycling characteristics were founded. Moreover, mathematical
models were formulated to describe cyclists’ acceleration and deceleration behavior.
The models were further estimated using the maximum likelihood estimator and evaluated
by several goodness-of-fit measures.
2014. , 76 p.