www.gusucode.com > 基于机动目标跟踪课题的整个算法matlab程序 > ex/Particle.m

    function Particle

% Particle filter 

x = 0.1; % 初始状态
Q = 50; % 过程噪声协方差
R = 50; % 测量噪声协方差
tf1 = 100; % 仿真长度
tf = 150;

N = 50; % 粒子滤波器粒子数

xhat = x;
P = 2;
xhatPart = x;

% 初始化粒子过滤器
for i = 1 : N
    xpart(i) = x + sqrt(P) * randn;
end

xArr = [x];
yArr = [-x^2 + sqrt(R) * randn];
xhatArr = [x];
PArr = [P];
xhatPartArr = [xhatPart];

close all;

for k =1:tf1;
    % 系统仿真
    x = -(x-50)^2 + 5*k + sqrt(Q) * randn;%状态方程
    y = -x^2 + sqrt(R) * randn;%观测方程
     %  卡尔曼滤波
    F = -2*(x-50) ;
    P = F * P * F' + Q;
    H = -xhat^2 ;
    K = P * H' * inv(H * P * H' + R);
    xhat = -(xhat-50)^2+5*k;%预测
    xhat = xhat + K * (y + xhat^2);%更新
    P = (1 - K * H) * P;
   
    for i = 1 : N
        xpartminus(i) = -(xpart(i) - 50)^2 + 5*k + sqrt(Q) * randn;
        ypart = -(xpartminus(i))^2;
        vhat = y - ypart;%观测和预测的差
        vhat0=sqrt(y.^2-ypart.^2);
        q(i) = (1 / (sqrt(R^2) * sqrt(2*pi))) * exp(-vhat^2 /( 2 * R^2));
    end
    %正常化的可能性，每个先验估计
    qsum = sum(q);
    for i = 1 : N
        q(i) = q(i) / qsum;%归一化权重
    end
    % 重采样
    for i = 1 : N
        u = rand; % 均匀随机数介于0和1
        qtempsum = 0;
        for j = 1 : N
            qtempsum = qtempsum + q(j);
            if qtempsum >= u
                xpart(i) = xpartminus(j);
                break;
            end
        end
    end
    xhatPart = mean(xpart);
    xArr = [xArr x];
    yArr = [yArr y];
    xhatArr = [xhatArr xhat];
    PArr = [PArr P];
    xhatPartArr = [xhatPartArr xhatPart];
    
    x0=100;
    xhat1 = x0;
    xhatPart1 = x0;  
    % 初始化粒子过滤器
for i = 1 : N
    xpart1(i) = x0 + sqrt(P) * randn;
end

xArr1 = [x0];
yArr1 = [3*x0 + sqrt(R) * randn];
xhatArr1 = [x0];
xhatPartArr1 = [xhatPart1];

close all;
    % 系统仿真
    x1 =  3*(x0+1)  + sqrt(Q) * randn;%状态方程
    y1 = 3*x1+ sqrt(R) * randn;%观测方程
     %  卡尔曼滤波
    F1 = 3 ;
    P1 = F1 * P * F1' + Q;
    H1 = 3*xhat1;
    K1 = P1* H1' * inv(H1 * P1 * H1' + R);
    xhat1 = 3 * (xhat1+1) ;%预测
    xhat1 = xhat1 + K1 * (y1 - 3*xhat1);%更新
    P1 = (1 - K1 * H1) * P1;
   
    for i = 1 : N
        xpartminus1(i) = 3* (xpart1(i)+1)  + sqrt(Q) * randn;
        ypart1 = 3*(xpartminus1(i));
        vhat1 = y1 - ypart1;%观测和预测的差
        vhat00=sqrt(y1.^2-ypart1.^2);
       q1(i) = (1 / (sqrt(R^2) * sqrt(2*pi))) * exp(-vhat1^2 /( 2 * R^2));
    end
    %正常化的可能性，每个先验估计
    qsum = sum(q1);
    for i = 1 : N
        q1(i) = q1(i) / qsum;%归一化权重
    end
    % 重采样
    for i = 1 : N
        u = rand; % 均匀随机数介于0和1
        qtempsum = 0;
        for j = 1 : N
            qtempsum = qtempsum + q(j);
            if qtempsum >= u
                xpart1(i) = xpartminus1(j);
                break;
            end
        end
    xhatPart1 = mean(xpart1);
    xArr1 = [xArr1 x1];
    yArr1 = [yArr1 y1];
    xhatArr1 = [xhatArr1 xhat1];
    PArr = [PArr P];
    xhatPartArr1 = [xhatPartArr1 xhatPart1];
    
t1 = 0 : tf1; t2=100:150;
      
  end
end

figure;
plot(t1, xArr, 'b.', t2, xArr1, 'b.',t1,xhatArr,'r',t2,xhatArr1,'r',t1, xhatPartArr, 'k-',t2,xhatPartArr1, 'k-');
xlabel('time step'); ylabel('state');
legend('True state','True state', 'KF', 'KF', 'Particle filter estimate');

figure;
m1=1:tf1;m2=100:150;
subplot(2,1,1),plot(m1,vhat0);
subplot(2,1,2),plot(m2,vhat00);