Understanding the dynamics of demand over space and time is essential for many applications across the entire transportation domain. Since we cannot directly observe the demand yet, we need to estimate the demand from whatever data and information are available. Unfortunately, the Dynamic Origin-Destination Estimation (DODE) is still one of the most challenging problems in transportation to date. The complexity in estimating dynamic OD matrices is that the problem is severely under-determined, a fact recognized since the early days of OD estimation and is emphasized in virtually all contemporary OD research.
Dynamic Traffic Assignment (DTA) models are a reliable tool to replicate complex traffic conditions, monitor and make short-term forecasts of the flow and speed of the network. The most crucial input in both online and offline dynamic traffic assignment is the OD demand. The success of the DTA applications relies on the quality of this fundamental input. Recent advances in real-time DTA applications in large networks call for robust and efficient methodologies for online OD demand estimation and short-term demand predictions.
With the rapid development of mobile-internet technologies, on-demand mobility services have become increasingly popular and largely shaped the way people travel. Therefore, demand prediction is a fundamental component in supply-demand management systems of ride-sourcing platforms. With accurate short-term demand prediction, the platforms make precise decisions on real-time matching and idle vehicle reallocation.
The main objective of this research project is to use Big Data resources, such as vehicle positions, to estimate and forecast demand in the form of Dynamic Origin-Destination matrices to be used as an input for DTA. In addition, gathering this data with machine learning techniques and artificial intelligence to prevailing demand patterns can be used for online demand forecasting, specifically predicting ride-sourcing demand.
Innovation is one of the strengths of the proposed research project. This research project focuses on developing a new methodology for Dynamic OD matrix estimation in large-scale congested urban networks, considering capacity constraints and hypercritical conditions, willing to overcome the limitations of other approaches. From the brief overview given in the state of the Art, it is clear that to solve a severely underdetermined problem like OD matrix estimation. One has to make assumptions related to using the data, the models that relate these data to the unobserved OD flows, and the solution methodologies to derive the OD flows from the combination of those data and models. This research project address how the information about congestion, either obtained from the underlying Dynamic Traffic Assignment method or directly measured, can be used within the DODE problem as a data source added to link flows. To this end, an analytical method is proposed to obtain a linear approximation of the dynamic network loading problem to be used within the DODE. Furthermore, to better reproduce the effects of hypercritical conditions and capacity constraints, a constant term is introduced, while the assignment matrix reflects the local sensitivity to demand variations. One of the key findings of this work is that, in the presence of queues, demand flow variations do not propagate along with the network accordingly with travel times; in particular, when a queue is present, adding some flow on the entry section of a link results in additional flow on the exit section when the queue vanishes.
Considering the popularity of ride-sharing services and their commercial value, there is a great interest but only limited knowledge insofar on their demand patterns. Notwithstanding, recent studies examined the impacts of ride-sourcing on the taxi market, passengers response to dynamic pricing. This research project proposes a methodology aims to identify meaningful spatial and temporal clusters in demand for ride-sourcing services to understand better the underlying patterns and support planners and service providers. Service providers can deploy proactive pricing and fleet management strategies to anticipate recurrent demand characteristics and thus contribute to the service provision. It can reduce passengers waiting and in-vehicle times. The approach and techniques proposed in this research can be transferred to other contexts and locations. It can be used to examine the relationship with the service offered by alternative modes, primarily the privately-owned car and fixed public transport, and their impact on the demand for ride-sourcing services.