Let's say there are three shifts a day in a manufacturing plant. The plant suffers from irregular power supply thereby affecting in how it is functioning in time and the shift's production. Therefore, you have the manufacturing process going on for a while and then pauses due to power outage(or lunch break) to be resumed when the power is back on. This on/off pattern of power is pretty much inconsistent and unpredictable.
By mining data of the past history, I want to come up with a model that given a point in a day of a shift I want it to be able to forecast the production for the rest of the shift(possibly accounting for the likelihood of experiencing power outages).
I was hoping for some perspective and idea in how to go about making this problem a machine learning problem and some advise in picking techniques amenable to the problem.