Bicycle Travel Demand Estimation Method

Abstract

Bicycle travel demand analysis remains relatively underexplored, yet it is becoming increasingly important for urban and campus planning. In the Netherlands—particularly within the TU Delft community—cycling is deeply embedded in daily life, creating a strong demand for supportive infrastructure. While congestion may not pose the same threat to bicycle transport as it does to motorized traffic, maintaining the performance and safety of the bicycle network is essential. A busy cycling network may not always be visible at a glance, yet it can present safety risks.

The primary aim of this research is to identify a modeling process and specifications that are compatible with the available data, while laying the groundwork for future improvements to the bicycle network, especially TU Delft Campus. This will help ensure the system remains adaptable and relevant for long-term planning.

To identify an appropriate modeling approach, an exploratory analysis of the data was conducted. A clear pattern emerged in bicycle traffic, characterized by short-interval fluctuations corresponding closely with lecture schedules. An additional notable observation is the occurrence of an average peak in bicycle traffic during midday. These findings support a dynamic analysis approach with a 5-minute interval.

Moreover, the model incorporates specialized variables defined by the study’s scope, focusing on trip generation and trip distribution within the established four-step modeling framework, specifically tailored for Origin-Destination (OD) matrix estimation in transportation engineering.

For trip generation, linear regression coupled with backward stepwise elimination via the Ordinary Least Squares (OLS) method was employed to identify significant predictors. For trip distribution, the Iterative Proportional Fitting (IPF) method was utilized. This approach was chosen based on the assumption that impedance is minimal for short-distance travel, a scenario particularly relevant within the TU Delft campus context.

Ultimately, this methodology provides a flexible and responsive framework tailored to the specific transportation dynamics at TU Delft, producing valuable insights for optimizing bicycle network planning.

The developed model is relatively simple but exhibits several shortcomings. One significant limitation is related to data collection, as the available data lack the temporal resolution necessary to fully capture the dynamic travel patterns targeted by the model. Additionally, the linear regression approach used for modeling trip production and attraction yielded unsatisfactory results, with $R^2$ values below 0.5. Another issue is potential underfitting, as indicated by the improved explanatory power of the model when trained on smaller datasets. Validation using RMSE and comparative plots of modeled versus actual flows further confirms that substantial improvement is needed in the model’s reliability and predictive capability.

The trip distribution process, conducted using the Iterative Proportional Fitting (IPF) method, reveals additional areas for improvement. The OD matrix underestimated total production by two bicycles in a 5-minute interval. Although seemingly small, this discrepancy underscores the necessity for more robust input data and methodological refinements. Additionally, direct validation of the OD matrix is crucial to enhance accuracy and reliability in representing actual travel flows.

Despite the shortcomings, the framework provides a balance between interpretability and flexibility, enabling both accurate representation of observed travel behavior and ease of scenario testing—making it a practical tool for supporting data-driven mobility planning and policy evaluation on campus.