Álvaro's answers are correct, but maybe these additions will help too … On Dec 9, 2012, at 12:20 PM, Fred Kanjelesa <[email protected]> wrote:
> I am using MATPOWER to run OPF for a 60 bus network modeled based on a real > power system. the output window indicates MIPS is the solver used and > converges in quite a few iterations. > > Question 1 > what i do not understand is whether MATPOWER uses Quadratic Programing (QP) > for power systems defined with quadratic generation cost functions or it uses > Linear Programing (LP) For the AC problem, it does not use QP or LP, it solves the full NLP problem. For DC OPF it uses a QP if the costs are quadratic, otherwise an LP. > Question 2 > If it uses LP is there a way of knowing in how many linear segments the > polynomial/quadratic cost function is converted to linearize it? MATPOWER can model both polynomial cost functions (only up to quadratic for DC OPF) or piecewise linear. For AC it uses the full non-linear cost function. For DC it picks the solver based on the cost function, but it will not automatically convert a polynomial to a piecewise linear cost. MATPOWER does include a utility function (poly2pwl.m) that can do this, but it is not used internally. > Question 3 > Is there a way of knowing the confidence level of the results obtained from > an OPF MATPOWER simulation, i.e. a value or something to use one can use to > convince others the certainity of the results. Each algorithm has a set of parameters it uses as termination criteria. Typically using smaller values will give more precise results, but there is a limit to how small you can go before you reach numerical limitations. I don't think there is a simple "confidence level" number MATPOWER can give you, but in practice if you solve the same problem with multiple solvers, the differences between the solutions should give you an estimate of the numerical accuracy of the solution. -- Ray Zimmerman Senior Research Associate 419A Warren Hall, Cornell University, Ithaca, NY 14853 phone: (607) 255-9645
