Currently we are investigating to know whether it is possible to use oemof-solph for doing an EMPC (economic model predictive control) for controlling different components (Li-ion Batteries, electrolyser, fuel cell) of a virtual power plant (or microgrid) in real-time. Being aware that the MPC which is based on solving an optimal control problem is often performed with dynamic programming (e.g. casadi, pyomo.dae) while oemof-solph deals with a LP or MILP, I just hope to get some experience sharing (should or should not) for using oemof-solph for doing EMPC of a microgrid or virtual power plant?
The answer heavily depends on what you define as “real time”. If you interpret the term as it is done in computer science (immediate reactions, even without clock cycles) it is not possible. If you have a pipeline of new data coming in and want to update your optimised schedule every now and then (e.g. every 5 Minutes), you can do that. There are two examples, I know about:
If you have a pipeline of new data coming in and want to update your optimised schedule every now and then (e.g. every 5 Minutes)
That is what we want to do indeed but probably with a lower time step (e.g. every 1 minutes). Your shared slides about CHP operating based on forecasts (I don’t know whether it is possible to have an accessible recording of this presentation it would be really nice) and Quartier EMS are really relevant.
Unfortunately, I already shared the best resources I have. Anyway, I guess, the level of detail would just tell you that it is doable, not how specifically the implementation looks. What I know is that both approaches focus on electricity market, making 15 minutes the relevant resolution.
Depending on model size and data availability, more frequent updates should be doable. However, there is a point where higher temporal resolution would need a more detailed model to really improve the results. I am not sure where this point is, but there is an analogue phenomenon in PV data. There improving the temporal resolution beyond 15 Minutes (even an hour?) is not worth it, if the spacial resolution of the weather data cannot correspond.