Machine Learning is an increasingly common aspect of modern life. Using the power of modern computing, it enables vast volumes of data to be harnessed, delivering valuable insight and improvements to everyday products or services, including fraud detection in online banking and tailored recommendations within streaming, shopping, and social media platforms.
Given its power, there is little doubt that Machine Learning will play a transformative role in shaping the public transport networks of tomorrow. Indeed, there is an expectation that it will pave the way to an entirely new generation of bus prediction algorithms, ensuring truly accurate passenger information, whatever the level of disruption on the roads.
Developments in this area have already begun: Trapeze is currently working on a project in which Machine Learning is being used to improve the reliability of public transport information systems.
Machine Learning technology enables us to bring together far more data points, thereby continually increasing the accuracy of real-time prediction algorithms. In this way we aim to ensure greater reliability of bus services, making public transport more attractive, and ultimately growing ridership.
The Quest for Prediction Accuracy
Bus arrival and journey time prediction accuracy is notoriously challenging due to the volume and complexity of factors involved.
At Trapeze, we have vast ITS experience and already use powerful algorithms to calculate prediction times. We are proud of our record and believe few organisations come close to matching us in this area.
Our ITS system has consistently delivered against challenging prediction accuracy KPIs in London and Singapore, despite their notoriously complex and busy environments with thousands of incidents occurring daily.
However, we recognise that passengers don’t care about complexity, the challenges of near-constant disruption, or the quality of our algorithm and team. From the public’s perspective, any time a bus fails to arrive when expected, it denies them the service they expect and demand.
Distilling the Complexity
Delivery of truly accurate predictions requires vast quantities of data. The Machine Learning solution which Trapeze is developing begins by meshing together current vehicle locations with historical measurements of buses along the same patterns. However, a range of additional input factors are required to harness the power of Machine Learning, including – but not limited to – the following:
Weather
Weather affects vehicle movements, so we blend in current weather conditions and use a Machine Learning algorithm to understand how this impacts predictions and then allocate ‘weightings’ to a range of scenarios.
Traffic
Current traffic conditions must also be considered. Average vehicle speed data can be sourced from mobile phones and satellite navigation systems but cannot be blindly relied upon. For example, the presence or otherwise of bus lanes is a significant factor directing whether a bus will be able to move through a busy city.
Traffic Lights
Traffic lights are perhaps the most significant impact on bus movements. Traffic light priority can play an essential role in keeping buses moving, but there is an additional challenge here because even when such a system is in place, historically, there has been no way to identify whether requests to change traffic lights were received and acted upon.
Our development team is therefore exploring ways to understand the impact of traffic light priority requests and responses on bus speeds, with the intention of building this into the algorithm.
Once all the data, including that outlined above, has been identified and pulled together, it is then the task of the Machine Learning algorithm to identify the impact of the full range of factors on each specific journey and use this to refine arrival and journey prediction times in real-time.
Mitigating Disruption
While Machine Learning can increase overall prediction accuracy, it is, of course, unable to support in instances of unforeseeable disruption. For example, an accident or a vehicle stopped in a bus lane. Or is it?
At Trapeze, we are also exploring the potential for Machine Learning to support decision making during unforeseen incidents and adjust predictions accordingly.
For example, in the event of a disruption that causes a service diversion, ITS Service Controllers could access previously defined and tested alternative routes (or ‘paths’), thereby enabling them to implement proven backup plans quickly. The recommended path will ensure minimal impact on passengers, maintaining maximum possible service quality on the route.
Interestingly, this is similar to what Trapeze’s rail business does for rail operators – creating pre-modelled plans that can be implemented during disruption.
In this way, when disruption strikes, train operator staff can choose from a number of proven solutions that can be implemented almost at the touch of a button. As a result, passengers may not even notice the existence of an incident that would otherwise have been incredibly disruptive to both the operator and the travelling public.
Naturally, there tend to be many more variables on roads than on train tracks, which makes this task even more complex. This is where the Machine Learning algorithm comes into play, by making it possible to manage a far greater volume of data and being adaptive to the changes.
And finally, don’t be so sure that technology can’t do anything about incidents such as vehicles stopped in bus lanes: Trapeze has also begun working with another sister company, Taranto Systems, which manages congestion and red route enforcement in major cities, including London.
We are at an early stage, but we certainly see potential in connecting such systems. One possible integration relates to Taranto providing early warnings of potential issues on the road network, such as a vehicle blocking a route. This data could be immediately factored into passenger information systems, with issues impacting bus service delivery flagged for priority resolution.
Bringing Machine Learning to Life
It is not possible to effectively compare prediction accuracy between regions because they are largely directed by the innate level of ‘chaos’ within the locality. This is unrelated to its size: some major cities are relatively simple, making it easy to deliver very high levels of accuracy, while small towns can have immense complexity, resulting in significantly lower accuracy levels.
We believe that the adoption of Machine Learning will deliver a transformative impact on prediction accuracy in towns and cities with very high complexity and disruption levels and provide opportunities to fine-tune algorithms to meet each city’s requirements.
Machine Learning is also becoming increasingly important in an electric vehicle world, as the same technology will underpin predictions for battery range. Of course, this subject is also incredibly complex, with many factors to consider, including the number of passengers on board, use of heating and air conditioning, and more.
Trapeze’s development team is harnessing the power of Machine Learning for bus predictions right now.
We know that service reliability remains an essential priority for bus users. Through Machine Learning, we can deliver a step change in relation to the accuracy of predictions, elevating the passenger experience, further restoring trust in bus travel, and driving essential ridership recovery.
