Short Term Traffic Prediction for the UK Motorway Network Using an Artificial Neural Network

Short Term Traffic Prediction for the UK Motorway Network Using an Artificial Neural Network

Nominated for The Neil Mansfield Award


Carl Goves, Transport Systems Catapult


This paper presents the results of applying an artificial neural network to estimate traffic conditions 15 minutes into the future on a section of motorway within the UK.


To be able to predict reliably traffic conditions over the short term (15 minutes into the future) may help realise significant benefits including the potential to reduce congestion on a transport system. For example, it could enable more proactive operational demand management of the network and / or instil more confidence in users with regard to the time and reliability of their planned journeys if this information is well communicated. With the emergence of large datasets describing the use of a transport network comes the opportunity to test the effectiveness of pattern recognition techniques to solve complex, non-linear problems such as the one in question.

This paper presents the results of applying artificial intelligence, specifically artificial neural networks (ANNs), to estimate traffic conditions a short time into the future given current / historic traffic information. Specifically, data collected from Highways England’s Motorway Incident Detection and Automatic Signalling (MIDAS) system has been used. MIDAS is installed along parts of the UKs busiest motorways and monitors traffic conditions (traffic flow and speed) at both short geographical intervals (approximately every 500 metres where installed) and over small timestamps (every minute). For this study, 2014 data (comprising over 3.2 million records) was extracted from MIDAS for approximately 20km of the M60, M62 and M602 motorway near Manchester, UK and used to build a short term prediction model. Once processed, data from 92 traffic detectors was used to train a model capable of predicting traffic density at each detector site 15 minutes into the future. To reduce the complexity of the problem, the number of input dimensions to the model was successfully reduced using a form of ANN known as an autoencoder. The final model developed exhibits very good predictive power with 90% of all predictions within 2.6 veh/km/lane of observed values.

The approach adopted in this research is one that can be transferred to other parts of the UK motorway network where MIDAS is installed, and once trained, the application of an ANN is straightforward. Potentially an algorithm such as the one derived has multiple applications which could include refining predictions within intelligent transport systems (ITS) and / or enabling traffic controllers to take proactive decisions to mitigate the impacts of expected congestion. It could also be the engine behind a “traffic-cast” system which could provide the public with a forecast of expected traffic conditions. This could result in benefits (such as reduced congestion) on the transport system being realised as accessibility to more accurate information could encourage beneficial behavioural changes in users.

The applied research conducted as part of this study forms part of a rolling programme being conducted by the Transport Systems Catapult, the UK’s innovation centre for Intelligent Mobility.


Association for European Transport