Extrapolation nonlinear data in matlab

Question

How to extrapolation following data for 850 and above in matlab?

x =  200.0000  205.0000  210.0000  215.0000  220.0000  225.0000  230.0000  235.0000  240.0000 245.0000  250.0000  255.0000  260.0000  265.0000  270.0000  275.0000  280.0000  285.0000 290.0000  295.0000  300.0000  305.0000  310.0000  315.0000  320.0000  330.0000  340.0000 350.0000  360.0000  370.0000  380.0000  390.0000  400.0000  410.0000  420.0000  430.0000 440.0000  450.0000  460.0000  470.0000  480.0000  490.0000  500.0000  510.0000  520.0000 530.0000  540.0000  550.0000  560.0000  570.0000  580.0000  590.0000  600.0000  620.0000 640.0000  660.0000  680.0000  700.0000  750.0000  800.0000

y =  0.8900    0.8600    0.8400    0.8200    0.8000    0.7900    0.7700    0.7500    0.7400   0.7200    0.7100    0.6900    0.6800    0.6700    0.6500    0.6400    0.6300    0.6200   0.6100    0.6000    0.5900    0.5800    0.5700    0.5600    0.5500    0.5400    0.5200   0.5100    0.4900    0.4800    0.4700    0.4500    0.4400    0.4300    0.4200    0.4100   0.4000    0.3900    0.3900    0.3800    0.3700    0.3600    0.3600    0.3500    0.3400   0.3400    0.3300    0.3200    0.3200    0.3100    0.3100    0.3000    0.3000    0.2900   0.2800    0.2700    0.2600    0.2600    0.2400    0.2200

Answer 1

If you plot log(y) vs log(x) you will see they follow a linear relationship. So we can do:

x =  [200.0000  205.0000  210.0000  215.0000  220.0000  225.0000  230.0000  235.0000  240.0000 245.0000  250.0000  255.0000  260.0000  265.0000  270.0000  275.0000  280.0000  285.0000 290.0000  295.0000  300.0000  305.0000  310.0000  315.0000  320.0000  330.0000  340.0000 350.0000  360.0000  370.0000  380.0000  390.0000  400.0000  410.0000  420.0000  430.0000 440.0000  450.0000  460.0000  470.0000  480.0000  490.0000  500.0000  510.0000  520.0000 530.0000  540.0000  550.0000  560.0000  570.0000  580.0000  590.0000  600.0000  620.0000 640.0000  660.0000  680.0000  700.0000  750.0000  800.0000];
y =  [0.8900    0.8600    0.8400    0.8200    0.8000    0.7900    0.7700    0.7500    0.7400   0.7200    0.7100    0.6900    0.6800    0.6700    0.6500    0.6400    0.6300    0.6200   0.6100    0.6000    0.5900    0.5800    0.5700    0.5600    0.5500    0.5400    0.5200   0.5100    0.4900    0.4800    0.4700    0.4500    0.4400    0.4300    0.4200    0.4100   0.4000    0.3900    0.3900    0.3800    0.3700    0.3600    0.3600    0.3500    0.3400   0.3400    0.3300    0.3200    0.3200    0.3100    0.3100    0.3000    0.3000    0.2900   0.2800    0.2700    0.2600    0.2600    0.2400    0.2200];

coeff = polyfit(-log10(x) , log10(y), 1)   % the '1' means linear
xp = [200:1000];
yp = 10^coeff(2)*xp.^(-coeff(1));
plot(x,y,'o',xp,yp)

And you get:

Answer 2

I am trying to use the above function for my own curve which i wannt to extend further , but facing some issue. I am new to this slm toolbox. please suggest how to avoid this error the time 1 is x-axis variable array having values range: 9.8682e-05 9.8687e-05

I would comment about the dangers of extrapolation. Nobody has ever said it better than Mark Twain, in Life on the Mississippi, here. Extrapolate too far, and expect random garbage as a prediction.

Having said that, there are many tools one could use. You might do an exponential fit of some ilk. The problem here is you need to find a model with the proper shape. This often takes some experimentation to find something that fits your data, if you don't already have a physical model that makes sense from first principles.

Better is a spline model, but the problem is an interpolating spline extrapolates poorly. So my recommendation is to use a tool like SLM, then building in your knowledge of what can/should happen in the region of extrapolation.

Here, I made some guesses as to reasonable properties of the curve as it is extrapolated out quite a ways. I've imposed a monotone decreasing constraint over the entire curve, with positive second derivatives. As well, the right end point was not allowed to lie below zero.

mdl = slmengine(x,y,'knots',[200:100:800,1000:500:2000],... 
   'decreasing','on','concaveup','on','rightminvalue',0,'plot','on');