Parameter estimation of known function based on experimental data

78 Views Asked by At

Assume following function:

$x = 1 - \exp( -t_s \cdot A \cdot \exp( ( \frac{-E}{R\cdot T} ) ) $

Parameters $t_s$ and $R$ are constant and known.

$T$ is variable.

How can I estimate $A$ and $E$ parameters based on experimental data (about 300 points of pairs $(T_i, x_i)$ )?

I was thinking about some kind of least square estimation, but I'm a little bit stuck because of that double exponential function.

The main objective is to minimize sum of squared error between experimental data and it's approximation with known function.

1

There are 1 best solutions below

4
On BEST ANSWER

For sure, it can be done provided that you have good estimates of $A$ and $E$ to start the nonlinear regression.

In a first step, take logarithms and rewrite the model as $$-\frac{\log(1-x)}{t_s}=A \exp\left(-\frac E{RT} \right)$$ Take logarithms again $$y=\log\left(-\frac{\log(1-x)}{t_s}\right)=\log(A)-\frac E{RT}=\alpha+\frac \beta T$$ A linear regression will provide $\alpha$ and $\beta$ from which $A=e^\alpha$ and $E=-R \beta$.

With these estimates (quite simple to get), you can start the nonlinear regression.

Just for illustration purposes, consider the following set of extremely noisy data $$\left( \begin{array}{cc} T & x \\ 450 & 0.1 \\ 475 & 0.2 \\ 500 & 0.3 \\ 525 & 0.5 \\ 550 & 0.7 \\ 575 & 0.8 \\ 600 & 0.9 \end{array} \right)$$ for which $t_s=\pi$ (why not ?) and $R=8.31$. So, the linear regression will use as values $$\left( \begin{array}{cc} T & y \\ 450 & -3.39510 \\ 475 & -2.64467 \\ 500 & -2.17566 \\ 525 & -1.51124 \\ 550 & -0.95910 \\ 575 & -0.66885 \\ 600 & -0.31070 \end{array} \right)$$ yielding to $$y=9.10976 -\frac{5605.85}{T}$$ So, the estimates are $A=9043$ and $E=46600$. Starting with these, the nonlinear regression leads, using two iterations only, to $A=9123$ and $E=46509$ (quite close) with $R^2=0.9988$. Now, compare the model prediction to the data $$\left( \begin{array}{ccc} T & x_{\text{exp}} & x_{\text{calc}} \\ 450 & 0.1 & 0.1075 \\ 475 & 0.2 & 0.1966 \\ 500 & 0.3 & 0.3260 \\ 525 & 0.5 & 0.4894 \\ 550 & 0.7 & 0.6643 \\ 575 & 0.8 & 0.8171 \\ 600 & 0.9 & 0.9218 \end{array} \right)$$