Dec 3, 2022
Understretching and overstretching
SBS Transit has apparently announced that they will use “AI” to control train service levels on the DTL.
This is being done in partnership with Siemens, the supplier of the DTL’s signalling system. Siemens has a webpage describing their Controlguide Airo MT solution which will power this new system. So we can probably thus assume this is already a commercialized product in other parts of the world; even if SBST claims to be the first in Asia-Pacific to use such a model.
Stuck in the past?
I am deeply skeptical of how well this is going to work. I’ve gone on for long enough previously on the blog about SBST’s current practices of staffing every train on both their MRT lines even though they are supposed to be able to drive themselves with minimal supervision. There’s also the extremely inflexible way they schedule their first and last trains, or respond to disruptions.
There are ways to survive such a change, but it will likely require so large an operational adaptation in procedures over so short a period that I’m not convinced SBST will be able to pull it off — after all, this system will be implemented “in the coming months” according to ST. Months, not years. And there’s no point in having such a system if the recommendations made by the system are constantly overridden for one reason or the other, or is forced to moderate its projections due to available manpower, and thus it becomes a useless paperweight.
For one, the flexible scheduling promised by Controlguide Airo, where a timetable will be prepared only the day before for the next day, means these train rover staff will likely only know the day before whether their services are required, after the system finalizes its timetable for that day. That’s the kind of work schedule that absolutely kills a social life, and I would not be surprised if SBST thus experiences further attrition of staff because of this. Alternatively, SBST might just choose to employ maximum staffing at all times, but that artificially increases their labour cost for not a lot of benefit.
What does this mean? SBST may be forced to learn, and quickly, how overseas operators of automated metros, such as Vancouver and Copenhagen — or even pre-2015 SMRT — do this, with only established patrol routes and station staff trained to respond to train issues. But it would be deeply counterproductive if in the meantime we saw a large amount of train cancellations — let alone any possible increase in service level — because of this lack of staff.
The proof is in the pudding
2023 will soon be upon us, that means the Marina Bay Countdown too, and like with other large-scale events, we need to think about public transport arrangements in the CBD.
Bus services being rerouted during a road closure also means bus arrival timings are unavailable throughout the disruption period. With the lengths and utilities of the affected bus routes being more than just the CBD alone, this may have knock on effects on far more people than the actual road closures. Paradoxically, this makes it harder to expect people to switch to public transport during the times when they ought to do so the most, as without arrival timings, public buses are made harder to use during this period.
To sidetrack a bit, I’m not sure if this is a system issue that can be fixed with a system upgrade, but this can be a bit of a shock if you open your bus app and it’s completely, totally empty with no bus arrivals. You can choose to sit down there and wait, but it’ll be a gamble. Especially for the Marina Centre shutdowns, where the bus terminal in the area is also forced to close because buses can’t physically access the terminal. This happened on National Day and F1 weekend, and will repeat itself during New Year’s, as it has in previous years.
So if buses are made much harder to use or even straight out unavailable due to such widespread road closures and service changes, who takes the slack? The trains. Apart from other journeys that are now made by rail, we need to consider what happens downtown. The idea behind Controlguide AIRO, according to Siemens marketing copy, is that it can ingest information from various sources such as newspaper articles and CCTV feeds to develop a timetable one day in advance.
Perhaps how it works in practice could be that the system might be able to extrapolate a service level required for the 2023 F1 weekend based on the 302,000 visitors to F1 2022 as well as prior statistics on the use of surrounding stations during the F1 period. The AI-developed timetable might then be able to time a ramp up in service at stations that need it just in time as events finish.
This will matter all the more during events throughout Marina Bay considering that serving the immediate vicinity of Marina Bay is what the Downtown Extension was built for. Three DTL stations serve the key activity nodes of Marina Bay: the Floating Platform/NS Square near Promenade Station, The Promontory@Marina Bay near Downtown Station, and the MBS waterfront promenade near Bayfront Station. Supplying the capacity to get all these people home quickly will require the schedule to provide a very high level of service, which hopefully the AI delivers.
But this doesn’t absolve DTL management or line control from the need for real-time monitoring and responding appropriately to real-time demand by keeping additional trains on standby in the right places. Even AI can get things wrong, as any practitioner will tell you.
Seen and unforeseen
There’s another more important reason why such steps ought to be taken, AI-controlled train schedules or not.
It probably says something when SBST’s PR people were helpfully tweeting suggestions to go take other lines when NEL services were disrupted on the 15th and 23rd of November — twice in two weeks. They could be adapting, but we can’t be too sure especially since they’re at the giving end and not the receiving end.
