www.gusucode.com > 混沌时间序列进行深度学习matlab编程 程序源码 > Main_Volterra_MultiStepPred.m
% 混沌时间序列的 volterra 预测(多步预测) -- 主函数 % 使用平台 - Matlab6.5 / Matlab7.0 % 作者:陆振波,海军工程大学 % 欢迎同行来信交流与合作,更多文章与程序下载请访问我的个人主页 % 电子邮件:luzhenbo@yahoo.com.cn % 个人主页:http://luzhenbo.88uu.com.cn clc clear close all %--------------------------------------------------- % 产生混沌序列 sigma = 10; % Lorenz 方程参数 a b = 8/3; % b r = 34; % c y = [-1,0,1]; % 起始点 (1 x 3 的行向量) h = 0.01; % 积分时间步长 k1 = 6000; % 前面的迭代点数 k2 = 5000; % 后面的迭代点数 (总样本数) z = LorenzData(y,h,k1+k2,sigma,r,b); X = z(k1+1:end,1); X = normalize_a(X,1); % 信号归一化到均值为0,振幅为1 %---------------------------------------------------- train_num = 500; % 训练样本数 test_num = 1000; % 测试样本数 %---------------------------------------------------- % 混沌序列的相空间重构 (phase space reconstruction) tau = 10 m = 3 p = 3 X = X(1:train_num+test_num); [xn_train,dn_train] = PhaSpaRecon(X(1:train_num),tau,m); [xn_test,dn_test] = PhaSpaRecon(X(train_num+1:train_num+test_num),tau,m); %---------------------------------------------------- [Wn,err_mse1] = volterra_train_lu(xn_train,dn_train,p); err_mse1 = err_mse1/var(X) %---------------------------------------------------- % 多步预测 len_pred = 300; x_start = X(train_num-(m-1)*tau:train_num); dn_pred = zeros(len_pred,1); for i=1:len_pred xn_start = PhaSpaRecon(x_start,tau,m); dn_pred(i) = volterra_test(xn_start,p,Wn); x_start = [x_start(2:end);dn_pred(i)]; end dn_test = X(train_num+1:train_num+len_pred); %---------------------------------------------------- % 作图 plot(train_num+1:train_num+len_pred,dn_test,'r',... train_num+1:train_num+len_pred,dn_pred,'b'); legend('真实值','预测值',0);