Up until very recently all train route planning in Sweden was done manually with huge sheets of paper and rulers. Scheduling trains is a very hard problem due to the inability for one train to pass another in many places and the constantly high congestion. Any change in the schedule affects basically the entire rail network.
In computer science it belongs to the class of “NP-hard” problems which are essentially intractable to solve optimally. There are various approximate methods that can be employed to get something that’s almost always good enough.
done manually with huge sheets of paper and rulers. Scheduling trains is a very hard problem
Just because it is a hard problem, it doesn’t mean it is easier to do with paper and rulers than with algorithms. So the problem being “hard” should be unrelated to Sweden being behind the times.
The problem with train scheduling is that computers don’t just make it better. There’s a very real chance computers could make it worse or result in approximately the same outcome but not require millions in investment into a new system.
The humans are likely following an algorithm of approximation too. If programmed correctly, the computer would be able to try way more approximations using the same algorithm the humans use, and hence likely find a better solution.
It is of course possible to program a computer badly, so you get worse outcomes than the existing human method. But that is kinda besides the point.
It an investment return problem as much as a technical one. A computer could maybe reduce a team of 6 to a team of 2, saving optimistically 500k/year. The cost of building that system is likely at least ten million, putting a break even at 20 years minimum. When you factor in the costs of maintenance and running the program, it’s going to make it take even longer. Also because it’s not actually a solvable problem, you have to pick an accuracy level you are comfortable with, too low and you need too many people fixing things anyway, too high and people are cheaper.
I’m having trouble understanding how that relates to the statement about train operators, the ones actually driving the trains because most trains are not autonomous.
Up until very recently all train route planning in Sweden was done manually with huge sheets of paper and rulers. Scheduling trains is a very hard problem due to the inability for one train to pass another in many places and the constantly high congestion. Any change in the schedule affects basically the entire rail network.
In computer science it belongs to the class of “NP-hard” problems which are essentially intractable to solve optimally. There are various approximate methods that can be employed to get something that’s almost always good enough.
Just because it is a hard problem, it doesn’t mean it is easier to do with paper and rulers than with algorithms. So the problem being “hard” should be unrelated to Sweden being behind the times.
The problem with train scheduling is that computers don’t just make it better. There’s a very real chance computers could make it worse or result in approximately the same outcome but not require millions in investment into a new system.
The humans are likely following an algorithm of approximation too. If programmed correctly, the computer would be able to try way more approximations using the same algorithm the humans use, and hence likely find a better solution.
It is of course possible to program a computer badly, so you get worse outcomes than the existing human method. But that is kinda besides the point.
It an investment return problem as much as a technical one. A computer could maybe reduce a team of 6 to a team of 2, saving optimistically 500k/year. The cost of building that system is likely at least ten million, putting a break even at 20 years minimum. When you factor in the costs of maintenance and running the program, it’s going to make it take even longer. Also because it’s not actually a solvable problem, you have to pick an accuracy level you are comfortable with, too low and you need too many people fixing things anyway, too high and people are cheaper.
I’m having trouble understanding how that relates to the statement about train operators, the ones actually driving the trains because most trains are not autonomous.