City Science successfully secured funding from Innovate UK to develop a novel suite of tools for automating transport model development. This commission was carried out in collaboration with the University of Exeter and AI specialists Sustainicity. The study aims to discover to what extent Automation and AI can expediate the development, calibration and validation of transport traffic models, reducing both cost and delivery time of models.
Overview
Traffic model development is currently a high-cost, high-effort activity: For traffic model development in particular, the demand model depends heavily on third party information such as mobile network data, with there being several known limits in the industry such as the need for multiple data sets, cost and granularity of data. AI algorithms can automate the processing and integration of diverse transportation data sources, including traffic flow data, LSOAs, MSOAs, OpenStreetMap, public transport schedules, and National Travel Surveys; streamlining the development of new transport traffic models.
City Science Response
The primary objective of this commission was to demonstrate the feasibility of automating the development of travel demand matrices by creating comprehensive data pipelines. The key innovation lies in the utilisation of a pivotal AI system developed by City Science and the University of Exeter to address the limitations of MND (e.g., short-distance trips). Our goal was to integrate this AI system into an end-to-end automated solution to enhance validation accuracy.
Throughout the project, we will leverage the latest advancements in automated modelling drawn from the existing literature. Additionally, we will conduct extensive user engagement activities led by Sustainicity to ensure that user needs are effectively incorporated into the solution development process.
The study was carried out according to the following 6 stages:
- Literature review – Undertakes a detailed literature review and consolidates/updates core elements associated with AI in transport modelling.
- Automated demand model – Data preparation, algorithm selection and design, developing the automated demand model.
- AI training – Model training, validation, evaluation and refining the algorithms.
- End-to-end integrating – Data exchange and integration of the AI model into operational systems. Testing and optimisation are imperative in this stage to verify the model integrates and performs as intended.
- User needs research – Identifying user groups, stakeholder engagement, use case scenarios and user interface.
- Interim & final report – Summary Report consolidating key study findings underpinned by Technical Appendices
Outcomes
To demonstrate the feasibility of automating the travel demand step using AI, we developed the end-to-end data pipelines including multi-modal demand, including key parameters based on identified open sources for these data. This means that input data, which was previously provided by manually built models, can then be used to train demand models, validating the accuracy and completing the integration of the demand model into the end-to-end automated solution. From this, user testing through a demonstration model was undertaken in order to validate the efficiency of the solution and the market proposition.
Next Steps
- Improve accuracy.
- Continue engagement with potential customers.
- Demonstrate effectiveness/accuracy to customers.
- Integrate into wider tools/services.
- Leverage efficiencies.
- Apply in international application.