基于神经网络的控制器设计与仿真

Posted 2022-01-12 fpga&matlab

        这个部分，主要是通过给定模型的输入和输出，然后通过网络进行训练，得到神经网络的辨识参数。

        对于网络辨识部分，其基本构架如下所示：

        上述的辨识结构，通过控制对象的输入和输出的延迟分别进入F网络和G网络，然后通过网络输出和实际输出的误差对网络F和网络G进行在线学习。

        具体理论论文讲的比较详细了，这里不再重复，仿真效果如下所示：

 部分核心程序：

clc;
clear all;
close all;
warning off;
addpath 'func\\'
pack;

Num_In     = 6;
Num_Hidden = 12;
Num_Out    = 1;   

load func\\result3.mat

[In,Out] = func_data(y3,y_out3);

parameter;

while Error>Err_goal & Iter < Max_iter
      u_delay1 = 0;
      u_delay2 = 0;
      u_delay3 = 0;
      u_delay4 = 0;
      y_delay1 = 0;
      y_delay2 = 0;
      y_delay3 = 0;   
      y_delay4 = 0;
      Err_tmp  = 0;
      for k=1:All_Length
          
          
          Data_Delays = [y_delay1;
                         y_delay2;
                         y_delay3;
                         u_delay2;
                         u_delay3;
                         u_delay4];
                     
          %隐层求和 
          [Y_hidden(k),Hidden1G,Hidden2G,Y_hiddenG(k),Hidden1F,Hidden2F,Y_hiddenF(k)] = func_Hiddern(Data_Delays,u_delay1,Num_Hidden,G_wight_In,G_wight_Inb,G_wight_Out,G_wight_Outb,F_wight_In,F_wight_Inb,F_wight_Out,F_wight_Outb);
          
          Err_tmp     = Out(k)-Y_hidden(k);

          %F和G网络 
          %G神经网络计算
          [dg_weight_in,dg_bweight_in,dg_weight_out,dg_bweight_out] = func_G_net(Err_tmp,Out(k),In(k),Y_hidden(k),G_wight_Out,Hidden2G,Data_Delays,Y_hiddenG(k),Hidden1G,Num_Hidden,Num_In); 
          %F神经网络计算
          [df_weight_in,df_bweight_in,df_weight_out,df_bweight_out] = func_F_net(Err_tmp,Out(k),In(k),Y_hidden(k),F_wight_Out,Hidden2F,Data_Delays,Y_hiddenG(k),Hidden1F,Num_Hidden,Num_In);

          %G网络权值更新
          [G_wight_In,G_wight_Out,G_wight_Inb,G_wight_Outb]=func_G_W_updata(Learn_Rate,alpha,...
                                                                            G_wight_In1,G_wight_Out1,G_wight_Inb1,F_wight_Outb1,...
                                                                            dg_weight_in,dg_weight_out,dg_bweight_in,dg_bweight_out,...
                                                                            G_wight_Outb1,...
                                                                            G_wight_In2,G_wight_Out2,G_wight_Inb2,G_wight_Outb2);
          %F网络权值更新
          [F_wight_In,F_wight_Out,F_wight_Inb,F_wight_Outb]=func_F_W_updata(Learn_Rate,alpha,...
                                                                           G_wight_In1,F_wight_Out1,F_wight_Inb1,F_wight_Outb1,...
                                                                           df_weight_in,df_weight_out,df_bweight_in,df_bweight_out,...
                                                                           G_wight_In2,F_wight_Out2,F_wight_Inb2,F_wight_Outb2);
                                                        
          %延迟
          u_delay4 = u_delay3;
          u_delay3 = u_delay2;
          u_delay2 = u_delay1;
          u_delay1 = In(k); 
        
          y_delay4 = y_delay3;
          y_delay3 = y_delay2;
          y_delay2 = y_delay1;
          y_delay1 = Out(k); 

          G_wight_In2   = G_wight_In1;
          G_wight_In1   = F_wight_In;
          F_wight_Out2  = F_wight_Out1;
          F_wight_Out1  = F_wight_Out;
          F_wight_Inb2  = F_wight_Inb1;
          F_wight_Inb1  = F_wight_Inb;
          F_wight_Outb2 = F_wight_Outb1;
          F_wight_Outb1 = F_wight_Outb;

          G_wight_In2   = G_wight_In1;
          G_wight_In1   = G_wight_In;
          G_wight_Out2  = G_wight_Out1;
          G_wight_Out1  = G_wight_Out;
          G_wight_Inb2  = G_wight_Inb1;
          G_wight_Inb1  = G_wight_Inb;
          G_wight_Outb2 = G_wight_Outb1;
          G_wight_Outb1 = G_wight_Outb;
          
      end

      Error        = sum((Out-Y_hidden).^2)/k;
      Error2(Iter) = Error;
      %如果当前训练误差反而增加，则放弃此次训练结果，重新训练
      if (Iter > 1) & (Error2(Iter) > 1.2*Error2(Iter-1))
         Iter      = Iter;
         Break_cnt = Break_cnt + 1;
         if Break_cnt > 20
            Error2(Iter) = [];
            break;
         end
      else
         Break_cnt = 0; 
         Iter = Iter+1; 
         fprintf('迭代次数：%d  ',Iter);
         fprintf('误差：%f',Error);
         fprintf('\\n\\n');
      end
end

figure; 
subplot(121);
plot(Times(1:1000),Out(1:1000),'r');  
hold on;
plot(Times(1:1000),Y_hidden(1:1000),'b--');  
hold on;
xlabel('Times(s)');
ylabel('Out'); 
legend('系统输出','网络输出');

subplot(122); 
semilogy(1:length(Error2),Error2,'r-o'); 
xlabel('Times(s)');
ylabel('Error');
grid on;

F_wight_In0   = F_wight_In;
F_wight_Out0  = F_wight_Out; 
G_wight_In0   = G_wight_In;
G_wight_Out0  = G_wight_Out;
F_wight_Inb0  = F_wight_Inb;
F_wight_Outb0 = F_wight_Outb; 
G_wight_Inb0  = G_wight_Inb;
G_wight_Outb0 = G_wight_Outb;

save NN_reg_signal3.mat F_wight_In0 F_wight_Out0 G_wight_In0 G_wight_Out0  F_wight_Inb0 F_wight_Outb0 G_wight_Inb0 G_wight_Outb0

A08-16