|
Perhaps what I have in mind is not easy to code, but what I'd want to try is a first run with the throttle at either 80% (just before the mixture gets enriched) or at the required cruise setting, or coasting. Ie: pretty much the way we drive. For the next iteration, I might try going from cruise to coasting a few seconds earlier in each instance, and seeing how the changes built up. Another kind of pass might look for opportunities to coast downhill, delaying acceleration until it is particularly easy. It might be necessary to run the data through several kinds of algorithms in each cycle, but the basic idea is to just change things to create a single new data set, not a sheaf. Then you compare that set to the previous one, and see where the best gains in the economy/time ratio has come about. You try more of the same on those places on the next pass, and back off to the previous setting where you hit some losers. You might need another register to see if the algorithms are working harmoniously or causing dithering instead of convergence. It might decide which one to try next, leaving some out on some passes.
Eventually, you'd get the program saying something like: OK, I can save you a buck at no time loss if you are willing to speed through the short valleys, drive slow over short hilltops, and coast sooner at lights. If you'll add a minute to your trip, I'll save you another buck in gas. If you'll give me two minutes, I can save two-fifty. . .etc.
|