Fitting of mathematical modeling

1 Overview of fitting method

Different from the interpolation problem, in the fitting problem it is not required that the curve must pass through a given point. The goal of the fitting problem is to find a function (curve) so that the curve is closest to all data points under a certain criterion, that is, the curve fits best (minimizes the loss function)

A small example

There is an existing file data1.xlsxthat saves the values ​​of x and y. Use Matlab to draw the graph as follows:

❗️Note:

• When importing data into Matlab, we created two variables respectively: x and y. Each variable only saves one column of data and saved the variables in a file data1.mat.

So next, how do we determine the fitting curve ?

• Send the sample to Matlab to draw the picture

2 Least Squares Method

2.1 Least squares geometric interpretation

2.2 Least squares solution

Use Matlab to solve least squares

3 Algorithm evaluation

After the fitting is completed, how do we evaluate the quality of the fitting?

How to determine a function that is linear to its parameters?

• In a function, parameters only appear in the form of a power of one, and cannot be multiplied or divided by any other parameters, and the
  composite function form of parameters cannot appear.

❗️Note:

• Although the fitting effect of a multivariate function may be better than that of a univariate function, our fitting function should be as simple as possible. For example, in extreme cases, polynomial interpolation is used. The SSE is obviously 0, but this conflicts with our original intention of using fitting.

Calculate goodness of fit using Matlab

4 Code writing

Use the least squares method to find k and b

clear;clc
load  data1
plot(x,y,'o')
% 给x和y轴加上标签
xlabel('x的值')
ylabel('y的值')
n = size(x,1);
k = (n*sum(x.*y)-sum(x)*sum(y))/(n*sum(x.*x)-sum(x)*sum(x))
b = (sum(x.*x)*sum(y)-sum(x)*sum(x.*y))/(n*sum(x.*x)-sum(x)*sum(x))
hold on % 继续在之前的图形上来画图形
grid on % 显示网格线

% 画出y=kx+b的函数图像 plot(x,y)
% 方法一：传统的画法：模拟生成x和y的序列，比如要画出[0,5]上的图形
% xx = 2.5: 0.1 :7  % 间隔设置的越小画出来的图形越准确
% yy = k * xx + b  % k和b都是已知值
% plot(xx,yy,'-')

% 方法二：匿名函数的基本用法。
% handle = @(arglist) anonymous_function
% 其中handle为调用匿名函数时使用的名字。
% arglist为匿名函数的输入参数，可以是一个，也可以是多个，用逗号分隔。
% anonymous_function为匿名函数的表达式。
% 举个小例子
%  z=@(x,y) x^2+y^2; 
%  z(1,2) 
% % ans =  5
% fplot函数可用于画出匿名一元函数的图形。
% fplot(f,xinterval) 将匿名函数f在指定区间xinterval绘图。xinterval =  [xmin xmax] 表示定义域的范围
f=@(x) k*x+b;
fplot(f,[2.5,7]);
legend('样本数据','拟合函数','location','SouthEast')

Calculate goodness of fit

y_hat = k*x+b; % y的拟合值
SSR = sum((y_hat-mean(y)).^2)  % 回归平方和
SSE = sum((y_hat-y).^2) % 误差平方和
SST = sum((y-mean(y)).^2) % 总体平方和
SST-SSE-SSR   % 5.6843e-14  =   5.6843*10^-14   matlab浮点数计算的一个误差
R_2 = SSR / SST

5 Curve Fitter

The result obtained using the curve fitter here is the same as the result we achieved using the code before.

Below we use a curve fitter to predict the US population

❗️ NOTE:

• The fitting function is not linear

Reference Code:

clear;clc
year = 1790:10:2000;
population = [3.9,5.3,7.2,9.6,12.9,17.1,23.2,31.4,38.6,50.2,62.9,76.0,92.0,106.5,123.2,131.7,150.7,179.3,204.0,226.5,251.4,281.4];
plot(year,population,'o')
cftool  % 拟合工具箱
% (1) X data 选择 year
% (2) Y data 选择 population
% (3) 拟合方式选择：Custom Equation (自定义方程)
% (4) 修改下方的方框为：x = f(t) = xm/(1+(xm/3.9-1)*exp(-r*(t-1790)))
% (5) 左边的result一栏最上面显示：Fit computation did not converge:即没有找到收敛解，右边的拟合图形也表明拟合结果不理想
% (6) 点击Fit Options，修改非线性最小二乘估计法拟合的初始值(StartPoint), r修改为0.02，xm修改为500 
% (7) 此时左边的result一览得到了拟合结果：r = 0.02735, xm = 342.4
% (8) 依次点击拟合工具箱的菜单栏最左边的文件—Generate Code(导出代码到时候可以放在你的论文附录)，可以得到一个未命名的脚本文件
% (9) 在这个打开的脚本中按快捷键Ctrl+S，重命名为createFit.m 将这个文件保存到当前文件夹。
% (10) 在现在这个文件中调用这个函数得到参数的拟合值和预测的效果
[fitresult, gof] = createFit(year, population)

t = 2001:2030;
xm = 342.4;   
r =  0.02735;
predictions = xm./(1+(xm./3.9-1).*exp(-r.*(t-1790)));  % 计算预测值（注意这里要写成点乘和点除,这样可以保证按照对应元素进行计算）
figure(2)
plot(year,population,'o',t,predictions,'.')  % 绘制预测结果图

Prediction result graph:

6 Simulate the data demonstration yourself

Prerequisite knowledge:

% （1）randi : 产生均匀分布的随机整数（i = int）  
%产生一个1至10之间的随机整数矩阵，大小为2x5；
s1 = randi(10,2,5)
%产生一个-5至5之间的随机整数矩阵，大小为1x10；
s2 = randi([-5,5],1,10)

%  （2） rand: 产生0至1之间均匀分布的随机数
%产生一个0至1之间的随机矩阵，大小为1x5；
s3 = rand(1,5)
%产生一个a至b之间的随机矩阵，大小为1x5；  % a + (b-a) * rand(1,5); 如：a,b = 2,5
s4= 2 + (5-2) * rand(1,5)

% （3）normrnd:产生正态分布的随机数
%产生一个均值为0，标准差（方差开根号）为2的正态分布的随机矩阵，大小为3x4；
s5 = normrnd(0,2,3,4)

% （4）roundn—任意位置四舍五入
% 0个位 1十位  2百位 -1小数点后一位  
a = 3.1415
roundn(a,-2)    % ans   =  3.1400
roundn(a,2)      % ans   =  0
a =31415
roundn(a,2)   % ans  = 31400
roundn(5.5,0)  %6
roundn(5.5,1) %10

Generate random samples and fit them using a curve fitter

clear;clc 
x = rand(30,1) * 10;  % x是0-10之间均匀分布的随机向量（30个样本）
y = 3 * exp(0.5*x) -5 + normrnd(0,1,30,1);
% cftool