基于MLP的数据预测与误差分析matlab仿真+仿真录像

clc; clear; close all; warning off; addpath(genpath(pwd)); N = 100; x = linspace(0, 1, N)'; t = x.^2 + unifrnd(-0.1, 0.1, N, 1); x = [ones(size(x, 1),1) x]; % t = [1 1 1 -1 -1 -1]; feature_number = size(x,2); w_hidden_node_number = 10; w_output_node_number= 1; % regression problem w_hidden = rand(feature_number + 1, w_hidden_node_number); w_output = rand(w_hidden_node_number + 1, w_output_node_number); w = rand(2, 1); n = 0.05; E = []; for i = 1:500 i y_hidden = tanh([ones(size(x, 1), 1) x] * w_hidden); y_output = logsig([ones(size(y_hidden, 1), 1) y_hidden] * w_output); y = y_output; e = t - y; % Backpropagation hidden layer delta_w_output = ((e .*(y_output.*(1-y_output))) .* [ones(size(y_hidden, 1), 1) y_hidden]); delta_w_hidden = (delta_w_output(:, 2:end) .* ((1-y_hidden.^2)))' * [ones(size(x, 1), 1) x]; w_output = w_output + n * sum(delta_w_output)'; w_hidden = w_hidden + n * delta_w_hidden'; E(i) = mse(e); subplot(2, 1, 1) plot(x(:, 2), t, '.b', x(:, 2), logsig([ones(N, 1) tanh([ones(N, 1) x] * w_hidden)] * w_output), 'r'); title('Actual vs. Prediction'); xlabel('x'); ylabel('y'); legend('Actual', 'Predicted'); subplot(2,1,2); plot(E); title('Error vs. Iteration'); xlabel('Iteration (n)'); ylabel('MSE'); drawnow; end