Youve got a measured value for something which can be approximated by some polynomial,youve also got a predicted value and therefore an error which can also be approximated by a polynomial,so the aim is the minimise the error,a bit of calculus or even excell will do the rest.