What impact could AI have on the rail sector in the coming decade?
By Henry Metcalf, Graduate Engineer
Our graduates are a constant source of pride and inspiration to us at PBA, routinely taking part in personal and professional development, industry initiatives, and finding innovative solutions to age-old problems. One such graduate engineer, Henry Metcalf, recently took part in the New Civil Engineer’s graduate competition ‘What impact could Artificial Intelligence (AI) I have on the airports, rail or highways sector in the coming decade?’. Choosing to focus on possibilities in the rail sector, Henry’s entry, below, combines the use of AI to improve our understanding of passenger movements and to roster trains and assign train paths to match this demand more efficiently.
Within rail, AI provides the opportunity for dynamic and demand responsive train services to be provided in real-time. This impact would vastly improve train services for customers.
The idea is simple:
1) Use AI to continuously analyse the vast amounts of train ticket sales data in real-time.
2) AI uses this analysis to immediately alter train timetables and train paths to provide the most efficient rail service possible.
In the first step, artificial intelligence learns the travel habits and preferences of UK rail travellers by analysing ticket sales information, gateline use, station car park usage, station retail sales, reports of train and station crowding and other metrics in real-time. The AI computer uses this to learn the complex patterns of ticket sales and predict train usage based on: weather, economic performance in different regions, sporting events and a myriad of other factors that it would have determined to be linked to train usage. Thus, AI’s most powerful tool, the ability to learn from vast sets of data can be harnessed to understand how patrons want to use the rail network like never before. Over time the information it monitors would grow -providing more data for analysis- and creating the perfect environment for an AI computer to constantly refine and improve itself.
In the second step, the AI computer produces a new railway timetable to match rail capacity with the expected demand found in the first step. Its algorithms would roster trains, crews and maintenance activities on a daily basis in the most efficient way possible to minimize costs to the rail industry (and hence consumers) whilst simultaneously improving services for customers. AI is best placed to do this because it can learn how the complex network of rail tracks not just operates but responds to service disruption, maintenance issues and unexpected occurrences – such as adverse weather, point failures or trespassers.
The UK rail network is a highly intricate array of gauge limits, speed restrictions, junctions, electrification and many other assets. It has 2,500 stations and over 16 million different train fares giving rise to more than 250 million possible journey fare combinations. Given this scale it is not feasible for humans to alter existing train paths in response to sudden changes to rail patronage or to predict how rail usage will change due to impending events such as: changing weather conditions, parking availability, shopping events and so forth and simultaneously taking into account how sections of track are performing at that particular time. AI, however, can achieve this.
Thus, AI can learn rail passengers’ travel habits and apply this learning to match the provision of trains on the network to user demand. If realised this will allow for a demand-responsive railway that dynamically adjusts to user demand in real-time with a minimum inefficiency.