> This system would run my home for about 8 hours in the winter and 4 in the summer before emptying, so I would need a lot more smarts in the box to try and anticipate price spikes without running out.

Most new grid-tie inverters can be controlled from a computer either over IP or a serial interface; you could fairly trivially control the flow of power by polling your electric provider's website (I scrape the RRTP API for current and future pricing data every 60 seconds) and commanding your inverter based on that.

I don't have a lot of experience with those inverters. How many cycles do you drop when switching over?

My first idea was to create a system that watched the immediate price as well as the day ahead AND the weather forecast to try and predict the best time to stay off the grid, but that's a little too ambitious.

My current idea is to grab my 5-minute history over the last year or two and model it against a simple buy-low-sell-high threshold. It should at least tell me how much the system would save each month.

(BTW about your earlier post, even if supply is 1c/kWh, there's still the delivery charge and taxes. I think supply needs to drop to something like -3.5c/kWh before the juice is "free". It's happened only once or twice in my history with the system.).