Well, the way I've done it is, run the control itself in a PID with per-cycle update rate, controlling to current or power setpoint, then an outer temp control loop at a lower update rate.
Current might be coil (rect. average, or peak) or switch (peak) current, or DC link (average inverter input). The latter of course makes power measurement simple, independent of coil Q.
Otherwise, you'll need some calculation of coil V, I, Q..., which isn't the easiest.
Partly control partly protection, they also go to pulse-skipping or burst mode at low output, including at high Q (when power loss per cycle is small, i.e. inverter gain is high). Which is partly how a pan is detected.
Tim