Strangely enough, this is very close to a possible scenario I detailed when advocating for the Springleaf bus hub. Especially the last case, where the CCL/NEL legs could easily be replaced with an eastbound bus from Springleaf station for Sengkang residents. But realistically, since the Springleaf bus hub might not be a lot of use for an Hougang or Punggol resident, how would this have played out?
As a driverless line with more than enough capacity, the TEL and perhaps CCL control rooms would have been able to dig up extra trains or extend the higher-intensity peak hour service to facilitate diversions via Caldecott or Bishan provided they had the flexibility to do so and were not constrained by manpower.
The whole concept of fully-manned trains on supposedly automated lines to supposedly improve system reliability is how we sacrificed that ability to flexibility respond to any issues that may arise on other parts of the network. If TEL control is open to the idea that any additional trains put in service as a response to this sudden increase in demand can be unmanned, that will work. If it doesn’t, this goes out the window. But whatever it is, they’re a different company with different practices, and I wager that SMRT might even ramp down the manpower allocation once TEL stability is in a better place.
The crux of the question is, will SBST ever be in a position to return the favour someday when the EWL mainline is unavailable? Well, they better. It might be an easier ask to stage extra feeder service from Beauty World to Bukit Batok and/or Jurong East should the EWL ever encounter difficulties, compared to the high frequency needed for bridging buses. Likewise, the appropriate connections should enable Bedok and Tampines residents to close the gap from a DTL3 or TEL station. But more importantly the extra train service levels will need to be provided in a time like this. And will the AI make it easier to do so? It really depends.
This might also need to be a greater change network-wide than just some PR people being helpful or one control room launching extra trains on its own initiative. Someone also has to be in a position to pick up the phone and make the necessary calls to order an increase in the service level of any diversionary routes by activating spare drivers and vehicles that are placed on standby, especially if they hold the power of the purse too.
EDIT: And who holds that power of the purse?? The regulators, who will also need to support such a cultural shift to ensure the success of such a system.
Chances are, that person is somewhere in the Land Transport Operations Centre. And such a greater mindset change means that the policies and procedures they depend on also need to take into account these possibilities, which are far more numerous and complex than putting up signs that basically say “you can take these buses to these MRT stations” — signs which get outdated quickly due to the pace of evolution of the bus network if not for the art and craft skills of local station managers.
What about us? I’d say the same — with the proviso that the signs have to be there first. It might not be obvious, though, and that is why the MyTransport app, amongst others, provides the ability to route around diversions. Once again, the value of live data cannot be overstated any more here, where people can be alerted at the outset on how their journeys are disrupted and what alternatives they can choose.
Holding fast
The last and most immediately concerning bit of the Siemens marketing language, to an untrained eye, is that it offers the opportunity to train operators to reduce mileage. Some might have their alarm bells ringing in that it might tempt SBST into cutting service to some degree —concerns that are not unwarranted, but in practice may be a rather silly idea considering that much of their costs are likely to be fixed due to running station and trackside infrastructure. Unlike Deutsche Bahn or something, there are still 34 fully-underground stations each with appropriate lighting, ventilation, and air cooling needs.
Or short turns to not go so far out to less busy stations up the line could also happen. The latter, I daresay, is quite unlikely to happen considering that Bukit Panjang and Tampines East, both at or near the end of the line, see considerable passenger traffic heading into town. What might help, though, is for the AI-generated timetable to tweak the stopping times at less busy stations, thereby speeding things up. It benefits both the passenger, as they get faster trips, and the operator, who need to use less trains to maintain a given service level. Or for a given amount of trains, they can even increase service, which would be a better outcome too.
It could even be possible to vary the stopping time at stations depending on the time of day. For example, Tan Kah Kee station at the heart of the Bukit Timah academic belt might not need long dwell times during the morning when students are in school or the PM rush hour when they’re at home; but at school dismissal time it may make sense to stop trains longer so more students can get on board. This could extend further to the counter-peak direction as well, so they can get to where they’re needed faster.
It’s not that the DTL needs to conserve trains, as they have way more than they could use now. But it makes me wonder whether optimizing train usage now could help reduce the fleet requirements for the Sungei Kadut extension. This avoids the millions of spending and awkward off-band replacement cycles that is likely to result if a new train order needs to be placed for the Sungei Kadut extension alone, which will likely have to be in place likely in the mid-2030s.
There are many possibilities here, and I will be watching with great interest.
