Automatically enable targeted load shedding when necessary

[danielolsen]

In REISE classic, we automatically shed load by 5% during every interval in which Gurobi returns any status besides optimal. REISE.jl has the ability to shed load in a targeted way, only shedding load at the buses and hours where it is necessary, but currently this feature is controlled via a flag in the source code. If it is enabled, it adds significant dimensionality to problem (one new variable for each of 30,000 load buses, in each hour), slowing the solve time and problem update time and consuming more memory, even when it is not needed for a whole scenario. If it is not enabled, load shedding is controlled by the crude and time-consuming iterative 5% step process.

Proposal

• Begin with load shedding disabled.
• Upon the first non-optimal status return, re-construct the model with load shedding enabled and re-run the same interval.
• Optional: if further non-optimal statuses are returned, enable additional solver settings which instruct the solver to take even more care in avoiding numerical issues.
• Optional: after N consecutive intervals in load shedding is enabled but no load is shed, if there are still may intervals left to go, re-build the model without load shedding enabled, to hopefully speed up the remainder of the scenario once we are past peak load.
• When extracting scenarios, we can no longer depend on knowing which results will be present based on the first result_*.mat file. We will need to be able to create a dataframe to hold results from load_shed if they are encountered in any interval, filling in zeros for all preceeding intervals.

Priority
Medium. It is not blocking anything, but if we can get it done soon then it will pay dividends for all future scenarios that we run.