How/What Does the Prius' CPU(s) "Learn?"

[Jack66]

Someone asked me today if you could save the values stored in memory before disconnecting the battery then put them back after the battery is reconnected. Is that even possible? If it is volatile memory you should be able to write to it so it sounds reasonable.

[qbee42]

It's only possible if you are given read/write access to the memory. I am not aware of any way to do this, otherwise it should be a common service practice.

If your friend is just replacing a battery, he can maintain 12 Volts on the system while he swaps out the old battery. Connect the maintaining source to the jump points under the hood.

Tom

[patsparks]

Most ECUs with a knock sensor will remember the advance curve that is optimum for the fuel and the condition of the engine so they don't have the engine knocking all the time.

[kocho]

... it seems the prius does *not* learn the driver's driving habits and adjust driveability to match them. Any info to the opposite?

[Rokeby]

Kudos to the many posters. You've brought up some most interesting
points.

From what's been said:
* it is highly unlikely that terrain and location data is stored as "Historical
Data."
* there are ICE/FE related data that change over time that are
necesssary for efficient ICE mangement and might be stored as Historical
Data.

The Tech Manual says that recovery of some of the data stored as
Historical Data can take 3 to 6 weeks to recover/replace.
It is hard to believe that such a long capture window is needed for ICE
specific data. I have read through the Prius specific sections of the Tech
Manual and it is perfectly clear that the CPU(s) are essentially
continuously monitoring conditions that affect the ICE. In the case of fuel
injector timing there are two correction regimes; short and long term. But
the time intervals may be 2 and 10 minutes respectively.

It seems to me that greatest variables affecting ICE operation are all
external to the ICE itself; the operator, ambient temperature, and load
carried in the vehicle. The CPU would have to make adjustments to the
latter two rather quickly to run cleanly and efficiently. This would seem to
indicate that the three week capture window is for operator related data.
But I found nothing in the Tech Manual that would support this idea.

In the past it took the early adopters great amounts of time to gather the
data needed to identify and then validate data that we now take for
granted: ICE stage transitions, stealth mode precursors, etc. So it would
not surprise me that it will take a while to identify "What the CPU(s)
learn."

It is most probable that the process will outlive this thread. My hope is
that we can make a first attempt at identifying data that the CPU(s)
might need to "learn" and store away, including tentatively identifying the
data that takes 3-6 weeks to collect.

For my part, I'm still having difficulty coming to grips with one of the
paragraphs extracted from the Tech manual:

"The battery ECU closely monitors energy consumption in Watts. By
calculating the amount of energy spent, recovered and stored, the
computer can calculate the required fuel burn. Fuel required to create
this amount of energy is compared against the engine ECU fuel injection
calculation to ensure accuracy."

It seems to me that regen energy, as it can come from hills and such is a
wild card and would make this form of determining fuel use equivilance
highly suspect.

I clearly have a lot to learn.

[racerbob]

"Long Term" to an ECU is only maybe every 2-5 seconds. STFT (short term fuel trim) is continuous and once it needs to deviate much more than 3% (plus or minus) from the programmed tables, it is stored as a LTFT correction and the short term reverts to a "new" zero point. LTFT may take several days to populate after a clear has occured (losing battery power or a ECU "reset" done with a scan tool). This is because many "blocks" must be populated. The only way this will happen is if the engine is cruising in every different rpm and MAP values that define the block tables. Normally the only driver specific habits that influence any stored values were related to transmission shift points. A Prius has a CVT so no shift points are necessary. It does however make sense that the Fuel Mileage Calculation could benefit by long term corrections that would get more accurate with a bigger database of information.

[Rokeby]

This summarizes and consolidates what posters have said so far.

In summary, data that the Prius' CPU(s) might collect and save as "Historical
Data" include:
1. Short Term and Long Term Fuel Trim adjustments.
2. Information that is used for control system tuning compensating for minor
differences from engine to engine.
3. Most ECUs with a knock sensor remember the optimum advance curve

Further, "...the Prius has no knowledge of terrain." qbee 42

The above three summary points come directly from these posts:

dogfriend first pointed out: "Short Term and Long Term Fuel Trim
adjustments. This is data that all OBD 2 systems store to adjust for
differences in individual engines." (Post #3)

In this same vein, racerbob said: "Long term fuel trim is a set of
"cells" in the engine ECU that store adjusted values of fuel mixture control.
These are adjusted according to engine characteristics such as age, pumping
ability (ring & valve condition, vacuum leaks, air filter restriction), cam
wear, etc. When 12v battery is disconnected these may take a couple of
days to fill up all the cells (blocks) with useful info." (Post #10)

And racerbob further added: "'Long Term' to an ECU is only maybe
every 2-5 seconds. STFT (short term fuel trim) is continuous and once it
needs to deviate much more than 3% (plus or minus) from the programmed
tables, it is stored as a LTFT correction and the short term reverts to a 'new'
zero point. LTFT may take several days to populate after a clear has occured
(losing battery power or a ECU 'reset' done with a scan tool). This is because
many "blocks" must be populated. The only way this will happen is if the
engine is cruising in every different rpm and MAP values that define the
block tables. Normally the only driver specific habits that influence any
stored values were related to transmission shift points. A Prius has a CVT so
no shift points are necessary. It does however make sense that the Fuel
Mileage Calculation could benefit by long term corrections that would get
more accurate with a bigger database of information." (Post #16)

qbee42 noted: "Most if not all new electronic control systems on
automobiles collect some information that is used for control system
tuning... Much of the information compensates for minor differences from
engine to engine. Yours may have more carbon on the plugs, or perhaps run
in a colder climate." (Post #7)

patsparks chimed in: "Most ECUs with a knock sensor will
remember the advance curve that is optimum for the fuel and the condition
of the engine so they don't have the engine knocking all the time." (Post #13)