Subsections


Non-linear Continuous Multi-Extremal Optimization

The test programs below are all run via one CE program: CEoptima.m

The following is a batchfile to run the programs: runTHIS.m

Rosenbrock function

$\displaystyle S(\mathbf{x}) = \sum_{i=1}^{n-1} 100  (x_{i+1} - x_i^2)^2 + (x_i-1)^2$ (1)

Rosen.m

Trigonometric function

$\displaystyle S(\mathbf{x}) = 1 + \sum_{i=1}^n 8\sin^2(\eta(x_i-x_i^*)^2) + 6\sin^2(2\eta(x_i-x_i^*)^2)+\mu(x_i-x_i^*)^2 \;.$ (2)

Paviani's function

$\displaystyle \sum_{i=1}^n \bigg(\ln ^2(x_i-2) + \ln ^2(10-x_i)\bigg) - \left( \prod_{i=1}^n x_i\right)^{0.2}.$ (3)

UniformPaviani.m

Paviani.m

Sphere model, first De Jong's function

$\displaystyle \sum_{i=1}^{n}x_{i}^2$ (4)

FirstJong.m

Third De Jong's function

$\displaystyle \sum_{i=1}^{n}\vert x_i\vert , \quad -2.048\le x_i\le 2.048$ (5)

ThirdJong.m

Fourth De Jong's function, Quartic function with noise, Quartic Gaussian function

$\displaystyle \sum_{i=1}^{n}ix_{i}^4 ,\quad -1.28\leqslant x_i \leqslant 1.28$ (6)

FourthJong.m

Rastrigin's function

$\displaystyle n*10 + \sum_{i=1}^{n}(x_i^2-10\cos(2\pi x_i)) , \quad -5.12\leqslant x_i \leqslant 5.12$ (7)

Rastrigin.m

Schwefel's function

$\displaystyle \sum_{i=1}^{n}-x_i\sin(\sqrt{\vert x_i\vert}) , \quad -512\leqslant x_i \leqslant 512$ (8)

Schwefel.m

Griewangk's function

$\displaystyle -\prod_{i=1}^{n}\cos\left(\frac{x_i}{\sqrt{i}}\right)+\sum_{i=1}^{n}\frac{x_i^2}{4000}+1 , \quad -600\leqslant x_i \leqslant 600$ (9)

Griewangk.m

Sine envelope sine wave function

$\displaystyle -\sum_{i=1}^{n-1}\left(\frac{\sin^2(\sqrt{x_{i+1}^2+x_{i}^2}-0.5)...
...01(x_{i+1}^2+x_{i}^2)+1)^2}+0.5\right)  ,\quad -100\leqslant x_i \leqslant 100$ (10)

Stretch V sine wave function (Ackley)

$\displaystyle \sum_{i=1}^{n-1}(x_{i+1}^2+x_{i}^2)^{0.25}(\sin^2(50(x_{i+1}^2+x_{i}^2)^{0.1})+1)  , \quad -10\leqslant x_i \leqslant 10$ (11)

AckleySine.m

Test function (Ackley)

$\displaystyle \sum_{i=1}^{n-1}\left(3(\cos(2x_{i})+\sin(2x_{i+1}))+\frac{\sqrt{x_{i+1}^2+x_{i}^2}}{e^{0.2}}\right)  ,\quad -30\leqslant x_i \leqslant 30$ (12)

Ackleyfunction.m

Ackley's function

$\displaystyle \sum_{i=1}^{n-1}\left(20+e^{-20}e^{-0.2\sqrt{0.5(x_{i+1}^2+x_{i}^...
...cos(2\pi x_{i+1})+\cos(2\pi x_i))}\right) ,\quad -30\leqslant x_i \leqslant 30$ (13)

Egg Holder

\begin{displaymath}\begin{split}& \sum_{i=1}^{n-1}\left(-(x_{i+1}+47)\sin\left(\...
...-x_{i})\right) \ & -512\leqslant x_i \leqslant 512 \end{split}\end{displaymath} (14)

EggHolder.m

Rana's function

\begin{displaymath}\begin{split}&\sum_{i=1}^{n-1}\Big((x_{i+1}+1)\cos\left(\sqrt...
...rt}\right)\Big)  , -500\leqslant x_i \leqslant 500 \end{split}\end{displaymath} (15)

Rana.m

Pathological test function

Notice: Possibly not OK.

$\displaystyle \sum_{i=1}^{n-1}\left(\frac{\sin^2\left(\sqrt{x_{i+1}^2+100x_{i}^...
...{i+1}x_{i}+x_{i}^2)^2+1.0}+0.5\right)  , \quad -100\leqslant x_i \leqslant 100$ (16)

Pathology.m

Michalewicz's function

$\displaystyle \sum_{i=1}^{n-1}\left(\sin(x_{i+1})\sin^{20}\left(\frac{2x_{i+1}^...
...left(\frac{x_{i+1}^2}{\pi}\right)\right)  , \quad 0\leqslant x_i \leqslant \pi$ (17)

Michalewicz.m

Master's cosine wave function

$\displaystyle -\sum_{i=1}^{n-1}e^{-\frac{1}{8}\left(x_{i+1}^2+0.5x_{i}x_{i+1}+x...
...x_{i+1}^2+0.5x_{i}x_{i+1}+x_{i}^2}\right)  , \quad -5\leqslant x_i \leqslant 5$ (18)

Keane's function

$\displaystyle \left\vert\frac{ \left(\sum_{i=1}^{n}\cos^4\left(x_{i}\right)-2\p...
...{n}\cos^2\left(x_{i}\right)\right)}{ \sqrt{\sum_{i=1}^{n}ix_{i}^2}} \right\vert$ (19)

Constraints: $ \prod_{i=1}^{n}x_{i}\ge0.75, \sum_{i=1}^{n}x_{i}\leqslant
7.5 n, 0\le x_{i} \le 10$ .

Hougen function (non-linear regression)

Hougen function is typical complex test for classical non-linear regression problems. The Hougen-Watson model for reaction kinetics is an example of this a non-linear regression problem. The form of the model is

   rate$\displaystyle = \frac{\beta_1 x_2 - x_3/\beta_5}{
1 + \beta_2 x_1 + \beta_3 x_2 + \beta_4 x_3},
$

where the betas are the unknown parameters, and the xs are the input variables. The parameters are estimated via the least squares criterion. That is, the parameters are such that the sum of the squared differences between the observed responses and their fitted values of rate is minimized. The following input data are used:

Table 1: Input data for the Hougen function
x(1) x(2) x(3) rate
470 300 10 8.55
285 80 10 3.79
470 300 120 4.82
470 80 120 0.02
470 80 10 2.75
100 190 10 14.39
100 80 65 2.54
470 190 65 4.35
100 300 54 13.00
100 300 120 8.50
100 80 120 0.05
285 300 10 11.32
285 190 120 3.13




cetoolbox www user
2004-12-17