Decarbonising railways with artificial intelligence

An innovation called Rail Decarb enables transport authorities to quickly and cost-efficiently appraise railway systems for a net zero future.

Retrofitting a railway to reduce its carbon emissions is a difficult task but one that is important for national – and ultimately global – emissions reduction. Rail clients recognise their role in achieving net-zero.

Electrifying railway systems is allowing a transition away from diesel powered rolling stock, but recent UK experiences have reminded the rail industry that retrofitting overhead contact systems to historic infrastructure can be complex and challenging. What is more, technology developments, particularly in batteries and hydrogen as a fuel, mean that a range of solutions is becoming available to transport authorities. More choice opens new opportunities but also additional considerations. Clients must optimise solutions not just for carbon but for cost, service quality and passenger experience too. There is a lot to consider. “The human brain is not very good at doing multi-criteria optimisations,” explains Sébastien Lechelle, head of traction power at Mott MacDonald.

Experienced professionals may be very adept at optimising projects based on cost, carbon or journey time, to build a strong business case. “But as soon as you have to take all three together in different proportions… it takes a very special person to do that, just with their brain,” says Sébastien.

Fortunately, technical advances are coming to the rescue. “In this day and age of artificial intelligence, computing power and big data, there are new things we can do here,” says Sébastien.

Rail Decarb helps identify where different combinations make the most sense in the full line electrification process.

Rail Decarb

The ‘new thing’ that Sébastien and his team have been working on is an intelligent optimisation tool called Rail Decarb. “We said OK, what if we had a tool employing artificial intelligence that we could feed data, and that would simultaneously consider the economic, social and environmental criteria to find an optimised solution?”

Things that make full line electrification difficult include tunnels, overbridges and the canopies of older stations, where there is insufficient clearance to accommodate the overhead line and pantograph; embankments with narrow crests and areas with poor ground conditions, making it difficult to erect gantries and masts; and distance from a high voltage power source, presenting physical and cost challenges in making a connection. To overcome these ‘obstacles’ requires electric trains with an on-board power source – traditionally diesel, but now available with batteries or with hydrogen as a fast-emerging technology. Trade-offs must be struck. For example, different modes of propulsion affect train speed and journey time; carrying lots of batteries or fuel increases train weight and consequently power demand as well as wear to the tracks.

Rail Decarb helps identify where different combinations make the most sense.

Data layers

Sébastien describes the large volume of information entered: It starts with baseline infrastructure, rolling stock and operational data on the existing network. Data laid on top of that includes the embodied and operational carbon and cost of equipment for electrification or fuelling, information about its maintenance requirements and lifespan; it also includes the embodied and operational carbon and cost, performance specifications, maintenance requirements and lifecycle of all potential rolling stock.

The next step is to provide Rail Decarb with essential design rules, such as how frequently substations are required to sit alongside electrified sections or how much power a battery can provide and what that means for train performance. The whole system is then analysed in exacting detail.

“What we do is a bit like the finite element method,” explains Sébastien speaking of the modelling used in structural analysis, where large areas are broken down into much smaller parts before applying an array of mathematical techniques to accurately calculate information about each element. These can then be reassembled to gain insight about the performance of a whole system. “We chop our railways into small cells. Each cell is examined in turn as being either electrified or not, and all possible combinations are explored to deliver an optimal result.”

Natural selection

Splitting a line into cells and then looking at them in different combinations means that the algorithm used to run the scenarios can present thousands, even millions of potential options for the final decarbonised railway. This is a step that has not been possible before because the options were always limited by human capacity.

The first outputs from the software are what Sébastien calls candidate solutions. “At this stage, we are not trying to determine something that's going to work exactly, we just generate lots of solutions that can be assessed in terms of net present value, carbon payback time and journey time.”

Net present value takes account of capital costs, operational costs and potential income. Rail Decarb looks at the carbon payback time for the lowest carbon options, taking account of embodied and operational carbon emissions. The third criterion is journey time. A candidate solution might show that use of a hybrid electric-battery train is as fast as a diesel-powered train over electrified sections of the network but slower when using battery power, resulting in longer journey time.

Setting weightings

“And then we use what's called fitness function,” explains Sébastien. This gives a score to each candidate solution and begins to reveal the trade-offs between cost, carbon and journey time. Rail Decarb allows the user to set different weightings for each parameter and explore all possibilities. Candidate solutions with the highest score are kept and those scoring poorly are abandoned. “The really clever bit is we use the genetic algorithm to optimise the decision: we take good candidate solutions and hybridise them.”

This is where the magic happens says Sébastien. “The beautiful thing is that over a certain number of iterations the best features of each solution combine and converge towards an optimum.”

This was something that Transport Scotland discovered during 2021 when a prototype of Rail Decarb was tested on the Fife to Levenmouth line. It ran 18,000 scenarios in just seven weeks, leading the organisation to comment that it had the potential to save years and millions of pounds in project appraisal. Transport Scotland had already investigated a range of electrification and low carbon solutions. Rail Decarb was used to retrospectively appraise the approach and options selected. The exercise was also an opportunity to better calibrate Rail Decarb with enhanced information on rolling stock maintenance and embedded carbon.

Future proof

Transport authorities are under pressure to reduce carbon and cost. These goals are not always easy to align, as illustrated by spiralling costs that led to cancellation of a host of English and Welsh railway electrification projects in 2017.

The ability to find robust solutions that will deliver value for money, and transparently show the workings, is essential: decision making must stand up to deep scrutiny. Sébastien credits his Mott MacDonald colleague Dr Joseph Cosgrave for first stating the problem and sowing the seeds of a solution. He recognised that engineers, no matter how experienced and thorough, cannot handle the quantities of information required to assess thousands of possible scenarios. Advances in AI and machine learning offered the opportunity to do so, and he set in train the idea of harnessing Mott MacDonald’s rail engineering know-how to that end.

Rail Decarb is the product – designed by engineers but with vastly greater data-crunching capacity and speed – delivering solutions hitting the optimal balance between carbon, cost and journey time, and improving all the time.

This article appeared first in Rail Professional magazine’s November 2022 issue.