MPG Cruise Control Algorithm

[GasSavers_maximilian]

The LUT would just be an "on the paper" estimate, so it can be pretty sparse, with a control loop finding the exact value. As I wouldn't want to compute different ones for each model of car, I can't get very specific anyway. While speed changes are pretty slow, so is the OBDII update rate, so I thought a table to get in the ballpark would enhance performance a bit. I guess I could have it cache values for future use. I can always interpolate the entries if needed to strike a compromise. That's the method of fuzzy logic controllers in a nutshell: LUT with a hack to interpolate.

I don't think any system not incorporating a GPS would be worth actually building. Maybe not even that.

I hear you on the space concerns. I'm writing a project with a LUT right now for an AVR tiny12 and that sucker has very little onboard. Fortunately the table is only 18x2, since the data range required is pretty small.

[GasSavers_maximilian]

I was thinking some about the particulars of implementing this and I think inferring the grade from the velocity and throttle data may not be sufficiently responsive, especially given the low ODB update rate. I know when I was using the ScanGauge I'd anticipate incline changes and then just fine tune it with the instantaneous mpg readout. If that's true, it'd imply adding an inclinometer. In theory all one needs is the incline and velocity to control throttle, but it'd need to be calibrated for each vehicle. Weight changes would show up in the changes in velocity, so that'd be no problem.

All in all, a lot of work. Better to wait for a Google Android device I want and then explore this via software. I keep trying to challenge that conclusion and failing. Oh well.

[GasSavers_maximilian]

After reading oneinchsidehop's great observation that a constant throttle seems to handle minor hill variations well, I can drastically simplify the basic algorithm. In fact, it gets so simple there's no need for pseudo code:

Stay at a constant throttle unless the speed goes too low or too high, then just maintain those limits.

No OBDII input required certainly is tantalizing. Having two inputs for min and max speed seems like it might be too complicated (although proper input design could avoid the max being less than the min ever). Thinking about how I drive around here in a 50 mph zone (default VT speed limit), I get a min speed of about 40 (slower than that and I'd be in serious danger of a road rage shooting) and a max speed of 60 (hills steeper than that I go into neutral and coast). +/- 10mph would be a lot in a 35 zone, so maybe the min & max could be a percentage of the set cruise speed? So that'd be +/- 20%, or ranging between 28 and 42 in a 35 zone. Still seems high, but you could always set the target below 35 I guess. Need to let this process a bit.

[theholycow]

Constant throttle is a different strategy, and has worked well for some. It's often called Driving With Load or DWL.

[GasSavers_maximilian]

I tried it yesterday and it seemed to work pretty well. Of course I was towing a laden trailer so I need to test it under normal conditions too. Keeping my foot totally still was harder than I expected. I still did other things like coast, but I'd interrupt a basic mpg cruise algorithm similarly.

[GasSavers_maximilian]

Did some more testing today and it appears I don't get to 20% above my level cruise speed before switching over to coasting. 10% is more like it. A range of 20% below and 10% above seems to match things quite well. One difficulty of using such a cruise system is that the naive approach would require it be set at the correct speed and also at the correct throttle. Tricky if there aren't a lot of level stretches (like around here). A table of throttles vs level travel speeds could allow it to assume the speed at the moment of setting and look up the appropriate target throttle. Still need to think some more.

My economy is definitely improving from the constant throttle technique.

[GasSavers_maximilian]

Having slept on it a bit I'm satisfied with the simpler algorithm, but now that I have something that could realistically be made, I have to face the fact that I am too much of a wimp to install an aftermarket cruise with experimental circuitry on my still under warranty car. Ah well, not like I really have the time anyway.