I have not seen any advanced intersection writeup. Most tutorials go over the signals. I'd suggest asking questions in the Factorio discord server in the #train-help channel. There are many experienced people there who can answer many of your questions. That being said, I will go over the rest of your post.
Parallel lanes on non-crossing turns are for consistency of throughput. They buffer trains against the randomness of the input. They are not as effective as buffering before and after crossings, of course.PavelH wrote: ↑Sun Oct 01, 2023 4:37 pm1a. A particular pattern that I don't understand is parallel lines on the non-intersecting turns - since both the input and output are single lane, I don't understand how this helps throughput. Is it somehow related to how the game selects which train has priority to enter the final block?
There are two things that could be going on here. One is that the exit blocks after crossings don't actually have to be directly after the crossing. They can be offset from the crossing at intervals of full train lengths if they are consistently a certain length.PavelH wrote: ↑Sun Oct 01, 2023 4:37 pm1b. Another pattern I don't understand are long non-signaled section that are not immediately after a crossing, or are much later than needed for the crossing. What role do these play? I understand having too dense signals causes throughout to drop due to trains not being able to speed up, but this seems excessive.
A specific example of this would be "parallel multicross turbo", where the blocks preceding the exit are at least 12 carriages long, with some other blocks also being seemingly unnecessarily long
In the Parallel Multicross Turbo, and a few of the others, they utilize high throughput merges. Regular merges get about 27 tpm for 2-4 trains. High throughput merges weave together traffic so that trains will merge at high speed instead of coming to a stop before merging. They can get about 33-34 tpm with this sort of merge.
You also said "having too dense signals causes throughout to drop". In general, the more dense the signals, the higher the throughput. After a crossing, the exit blocks must be the length of your longest trains to prevent deadlock.
Signals are used in one line for mirror symmetry I assume. I also offset 1 tile for my straight sections. It doesn't really matter. The blocks should be sliced the same regardless.PavelH wrote: ↑Sun Oct 01, 2023 4:37 pm2. Is there a reason all intersections are using signals in one line? Personally, I have an offset of 1 tile between the "in" and "out" signals, because that enables the straight section to be a rotationally symmetric blueprint, and (according to the indicator when placing the signals) the train stops in the same place. Does this matter at all?
I've seen intersections that have 4 lanes horizontally and 2 lanes vertically, for example. Your adaptation was flawed. You would need to adapt the Parallel Multicross 4 Lane instead, by reducing the 3 sides from 4 lanes to 2 lanes. Then you'd probably see some throughput increase over a 2 lane Parallel Multicross. Personally, I would just have 2 <-> 4 lane conversions and leave the intersections alone. Intersections are the major bottlenecks in a rail system. Don't have merges/splits before/within/after them if at all possible. In reality, 4 lanes are rarely needed anyhow. Better rail system design is much better at preventing congestion.PavelH wrote: ↑Sun Oct 01, 2023 4:37 pm3. Does it make sense to try designing an intersection with 4 lanes on one side and 2 lanes on the rest? Or am I unlikely to get much better than joining 4 lanes into 2 before the intersection?
The use case I imagine for this would be a 4 lane "spine" network, with several 2 lane "local" networks connected to the spine.
Since all of the large designs parallelize each path, I tried adapting some of them (namely "parallel multicross" and "parallel multicross turbo") to 4 lanes on one side, but the benchmark didn't really show any difference in throughput. I'm not sure whether this is because my design is bad or because the train direction mix used in the benchmark doesn't really match where such an intersection would be used. Any thoughts?
Once again, I'd suggest going to the discord server. They're very helpful and friendly!
Avona
