Spatio-Temporal Forecasts for Bike Availability in Dockless Bike Sharing Systems
Abstract
Keywords
Acknowledgements
Abbreviations
1
Introduction
1.1
Context
1.2
Objective
1.3
Related work
1.3.1
Forecasting in station-based systems
1.3.2
Forecasting in dockless systems
1.4
Approach
1.5
Outline
2
Theoretical background
2.1
Time series definition
2.2
Time series characteristics
2.2.1
Autocorrelation
2.2.2
Stationarity
2.2.3
Spectral entropy
2.3
Time series components
2.3.1
Definitions
2.3.2
Classical decomposition
2.3.3
STL decomposition
2.4
Time series forecasting
2.4.1
Forecasting models
2.4.2
ARIMA
2.4.3
Naïve forecasts
2.4.4
Seasonal forecasts
2.5
Time series clustering
2.5.1
Dissimilarity measures
2.5.2
Hierarchical clustering
2.5.3
Spatial time series clustering
3
System architecture
3.1
Overall design
3.2
Software
3.3
System area
3.4
Database
3.4.1
Distance data
3.4.2
Usage data
3.5
Forecast request
3.6
Cluster loop
3.7
Model loop
3.8
Forecast loop
4
Data and experimental design
4.1
Data source
4.2
Data retrieval
4.2.1
Distance data
4.2.2
Usage data
4.3
Experimental design
4.3.1
Training and test periods
4.3.2
Additional software
5
Results and discussion
5.1
Clustering
5.2
Model building
5.3
Forecasting
5.4
Limitations and recommendations
5.4.1
Limits of forecastability
5.4.2
Exogenous variables
5.4.3
Residual distributions and prediction intervals
5.4.4
GPS accuracy
6
Conclusion
Appendix
A
Code
B
Models
B.1
Bayview model point
B.2
Downtown model point
B.3
Residential model point
B.4
Presidio model point
References
Published with bookdown
References