NORTHMOST

Northmost stands the Northern Training Hub on the Mathematics Of Systems of
Transport. It is intended to be an informal set of meetings between researchers, and practitioners interested in mathematics and transportation systems, both inside and outside the northern powerhouse.

The Next Northmost Meeting

The next northmost meeting will take place at the School of Mathematics at the University of Manchester on 14th June 2018.

Speakers:

Abraham Nahr — Transport Systems Catapult

Richard Gibbens — University of Cambridge, Alan Turing Institute

Peter Heywood — University of Sheffield

Alex Jirat & Chris Talbot — Transport for Greater Manchester

Steven O’Hare — Mott Macdonald

Damon Wischik — University of Cambridge, Alan Turing Institute

Philip James — Urban Observatory Newcastle

Registration:

Registration is free. So we can arrange lunch, please register by email the website below with your name and institution:

northmost.meeting@gmail.com

Location:

Alan Turing Building — Frank Adams Room
School of Mathematics
The University of Manchester
Manchester
M13 9PL

Titles, Slides, and Abstracts:


Abraham Narh [SLIDES]

REAL-TIME TRAFFIC PREDICTIONS ON UK MOTORWAY BASED ON NEURAL NETWORK

The presentation gives insights on the potential use of artificial intelligence techniques such as Neural Network to forecast traffic density and speed in real time on the motorway in the UK.  This is based on the outcome of a research undertaken by TSC on behalf of Highways England to support the strategic management of the Strategic Road Network. In this research, we used data gathered from the MIDAS database over a period of one year, specifically from April 2015 to March 2016. The NN model was tested data from the last two weeks; the remaining data was used as follows – 80% of the data for training of the network and 20% for validation purposes. The results show that the density models gave a high correlation coefficient of 91-92% between the observed and predicted values, whilst the equivalent figures for speed is 65-91%.  In terms of the error margins, we found that 96-98% and 84-93% of 1,021 predictions for density and speed respectively were within an error margin of ±10 veh/km/lane.  The research outcomes prove the suitability of the use of Neural Network prediction algorithms to forecast traffic conditions.  The research output paves the way for the potential use of Neural Network algorithms linking into existing systems to support the proactive traffic management of the motorway network.


Alex Jirat & Chris Talbot [SLIDES]

Manchester as a Smart City: From the Bottom Up

Transport of Greater Manchester holds the responsibility for monitoring and managing the performance of its transport networks.  With Smart Cities and the Internet of Things playing an increasingly prominent role in the strategic planning of the future of the region, there is a growing demand for more robust and granular data.  By examining the existing functions of Transport for Greater Manchester, this presentation aims to answer the question: what does the future hold for Manchester and how can it continue its development and fully embrace its ambition of becoming a full integrated smart city?


Peter Heywood [SLIDES]

Improving Transport Simulation Performance using Graphics Processing Units

The scale, complexity and quantity of transport simulations is limited by the long run-times of simulation software. This talk will give insight into how many-core processing architectures, such as Graphics Processing Units (GPUs), can be applied to transport simulations. Significant performance improvements can be achieved, both reducing simulation runtime and enabling larger-scale simulations than possible with traditional CPU based software.


Steven O’Hare [SLIDES]

Transferring Link Data between Networks: An Algorithmic Approach

A common challenge that transport modelling practitioners encounter is that of how to transfer data between different network representations of the same transport network; for example, transferring Traffic Master journey time data from OS ITN base to a SATURN model representation. Frequently such representations do not overlap, making spatial approaches inaccurate, do not come ready packaged with correspondence lists. Standard practice is therefore manually transfer data for a small subset of links that of interest, e.g. that form routes for journey time validation. This presentation presents a technique that makes use of simple path searching algorithms to transfer data en masse between networks. The advantage of having data for a whole network rather than just part of it is that it makes it possible to calibrate model parameters, such as link cruise speeds and speed flow curves, and also (potentially) to imagine the automation of the whole network build process.


Philip James

Smart Cities, IoT and sensing in the real world.  Lessons from the frontline for Urban Monitoring


Richard Gibbens [SLIDES]

Data, modelling and inference in road transport networks

Future cities present a range of challenges for road transport modelling. In this talk we discuss the ways in which data collected by various types of sensor can, with suitable statistical and computational modelling techniques, help us tackle these challenges. We study UK road traffic data and explore a range of modelling and inference questions that arise from them. Loop detectors on the M25 London orbital motorway record speed and flow measurements at regularly spaced locations as well as the entry and exit lanes of
junctions. An exploratory study of this data helps us to better understand and quantify the nature of congestion on the road network. From a traveller’s perspective it is crucially important to understand the overall journey times and we look at statistical methods to improve our ability to predict journey times given access to both real-time and historical loop detector data. A second data source is gathered from GPS traces of buses within Cambridge, UK. With this data we have been able to describe the daily profile of journey times and the spread of congestion relating to incidents in the urban road network.


Damon Wischik [SLIDES]

Cities and fleets

How should cities respond to ride sharing fleets, last mile delivery fleets, and perhaps fleets of autonomous vehicles? We need to model the interplay between three types of agent, each with their own goals: users, city traffic management, and fleet operators. Fleet operators control sizable amounts of traffic in a city, they have rich real-time data feeds, and their algorithms can react in an instant by setting surge prices or by reassigning jobs. This can lead to problems e.g. gridlock on side streets allegedly due to Waze. It could also allow better solutions for managing traffic, thanks to better signalling and better options for shifting load. In this talk I will describe a simple user+fleet+city traffic model, and show perverse outcomes in the spirit of Braess’ paradox. I will suggest some ways that cities might harness the adaptability of fleets, to improve traffic management. Finally, I will describe the sort of visualisation and modelling capabilities that cities will need if they are to keep up with modern fleets and their teams of data scientists.


Program

Coffee and arrival: 10:00-10:30

Welcome: 10:30

Peter Heywood — 10:45

 

Steven O’Hare — 11:20

Abraham Narh — 11:55

Lunch  — 12:30–13:30

Richard Dolphin — 13:30

Richard Gibbens — 14:05

Coffee— 14:35 — 15:05

Philip James — 15:05

Damon Wischik– 15:40

Break— 16:15-

Panel Discussion — 17:00-17:30

Drink (optional): Go for a drink at Picadilly Tap.

Organizers:

Richard Connors (Institute for Transportation Studies, Leeds University)

Mike Smith (York University)

Neil Walton (Manchester University)

David Watling (Institute for Transportation Studies, Leeds University)

Sponsors:

resize.jpg

         Data Science Institute, The University of Manchester

 

the_alan_turing_institute_logo.png

The Alan Turing Institute