Dear Dr Zimmerman,
Big thanks for your instructions.
I tried this way to add my cost function, but I got an error. Could you
please take a look ?
Here is the error description:
*Error using branch_loss_costfcnToo many output arguments.Error in
toggle_branch_loss>@(x)branch_loss_costfcn(x) (line 33)fcn1 =
@(x)branch_loss_costfcn(x);Error in opt_model/eval_nln_cost (line 48)
[fk, dfk] = fcn(xx); %% evaluate kth cost and gradientError in
opf_costfcn (line 45) [f, df] = om.eval_nln_cost(x);Error in
mipsopf_solver>@(x)opf_costfcn(x,om) (line 125)f_fcn = @(x)opf_costfcn(x,
om);Error in mips (line 349)[f, df] = f_fcn(x); %% costError in
mipsopf_solver (line 129) mips(f_fcn, x0, A, l, u, xmin, xmax, gh_fcn,
hess_fcn, opt);Error in opf_execute (line 86) [results, success, raw]
= mipsopf_solver(om, mpopt);Error in opf (line 232) [results, success,
raw] = opf_execute(om, mpopt);Error in runopf (line 75)[r, success] =
opf(casedata, mpopt);*
I used the following script to run OPF:
*mpc = loadcase('case9Q.m');mpc = toggle_branch_loss(mpc, 'on');results =
runopf(mpc);*
with the following functions:
*function mpc = toggle_branch_loss(mpc, on_off)%fcn toggle_loss_costfcn*
*%mpc = toggle_branch_loss(mpc, on_off) %% add callback functions
%% note: assumes all necessary data included in 1st arg (mpc, om, results)
%% so, no additional explicit args are needed if
strcmp(upper(on_off), 'ON') mpc = add_userfcn(mpc, 'formulation',
@userfcn_branch_loss_formulation); mpc.userfcn.status.branch_loss =
1;elseif strcmp(upper(on_off), 'OFF') mpc = remove_userfcn(mpc,
'formulation', @userfcn_branch_loss_formulation);
mpc.userfcn.status.branch_loss = 0;elseif strcmp(upper(on_off), 'STATUS')
if isfield(mpc, 'userfcn') && isfield(mpc.userfcn, 'status') && ...
isfield(mpc.userfcn.status, 'branch_loss') mpc =
mpc.userfcn.status.branch_loss; else mpc = 0; endelse
error('toggle_reserves: 2nd argument must be ''on'', ''off'' or
''status''');end%%----- formulation
--------------------------------------------------function om =
userfcn_branch_loss_formulation(om, mpopt, args)%om =
branch_loss_formulation(om, mpopt, args)%% add cost to the modelfcn1 =
@(x)branch_loss_costfcn(x);om.add_nln_cost('mycost1', 1, fcn1);endend*
As you can see *toggle_branch_loss.m *consist of only one callback function
- *userfcn_branch_loss_formulation.m* because in my case I dont need to
define other callback funtions (I guess).
Here is the my additional cost function *branch_loss_costfcn.m* which is
used as a hande function in a solver stage (as I understand). So I'm trying
to add a cost function to reduce transmission system losses min
*Ploss=Gij*(Vi^2+Vj^2-2*Vi*Vj*cos**θij)*:
*function f = branch_loss_costfcn(x)%fcn branch_loss_costfcn determines an
additional multi-variable cost fcn%f = branch_loss_costfcn(x)%% get some
datampc = om.get_mpc();[baseMVA, gen, gencost, bus, branch] =
deal(mpc.baseMVA, mpc.gen, mpc.gencost, mpc.bus, ... mpc.branch);%%
problem dimensionsnxyz = length(x); %% total number of control vars
of all typesnn=length(branch(:,1)); %% total number of
branchesnft=zeros(nn,2);nft(:,1)=mpc.branch(:,1);nft(:,2)=mpc.branch(:,2);%%
grab Va & VmVa = x(vv.i1.Va:vv.iN.Va <http://v.iN.Va>); %% voltage
anglesVm = x(vv.i1.Vm:vv.iN.Vm); %% voltage magnitudes%% build admittance
matrices[Ybus, Yf, Yt] = makeYbus(mpc.baseMVA, mpc.bus, mpc.branch);%% add
cost for my variablef=0;for
i=1:nnf=f+real(Ybus(nft(i,1),nft(i,2)))*(Vm(nft(i,1))+Vm(nft(i,2))-2*Vm(nft(i,1))*Vm(nft(i,2))*cos(abs(Va(nft(i,1))-Va(nft(i,2)))));endend*
I still do not understant how is that possible that *branch_loss_costfcn.m
*causes the error 'Too many output arguments' because f (output) is a
scalar.
Also according to the descriptoin of *add_nln_cost.m, *additional cost
function should take the following form: *f=fcn(x)* or *[f, df]=fcn(x)*,
or *[f, df, d2f]=fcn(x)*. So X - the optimization vector - is only one one
(allowed) input vector. Is that possible to use *om.get_mpc()* and define
Ybus, ..., etc inside *branch_loss_costfcn.m *as I did* ?*
Big thanks in advance for your time ! This case in highly important for
my research and I appriciate for any information of you regarding this !
чт, 28 янв. 2021 г. в 19:00, Ray Daniel Zimmerman <r...@cornell.edu>:
> Hi Veniamin,
>
> The legacy cost function only supports costs of a particular form, which I
> don’t believe yours can be made to fit. So the only option is to add a
> non-linear cost. Unfortunately, I never got around to implementing the
> direct specification of non-linear costs (like we have for non-linear
> constraints), so you’ll have to do it via a callback function and the use
> of add_nln_cost().
>
> If you look in opf_setup.m, you can see examples of two functions that
> are used with add_nln_cost(), namely, opf_gen_cost_fcn() for handling
> higher order polynomial generator costs, and opf_legacy_user_cost_fcn()
> for the general legacy user cost. In both cases, a function handle that
> only takes *x* as an input is created before passing it to add_nln_cost().
>
> You’ll want to have a look at some of the toggle_*.m functions, like
> toggle_reserves.m to see how to create the callback functions (see
> Sections 7.3-7.6 in the manual).
>
> I hope this helps, but don’t hesitate to ask if you have more specific
> questions in the process. I am working to make these sorts of
> customizations simpler in the future.
>
> Ray
>
>
> On Jan 27, 2021, at 4:31 AM, Вениамин Сукровицин <venya.s...@gmail.com>
> wrote:
>
> Dear Dr Zimmerman and Matpower Community,
>
> Despite the ways to add user defined variables, cost and constraints have
> been already discussed, I haven't found a solution for my case.
> I'm working on multi-objective reactive optimal power flow (ROPF) and want
> to expand a Matpower cost function which minimizes a cost of reactive power
> generation with a standard cost function for ROPF (attached picture).
> As you can see, the additional cost function depends on voltages and
> branch angles.
> The question is what is the best way to implement this cost function ?
> I tried to use add_nln_cost but failed because i wasnt able to determine
> an appropriate handle function.
> I also tried this approach (
> https://www.mail-archive.com/matpower-l@cornell.edu/msg05517.html),
> which describes (as i understand) a legacy cost specification for Vm.
>
> But I want to implement multi-variable cost as I described above.
>
> I will be very grateful for your advice and examples.
>
> Big thanks in advance!
>
> --
>
> Veniamin S.
> <F.JPG>
>
>
>
--
Veniamin S.
