Hi!
When running a scenario simulation, is there randomness introduced in the grid behavior (e.g., monte carlo) or is it static? In other words, if I run the same simulation with all the inputs (network, scenarios, etc.) the same, will I get the same results or different each time?
Thanks!
This is a great question! There certainly can be some randomness in the solution.
One potential source of “randomness” can come from the MIP Gap. Solving a mixed integer optimization problem (a “MIP” problem) is mathematically challenging. I recommend checkout out this MIP Gap explanation from Gurobi.
A slightly more nefarious (in my opinion) potential source of “randomness” can come from degenerate solutions. Degenerate solutions are a complex topic in optimization, but within the context of production cost modeling you can think of them as solutions with different results but the same objective function value. For example, if you have a power system with only 2 generators and each have a cost of 5 [$/MWh], which generator would be dispatched in an optimal solution? Since the solver can’t distinguish which is better, you may observe different behaviors each time you solve the problem. When you consider this in conjunction with the need to have a MIP Gap for production cost models (since binary variables are required), things can get challenging.
This tends to show up strongest with respect to storage. Different solutions with different storage charging behaviors can often result in degenerate solutions. For example, if I have a solution with an energy price of 0 [$/MWh] in the first two timesteps, should storage charge in the first timestep or the second (assuming it must pick in this simple example)?
One strategy to overcome this is to embed small costs (in the form of VOMPrice, VOMGPrice, and/or VOMDPrice) for the different objects in a network to help “break the degeneracy”. This can lead to more stable solutions (and faster solve times). However, one must be careful with such assumptions and be aware of the potential side-effects they may need to account for in their analysis of the results.
I hope this helps! Happy modeling 