matlab-基于模糊小波神经网络FWNN的训练和预测matlab仿真-源码

clc; clear; close all; warning off; addpath(genpath(pwd)); %**************************************************************************** %更多关于matlab和fpga的搜索“fpga和matlab”的CSDN博客: %matlab/FPGA项目开发合作 %https://blog.csdn.net/ccsss22?type=blog %**************************************************************************** % 关键变量 d = 5; % 用于输入的宽度 m = d; % 输入信号的个数 n = 5; % 关系函数的个数，模糊判断的个数，小波函数的个数 epoch = 5000; % 迭代次数 num_yangben = 49; % 数据个数 num_test = 12; rate = 0.08; % 学习速率 mom = 0.5; % 冲量 % 产生试验数据 data = indata(); %result = plant(data); result = data(:,d+1); % TEST file_yangben = '测试集.dat'; fid = fopen(file_yangben); %u = fread(fid,[size_input_x,size_input_y],'float'); u_test = dlmread(file_yangben,','); fclose(fid); % 随机初始化各个参数于（0，1） c = rand(m, n); q = rand(m, n); % 注意：不能为零 a = rand(n, m); b = rand(n, m); w = rand(1, n); % t-1迭代的参数值 pc = c; pq = q; pa = a; pb = b; pw = w; % t+1迭代的参数值 nc = zeros(m, n); nq = zeros(m, n); na = zeros(n, m); nb = zeros(n, m); nw = zeros(1, n); % 用于画图的数据 tu = zeros(epoch, num_yangben); E = zeros(epoch, num_yangben); % 训练过程 tic % 开始计时 for loop1 = 1 : 1 : epoch for loop2 = 1 : 1 : num_yangben % 初始化中间数据 x = zeros(1, m); g = zeros(m, n); U = zeros(1, n); p = zeros(1, n); W = zeros(1, n); %y = zeros(1, n); % 给输入节点赋值 for i = 1 : 1 : d x(i) = data(loop2,i); end %for i = 1 : 1 : d % x(m + 1 - i) = result(loop2 - i); %end % 计算第二层节点的输出值 for i = 1 : 1 : m for j = 1 : 1 : n g(i, j) = relation(x(i), c(i, j), q(i, j)); end end % 计算第三层节点的输出，同时记录第三层节点的选择信息 for i = 1 : 1 : n [min, which] = fuzzy(g, m, i); U(i) = min; p(i) = which; end % 计算第四层节点的输出 for i = 1 : 1 : n for j = 1 : 1 : m W(i) = W(i) + wavelet(x(j), a(i, j), b(i, j)); end end y = w .* W; % 计算最终的输出 u = defuzz(U, y, n); tu(loop1, loop2) = u; % 计算误差 temp1 = u - result(loop2); E(loop1, loop2) = temp1^2 / 2; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 下面使用梯度下降算法修正参数 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % 计算E对w的偏导数 temp2 = sigamau(U, n); Etow = zeros(1, n); sima = zeros(1, n); for i = 1 : 1 : n Etow(i) = temp1 * U(i) * W(i) / temp2; sima(i) = temp1 * U(i) * w(i) / temp2; end % 计算几个下面会重复使用的数据 temp3 = zeros(n, m); temp4 = zeros(n, m); for i = 1 : 1 : n for j = 1 : 1 : m temp3(i, j) = varz2(x(j), a(i, j), b(i, j)); % temp3= z^2 temp4(i, j) = exp(-temp3(i, j)/2) / sqrt(abs(a(i, j))^3); % temp4 = ? end end % 计算E对a的偏导数 Etoa = zeros(n, m); for i = 1 : 1 : n for j = 1 : 1 : m temp5 = temp3(i, j); % temp5 = z^2 Etoa(i, j) = sima(i) * (3.5 * temp5 - temp5^2 - 0.5) * temp4(i , j); end end % 计算E对b的偏导数 Etob = zeros(n, m); for i = 1 : 1 : n for j = 1 : 1 : m temp5 = temp3(i, j); % temp5 = z^2 Etob(i, j) = sima(i) * (3 * temp5 - temp5^3) * temp4(i , j); end end % 计算E对c、q的偏导数 utoU = zeros(1, n); for i = 1 : 1 : n utoU(i) = (y(i) - u) / temp2; % temp2 = sigama(U) end [Utoc, Utoq] = Utocq(g, x, c, q, m, n, p); Etoc = zeros(m, n); Etoq = zeros(m, n); for i = 1 : 1 : m for j = 1 : 1 : n temp6 = temp1 * utoU(j); Etoc(i, j) = temp6 * Utoc(i, j); Etoq(i, j) = temp6 * Utoq(i, j); end end % 对参数修正 nw = w - rate * Etow + mom * (w - pw); na = a - rate * Etoa + mom * (a - pa); nb = b - rate * Etob + mom * (b - pb); nc = c - rate * Etoc; nq = q - rate * Etoq; % 修改参数t-1，t pw = w; w = nw; pa = a; a = na; pb = b; b = nb; c = nc; q = nq; end end toc % 结束计时，并显示时间 figure(1) % 图形显示统计信息 k = 1 : 1 : num_yangben; ttu = tu(epoch, :); plot(k, result, '-', k, ttu, '-r') legend('样本集结果', '样本集预测值') title('训练结果'); xlabel('样本'); ylabel('优先级'); %%============================================================' %%测试 %%=========================================================== for loop2 = 1 : 1 : num_test % 初始化中间数据 x = zeros(1, m); g = zeros(m, n); U = zeros(1, n); p = zeros(1, n); W = zeros(1, n); %y = zeros(1, n); % 给输入节点赋值 for i = 1 : 1 : d x(i) = u_test(loop2,i); end %for i = 1 : 1 : d % x(m + 1 - i) = result(loop2 - i); %end % 计算第二层节点的输出值 for i = 1 : 1 : m for j = 1 : 1 : n g(i, j) = relation(x(i), c(i, j), q(i, j)); end end % 计算第三层节点的输出，同时记录第三层节点的选择信息 for i = 1 : 1 : n [min, which] = fuzzy(g, m, i); U(i) = min; p(i) = which; end % 计算第四层节点的输出 for i = 1 : 1 : n for j = 1 : 1 : m W(i) = W(i) + wavelet(x(j), a(i, j), b(i, j)); end end y = w .* W; % 计算最终的输出 result_test(loop2) = defuzz(U, y, n); end figure(2) % 图形显示测试信息 k = 1 : 1 : num_test; plot(k, u_test(:,6), 'g',k, result_test, 'r') legend('测试集结果', '测试集预测值') title('测试结果'); xlabel('样本'); ylabel('优先级');