I help to build hydroelectric plant control systems for one of my customers
and we use PID controllers to control the gates that let the water into the
generators (which controls the power output of the generator).

Tuning these controllers is always a pain in the neck, and unfortunately any
given tuning applies to a relatively narrow band of opeation. Temperature
changes, load level, wear and aging and, I swear, the phase of the moon,
affect the unit's response dramatically.

We've thought about trying to build an adaptive controller of some sort.
This controller would analyze the response of the unit to the control
signals we send it, and then adjust the algorithm to result in smoother
control (ie: quick response without overshoot).

One possibility would be to dynamically create a function of all the
appropriate variables (temperature, load, phase of moon, etc.) to modify the
PID loops constants. This function would be refined over time (days, months,
years).

Alternately, and the technique that I feel has the best chance of working,
we can simply watch the unit over a midrange timer interval (perhaps an hour
or so), and adjust the algorithm based on 'recent' behavior.

One additional issue with these controllers is that 'overcontrol' is a very
bad thing because you cause the mechanics of the gates to wear quickly. Its
not like we're just changing a PWM input into a motor. The controller
actually generates pulses that operate hydraulic valves that move a huge
cylinder that slews the gates around.

Does anybody have any good references/sources for such an adaptive system?

Bob Ammerman
RAm Systems
(contract development of high performance, high function, low-level
software)

--
http://www.piclist.com hint: The list server can filter out subtopics
(like ads or off topics) for you. See http://www.piclist.com/#topics