可编程线性二次调节器simulink实现.rar

%% ANN-based programmable LQR for inverter-fed induction motor % % Bartlomiej Ufnalski % Institute of Control and Industrial Electronics % Warsaw University of Technology % % The idea (dated back to 2006) is to use linear-quadratic regurator (LQR) % concept to synthesize optimal controller for plants with the non-constant % state matrix. It is assumed that this matrix changes with working point. % An induction motor can serve here as an example. % % Basically, this model illustrates the concept described in detail here: % http://dx.doi.org/10.1109/IJCNN.2006.246811 % (Neural-Network-based Programmable State Feedback Controller for % Induction Motor Drive) % % rev. 2014.10.02 for Matlab Central % clear all training = 0; % this submission already includes ANN stored in .mat --> you can run the model skipping the training stage; % training = 1 is needed if you experiment with different Q % and R matrices (see: % http://www.mathworks.com/help/control/ref/lqr.html) with_noise = 1; % % Plant parameters (induction motor) Ls = 0.1004; Lr = 0.0969; Lm = 0.0915; sigma = 1-Lm*Lm/Lr/Ls; lambda = Ls*Lr-Lm*Lm; Rr = 1.294; Tr = Lr/Rr; Ts = 1e-4; Tp = Ts; Rs = 1.54; Rs_ident_error = 1; % Rs in EMF estimation = Rs * Rs_ident_error wc = 1/Ts; pb = 3; psi_ref = 0.7; J = 0.15; slip_max = 280; omega_max = 300; % omega flux % if training == 0, load net_lqr.mat open('sfoc_lqr_ann.slx'); disp('Simulating...'); set_param('sfoc_lqr_ann','SimulationCommand','start'); return end % % The ststes: isx isy psisx omegam p1-psi p2-omega % k = 1; disp('LQR: calculating gain matrices K for different slips at different speeds...'); for a1 = -omega_max:20:omega_max, % disp(['LQR: calculating gain matrices for different slips at ',num2str(a1),'rad/s']); for a2 = -slip_max:20:slip_max, A = [-(Rr*Ls+Rs*Lr)/lambda a2 Rr/lambda 0 0 0; -a2 -Rr*Ls/lambda a2*Lr/lambda 0 0 0; -Rs 0 0 0 0 0; 0 0 -1/J*3/2*pb*psi_ref*a1/Rs 0 0 0; 0 0 1 0 0 0; 0 0 0 1 0 0]; B = [Lr/lambda 0; 0 0; 1 0; 0 1/J*3/2*pb*psi_ref/Rs; 0 0; 0 0]; % % Q and R by guessing and checking. The Bryson's rule might be helpful. % Some hints on choosing Q in augmented systems (i.e. with % auxiliary state variables) can be found here: % http://dx.doi.org/10.1109/TIE.2014.2334669 % (Particle Swarm Optimization of the Multi-oscillatory LQR for a 3-phase 4-wire Voltage Source Inverter with an LC Output Filter % AKA Particle Swarm Optimization of the Multi-Oscillatory LQR % for a Three-Phase Four-Wire Voltage Source Inverter with an LC Output Filter) % or here: % State-Space Current Control For Four-Leg Grid-Connected PWM % Rectifiers With Active Power Filtering Function (PEMC 2014) % or soon here: http://pe.org.pl/last.php?lang=1 (A. Galecki, % B. Ufnalski, A. Kaszewski, L.M. Grzesiak). % Q = diag([1/100 1/100 1/0.7^2 1/10000 1/0.7^2 1/10000].*[0.1 0.1 0.002 0.1 100000 100]); R = diag([1/40000 1/40000].*[0.01 0.01]); % [Klqr,S,E] = lqr(A,B,Q,R); inputANN(:,k) = [a1;a2]; outputANN(:,k) = [Klqr(1,:)';Klqr(2,:)']; k = k+1; end end % norm_input = max(abs(inputANN'))'; norm_inputANN = repmat(norm_input',max(size(inputANN)),1)'; norm_output = max(abs(outputANN'))'; norm_outputANN = repmat(norm_output',max(size(outputANN)),1)'; P = inputANN./norm_inputANN; T = outputANN./norm_outputANN; s1 = 3; s2 = 5; [s3,pom] = size(T); % net = feedforwardnet([s1,s2]); net.layers{1}.transferFcn = 'tansig'; net.layers{2}.transferFcn = 'tansig'; net.layers{3}.transferFcn = 'purelin'; net.divideFcn='dividetrain'; net.trainParam.show = 5; net.trainParam.epochs = 100; disp('Training...'); net = train(net,P,T); % gensim(net,-1); % The net has to be copied manually into .slx. Sorry for that. I could do better. net.userdata.norm = [norm_input' norm_output']; save net_lqr net % Some graphs can be plotted using lqr_ann_graphs script. % That's all folks!