Manuka, reliable? maybe not as relaible as a 'simple' system but as with any automation (eg autopilot) you should always have the ultimate cop-out clause, namely; manual override!
How can a smart system anticipate? By doing whatever is required such that it does not need to. Put simply, energy needs are whatever they need (excluding efforts at efficiency) so the system needs to cater for maximum demand. Hot water can be stored but if it gets all used up, then you need to make more. Knowing that in advance will not help. Peak electric useage can be reduced with simple tricks such as 'fridge goes off when kettle comes on' and many other such examples of devices which do not require power at specific set times. However, you don't want the fridge off all day because somebody left all the lights on so there needs a little more intelligence to override. eg allow fridge to rise by 2 degrees rather than turn it off. Ultimately, if has to be on, it has to be on.
Then there is "dynamic demand" which looks at mains frequency to determine grid loading. When the latest soap on TV has an add, all the kettles go on, the line frequency drops and anything that does not need to be on right now can go off. eg immersion, fridge or electric oven.
My interest is in "use it or lose it" PV generation.
Lots of testing to be done but here's my biggest concern.
Scenario:-
PV generating say 2kW
House using say 1kW
Spare = 1kW
Immersion element rated at 3kW
If you switch on and off at 60S intervals, you will lose 1kW for 60s then draw 2kW from the grid for 60s. (worse than no control)
So, just how quick can the utility meter work things out?
What if we burst fire over say 1s?
How about phase angle control? Even then, the INSTANT current drawn will still be over and above that available from the PV.
If the meter is PERFECT, it will register the use (even though the NET is zero).
I've heard/read that most meters will allow a certain amount of 'bouncing' before they clock up units but I've not yet tested this on my own unit. There's no way a PICAXE could keep track of phase-fire so it's back to good old anlogue techniques with the PICAXE being fed data that is averaged over several cycles.