Re: DCOPF with non-convex obj. function

[Carlos E Murillo-Sanchez]

Enrico: I would certainly experiment with the values of the reactive component of loads.  In fact, an extreme measure intended to make this problem more like a DC OPF, while still retaining the AC solver, would be to add synchronous capacitors (generators with Pmin=Pmax =0) to as many load buses as possible, with voltage setpoint of 1.0 .  Then you'd have automatic reactive compensation essentially everywhere, and the active dispatch would pretty much be free of reactive considerations. Carlos. Enrico Vaccariello wrote: Dear Dr. Sanchez, Thank you very much for your input. According to the manual, the DC-OPF should not take in consideration voltage magnitudes and reactive powers. I believe I have understood that what you are suggesting is to "manually" implement the simplifying assumptions of a DC-OPF in such a way that the AC-OPF provides similar results. I have already set all the branch resistances and charging susceptances to zero. Anyway, I am still having some convergence issues, but I believe they are not being merely numerical as before. Now, do you think it would be right to set equal to zero all the bus reactive loads as well? In fact, since my original intentions were to run the simulation in AC, all my loads had been originally defined as ohmic-inductive with non-zero reactive demand. The necessity of simplifying my problem down to a DC model was born after I faced too many unexplained convergence issues. Thank you. Best regards,       Enrico 2017-02-25 22:01 GMT+01:00 Carlos E Murillo-Sanchez <ce.murillosanc...@gmail.com>: Perhaps you can try to run an AC OPF with branches that are represented only by the series reactance (no line charging capacitance nor resistance); one of the AC solvers might be able to solve the resulting problem. Carlos. Enrico Vaccariello wrote: Dear all MATPOWER developers and users, I will briefly explain my issue. I am trying to run a DC-OPF on my 285-bus case, whose generators costs' curves have been defined as second-order polynomials. Some of these polynomials have negative convexity, as the one shown by the red line of the graph I am attaching. When I try to run the DC-OPF it does not converge. This happens with all the simplex, dual-simplex and primal-simplex OT and the MIPS solver. In the case of the first three linprog solvers, I get the error message: "The problem is non-convex". Therefore, I have changed the convexity of my cost curves polynomials, just as a test, by modifying their first coefficient: mpc.gencost(:,5)=abs(mpc.gencost(:,5)). The result of the changes is shown by the blue curve of the graph I am attaching. Both the curves refer to the same CCGT generator. Performing the same DC-OPF simulation with such modifications leads to convergence. Therefore (excuse me if that's trivial) the convexity of the cost curves guarantees the convexity of the objective function. Anyway, since my actual cost curves are non-convex, could you please indicate any other solver allowing to work with non-convex objective functions? Thank you, any help will be very appreciated. Best regards,       Enrico