Deep learning Anticipated Urban Mobility peaks
Duration: April 2021 - April 2024
The COVID-19 crisis has led to profound changes in mobility patterns, including all aspects of life. The changes were in some cases sudden and in some occurred over an adjustment period of time. Triggers of the change can be found in soft and hard factors. With the notion “hard” we refer to regulations and guidelines given by various authorities on local and national level. “Soft” refers to information spread within society including social media trends and other false and true information spreading fast. The goal of this project is to relate the onset of such trends with mobility patterns. We utilize data fusion of mobility data, social media data and product price data as input for deep learning methods. The data are further input for simulation of urban traffic patterns to show crowding and congestion effects, as well as the disruption in typical daily mobility patterns. This joint supply simulation and learning of demand dynamics will lead to a comprehensive evaluation of the effectiveness of specific policy measures to avoid undesired travel peaks and demand peaks for specific facilities or products. The study therefore uses data fusion and artificial intelligence to promote resilient and safe societies. Even though we focus on data observed during the still ongoing Covid-19 crisis we aim to develop a framework with general applicability to larger disruptions.
Spreading the peak has been one of the most used phrases to describe policies dealing with the COVID crisis. Whereas this is mostly applied to the infection numbers, it can also be applied to mobility as crowding in public transport as well as in public places increases infection risks. In some cases the risk has in fact been amplified through additional trips and long queues at shops that are selling items that are likely in shortage during the main phase of the disaster. To avoid demand peaks therefore early identification of changes in behavioural patterns and system bottlenecks is crucial. The overall objective of this project is accordingly to develop a methodology to identify and predict triggers of behavioural change in the context of disruptive events.
The consortium will utilise the opportunities that have been arising through the availability of large, new datasets. Some of these such as google trend data and data from social media can be freely collected or extracted. Other data sources such as telecommunication data that provide aggregated information on movements and commodity purchase data are powerful to explain behavioural change and will further be utilised. The first main contribution and outcome of this project is therefore a framework to combine such data to jointly describe mobility patterns. A second outcome is the development of an artificial intelligence platform to learn causal relations within and between these data sets. A third outcome is then a description of how these findings can be used to facilitate a “dashboard” for city planners that show the effectiveness of different policies for mobility peak smoothing.
Please refer to the Project GitHub Page for more details.
Lyu, C., Lu, Q. L., Wu, X., & Antoniou, C. (2024). Tucker factorization-based tensor completion for robust traffic data imputation. Transportation Research Part C: Emerging Technologies, 160, 104502.
Lu, Q. L., Qurashi, M., & Antoniou, C. (2024). A two-stage stochastic programming approach for dynamic OD estimation using LBSN data. Transportation Research Part C: Emerging Technologies, 158, 104460.
Lu, Q. L., Mahajan, V., Lyu, C., & Antoniou, C. (2024). Analyzing the impact of fare-free public transport policies on crowding patterns at stations using crowdsensing data. Transportation Research Part A: Policy and Practice, 179, 103944.
Lu, Q. L., Qurashi, M., & Antoniou, C. (2023). Simulation-based policy analysis: The case of urban speed limits. Transportation Research Part A: Policy and Practice, 175, 103754.
Yang, N., Lu, Q. L., Lyu, C., & Antoniou, C. (2024). Transfer learning for transportation system resilience patterns prediction using floating car data. In 103th TRB Annual Meeting 2024.
Lu, Q. L., Sun, W., Dai, J., Schmöcker, J.D., & Antoniou, C. (2023). Surrogate modeling for recovery measure optimization to improve traffic resilience. In the 9th International Symposium on Transport Network Resilience 2023.
Lu, Q. L., Qurashi, M., & Antoniou, C. (2023). A Two-Stage Stochastic Programming approach for Dynamic OD Estimation. In 102th TRB Annual Meeting 2023.
Lu, Q. L., Qurashi, M., & Antoniou, C. (2022). A Stochastic Programming Method for OD Estimation Using LBSN Check-In Data. In 4th Symposium on Management of Future Motorway and Urban Traffic Systems.