matlab lsqcurvefit 非线性拟合

非线性曲线拟合是已知输入向量xdata和输出向量ydata，并且知道输入与输出的函数关系为ydata=F(x, xdata)，但不知道系数向量x。今进行曲线拟合，求x使得输出的如下最小二乘表达式成立：

例 1

拟合如下非线性函数

函数曲线如下所示。

fit_fun.m

function diff = fit_fun(param, X)

a = param(1);
m = param(2);
c = param(3);
diff = c + 1./(1 + X.*a).^m;

main.m，调用lsqcurvefit函数，完成非线性数据拟合。

clear all;
close all;
clc;


X = 0:0.01:50;

a = 0.15;
m = 1.8;
c = 0.1;
Y = c + 1./(1 + X.*a).^m;

param_0 = [1 1 1];

param = lsqcurvefit(@fit_simp, param_0, X, Y)

Y_new = param(3) + 1./(1 + X.*param(1)).^param(2);

plot(X, Y, '*r', X, Y_new, 'b');

拟合结果如下：

>> x =  0.1500    1.8000    0.1000

例 2

拟合如下三维曲线

fun.m

function diff = fit_simp(param, Xin)

a = param(1);
b = param(2);
m = param(3);
c = param(4);

x = Xin(:, 1);
y = Xin(:, 2);

diff = c + 1./(1 + x.*a + y.*b).^m;

main.m

xdata = 0:1:50;
ydata = 0:1:40;

Xin = zeros(length(xdata),2);
Xin(:, 1) = xdata;
Xin(:, 2) = linspace(min(ydata), max(ydata), length(xdata));

a = 0.15;
b = 0.1;
m = 1.8;
c = 0.1;
zdata = c + 1./(1 + Xin(:, 1).*a + Xin(:, 2).*b).^m;

param0 = [1 1 1 1];

param = lsqcurvefit(@fit_simp_3, param0, Xin, zdata)

Z_new = param(4) + 1./(1 + Xin(:, 1).*param(1) + Xin(:, 2).*param(2)).^param(3);

plot3(Xin(:,1), Xin(:,2), zdata - Z_new, '-*b');

拟合结果为：

>> param = [ 6.3753 -7.7816 1.8 0.1]

拟合残差如图所示。由图可知残差非常小，说明拟合效果理想。

注：由于初始参数的设置问题，拟合出的参数与实际参数并不一致， 这是因为lsqcurvefit求解的并不是全局最优解，而是局部最优。