传统控制器与模糊控制器用于调节隔离DCDC转换器的电压和电流matlab代码.rar

%% Inicializacion de Matlab clc, clear Ts = 1e-6; %% Modelo Simulink open("RL_TF_SS.slx") %% Creacion de entorno % Observation info obsInfo = rlNumericSpec([4 1]); obsInfo.LowerLimit = [-inf 0 0 0]'; obsInfo.UpperLimit = [inf 12 12 10]'; obsInfo.Name = "Observations"; obsInfo.Description = "Error Voltage SetPoint Control"; % Actions Info actInfo = rlNumericSpec([1 1]); actInfo.LowerLimit = 0; actInfo.UpperLimit = 10; actInfo.Name = "Voltage"; % Create environment object env = rlSimulinkEnv("RL_TF_SS", "RL_TF_SS/RL_Charger", obsInfo, actInfo); env.ResetFcn = @(in)LclRstFcn_1(in); rng(0) %% Critic criticNet = [ featureInputLayer(prod(obsInfo.Dimension)) fullyConnectedLayer(5) reluLayer fullyConnectedLayer(10) reluLayer fullyConnectedLayer(15) reluLayer fullyConnectedLayer(20) reluLayer fullyConnectedLayer(1) ]; criticNet = dlnetwork(criticNet); summary(criticNet) critic = rlValueFunction(criticNet, obsInfo); getValue(critic, {rand(obsInfo.Dimension)}) %% Actor % Input path inPath = [ featureInputLayer(prod(obsInfo.Dimension), Name="netObsIn") fullyConnectedLayer(prod(actInfo.Dimension), Name="infc") ]; % Mean value path meanPath = [ tanhLayer(Name="tanhMean") fullyConnectedLayer(10) reluLayer fullyConnectedLayer(5) reluLayer fullyConnectedLayer(prod(actInfo.Dimension)) scalingLayer( ... Name="netMout", ... Scale=actInfo.UpperLimit) ]; % Standard deviation path sdevPath = [ tanhLayer(Name="tanhStdv") fullyConnectedLayer(10) reluLayer fullyConnectedLayer(5) reluLayer fullyConnectedLayer(prod(actInfo.Dimension)) softplusLayer(Name="netSDout") ]; % Add layers to network object actorNet = layerGraph; actorNet = addLayers(actorNet, inPath); actorNet = addLayers(actorNet, meanPath); actorNet = addLayers(actorNet, sdevPath); % Connect layers actorNet = connectLayers(actorNet, "infc", "tanhMean/in"); actorNet = connectLayers(actorNet, "infc", "tanhStdv/in"); actorNet = dlnetwork(actorNet); summary(actorNet) actor = rlContinuousGaussianActor(actorNet, obsInfo, actInfo, ... ActionMeanOutputNames="netMout",... ActionStandardDeviationOutputNames="netSDout",... ObservationInputNames="netObsIn"); getAction(actor, {rand(obsInfo.Dimension)}) %% Creacion de agente agent = rlACAgent(actor, critic); agent.SampleTime = Ts; agent.AgentOptions.NumStepsToLookAhead = 32; agent.AgentOptions.DiscountFactor = 0.99; % Ajuste del factor de descuento agent.AgentOptions.CriticOptimizerOptions.LearnRate = 10e-3; % Ajuste de la tasa de aprendizaje agent.AgentOptions.CriticOptimizerOptions.GradientThreshold = 1; agent.AgentOptions.ActorOptimizerOptions.LearnRate = 10e-3; % Ajuste de la tasa de aprendizaje agent.AgentOptions.ActorOptimizerOptions.GradientThreshold = 1; %% Entrenamiento trainOpts = rlTrainingOptions(... MaxEpisodes = 206, ... MaxStepsPerEpisode = 200, ... ScoreAveragingWindowLength = 20, ... Verbose = true, ... Plots = "training-progress", ... StopTrainingCriteria = "AverageReward", ... StopTrainingValue = 10000); trainingStats = train(agent, env, trainOpts);