Trying to converte DC Multi-Period OPF to AC Multi-Period OPF
DiogoFreitas60 opened this issue · 7 comments
Greetings
On the light of my dissertation where i am trying to study the impact of the addition of battery storage systems to a grid and analyze its impacts on the OPF for the system.
After a conversation with te professor that is guiding me, we talked about trying to use the DC Multi Period OPF from MOST and try to adapt it to work an AC OPF.
Was just posting this here in order to see if anyone would have any idea or suggestion of how to do this
Thank you all for your time
Best Regards
@carlosmurillosanchez, who implemented the early prototypes of MOST, has done some work on AC versions. I'll let him respond with his suggestions.
The original formulation for MOST tried to accomplish the AC version of the algorithm by means of a decomposition involving, on one hand, a central MIQP problem involving all intertemporal constraints and binary variables, but only a relaxed version of the flow, and on the other hand, AC flows (OPFs) for each operating state in the model. A decomposition coordination scheme based on the Auxiliary Principle of G. Cohen was then applied to this, using a simple gradient ascent rule for the dual iterations. This scheme exhibited convergence problems due to the fact that in intertemporally-coupled problems it is possible to have negative nodal prices in some periods and buses, which makes the OPF subproblems unstable. Not a lot of progress has been made on this until recently, when a PhD student agreed to tackle this problem. We have been working (starting from the last code that Ray and I shared before Ray transformed it into MOST using the object-oriented problem description mechanism in MATPOWER, and which includes some hydro-specific additions) on re-implementing the whole scheme using an islanding technique and the generalized OPF capabilities in MATPOWER. This would allow us to solve continuous (no unit commitment) AC problems that fit the MOST paradigm, but it would work for small problems only. The reason why we are doing this is so that we can solve some small-sized problems and then compare upcoming decomposition schemes with known-solved problems. For the large scale case, we are looking at techniques for the dual update using second order information for the dual iteration; generalized Benders; and decomposition techniques at the Newton step level of a huge IPM for the whole problem.
I do not know how large are the systems that you want to solve. Perhaps, when we are done with the islanded implementation, we can share it; I do not know how viable it will be to include it alonside MOST (it would need extra work to meet the data format, quality standards and provenness already in MOST).
Carlos.
Diogo: If you do not need the stochastic aspect of the problem formulation in MOST, then trying to modify MOST to be able to work with AC modeling is probably overkill. Do you need the stochasticity in renewable generation? Do you need the N-1 security? The ramp constraints? Unit commitment? Market treatment of goods such as reserve and ramp reserve? Let's first try to decide if you really need the full-fledged MOST capabilities or something simpler, in which case I can guide you on how to implement that using the standard generalized OPF in MATPOWER (and move the discussion out of github).
DIogo: let's continue this via email at carlos_murillo@ieee.org
Feel free to re-open this issue if there is more discussion that belongs here.