Approach for Neural network

I wanted to create a Neural network that can be replica of runpf() in MATPOWER but only active and reactive power using the Neural Network fitting toolbox. is this the right approach. Later I want to use Neural network for different projects

mpc2=loadcase('case39');
load_min = 0;
load_max = mpc2.bus(:,3);
LP=zeros(39,5000);
LQ=zeros(39,5000);
V=zeros(39,5000);
%c=1;
j=1;
for i=1:5000
    load= rand .* (load_max - load_min) + load_min; % Generating Random value for load
    mpc2.bus(:,3)=load;
    PF = runpf(mpc2);
    if (PF.success)
        LP(:,j)=PF.bus(:,3);
        LQ(:,j)=PF.bus(:,4);
        L = [LP; LQ];
        V(:,j)=PF.bus(:,8);
        j =j+1;
    end
end

Answer 1

[Ritam]

You might consider framing this as a surrogate that learns the mapping from load inputs (P,Q) to state outputs (bus voltage magnitudes and angles) rather than predicting P/Q themselves. Your data-generation loop seems fine for building a dataset from successful runpf cases; you could treat inputs as X = [LP; LQ] and targets as Y = [V; θ], with per-bus normalization and a train/validation/test split. A multi‑output regression network (e.g., NN Fitting Toolbox/fitnet with a few hidden layers and MSE loss) might be a practical starting point, evaluated against runpf with RMSE/MAE. For broader applicability, you may want to include operating conditions (generator setpoints, taps, topology) in the inputs or use the NN primarily as a fast initializer with runpf performing the final solve.