MATLAB强化学习训练simulink模型优化

万木逢春

发布于 2020-04-14 15:57:12

2.4K7

发布于 2020-04-14 15:57:12

文章被收录于专栏：帮你学MatLab帮你学MatLab

这是训练得到的效果

simulink物理结构模型用的是前面文章matlab倒立摆环境建模当在solidworks绘制后导入的，去除子模块封装，添加驱动以及传感器接口，就像这个样子

强化学习模型参考前面的训练文章MATLAB强化学习训练simulink模型，大体形式不变

但是在判断是否结束的环节添加杆的旋转角度限制，如果杆在持续转圈就停止

实现的效果是如果杆在某一方向旋转弧度超过6rad就停止、同时给reward加上-100，绝对值模块就是为了把顺时针和逆时针的旋转角度统一

添加方法和滑块的位置限制类似

再用OR逻辑模块把滑块超出固定横杆的判定和持续旋转的判定合并处理

模型建立后还需要将env的输入输出告知agent

%%
mdl = 'CartPole_RL';
open_system(mdl)
%%
actionInfo = rlNumericSpec([1 1],...
    'LowerLimit',-0.5,...
    'UpperLimit',0.5);
actionInfo.Name = 'force';
%%
observationInfo = rlNumericSpec([5 1],...
    'LowerLimit',-Inf,...
    'UpperLimit',Inf);
observationInfo.Name = 'observations';
%%
env = rlSimulinkEnv(mdl,[mdl '/RL Agent'],observationInfo,actionInfo);
这样就会在simulink模型文件中绑定agent模块了，接下来就是设置agent参数
%%
numObservations = observationInfo.Dimension(1);
%% 
Ts = 0.02;
Tf = 25;
rng(0)
%% 初始化agent
statePath = [
    imageInputLayer([numObservations 1 1],'Normalization','none','Name','observation')
    fullyConnectedLayer(128,'Name','CriticStateFC1')
    reluLayer('Name','CriticRelu1')
    fullyConnectedLayer(200,'Name','CriticStateFC2')];
actionPath = [
    imageInputLayer([1 1 1],'Normalization','none','Name','action')
    fullyConnectedLayer(200,'Name','CriticActionFC1','BiasLearnRateFactor',0)];
commonPath = [
    additionLayer(2,'Name','add')
    reluLayer('Name','CriticCommonRelu')
    fullyConnectedLayer(1,'Name','CriticOutput')];
criticNetwork = layerGraph(statePath);
criticNetwork = addLayers(criticNetwork,actionPath);
criticNetwork = addLayers(criticNetwork,commonPath);
criticNetwork = connectLayers(criticNetwork,'CriticStateFC2','add/in1');
criticNetwork = connectLayers(criticNetwork,'CriticActionFC1','add/in2');
figure
plot(criticNetwork)
criticOptions = rlRepresentationOptions('LearnRate',1e-03,'GradientThreshold',1);
critic = rlRepresentation(criticNetwork,observationInfo,actionInfo,...
    'Observation',{'observation'},'Action',{'action'},criticOptions);
actorNetwork = [
    imageInputLayer([numObservations 1 1],'Normalization','none','Name','observation')
    fullyConnectedLayer(128,'Name','ActorFC1')
    reluLayer('Name','ActorRelu1')
    fullyConnectedLayer(200,'Name','ActorFC2')
    reluLayer('Name','ActorRelu2')
    fullyConnectedLayer(1,'Name','ActorFC3')
    tanhLayer('Name','ActorTanh1')
    scalingLayer('Name','ActorScaling','Scale',max(actionInfo.UpperLimit))];
actorOptions = rlRepresentationOptions('LearnRate',5e-04,'GradientThreshold',1);
actor = rlRepresentation(actorNetwork,observationInfo,actionInfo,...
    'Observation',{'observation'},'Action',{'ActorScaling'},actorOptions);
agentOptions = rlDDPGAgentOptions(...
    'SampleTime',Ts,...
    'TargetSmoothFactor',1e-3,...
    'ExperienceBufferLength',1e6,...
    'MiniBatchSize',128);
agentOptions.NoiseOptions.Variance = 0.4;
agentOptions.NoiseOptions.VarianceDecayRate = 1e-5;
agent = rlDDPGAgent(actor,critic,agentOptions);
%% 设置训练参数
maxepisodes = 2000;
maxsteps = ceil(Tf/Ts);
trainingOptions = rlTrainingOptions(...
    'MaxEpisodes',maxepisodes,...
    'MaxStepsPerEpisode',maxsteps,...
    'ScoreAveragingWindowLength',5,...
    'Verbose',false,...
    'Plots','training-progress',...
    'StopTrainingCriteria','AverageReward',...
    'StopTrainingValue',400,...
    'SaveAgentCriteria','EpisodeReward',...
    'SaveAgentValue',400);
%% 并行学习设置
trainingOptions.UseParallel = true;
trainingOptions.ParallelizationOptions.Mode = "async";
trainingOptions.ParallelizationOptions.DataToSendFromWorkers = "Experiences";
trainingOptions.ParallelizationOptions.StepsUntilDataIsSent = -1;
然后就是漫长的训练时间
%% 训练
trainingStats = train(agent,env,trainingOptions);
%% 结果展示
simOptions = rlSimulationOptions('MaxSteps',500);
experience = sim(env,agent,simOptions);
totalReward = sum(experience.Reward);
% bdclose(mdl)