I'm working on a system written in Java which is able to perform forecasts using history data. The algorithm being used is a Java port of <a href="http://adorio-research.org/wordpress/?p=1230" rel="noreferrer">this Holt-Winters implementation</a> (multiplicative seasonality).
I have several time series that we would like to analyse and we need different smoothing coefficients for those time series. The algorithm seems to work really well at the moment, the only problem is that how to determine the most sensible values for the smoothing coefficients (alpha, beta, gamma).
I know that I need some sort of non-linear optimization but I am not a mathematician at all so I'm a bit lost among all those theories and concepts.
<strong>EDIT</strong>:
I have a lot of different time series to analyse, I'd like to know if there is a standard/good enough technique (library would be better) for calculating the smoothing parameters that I should give to the Holt-Winters algorithm.
I have several time series that we would like to analyse and we need different smoothing coefficients for those time series. The algorithm seems to work really well at the moment, the only problem is that how to determine the most sensible values for the smoothing coefficients (alpha, beta, gamma).
I know that I need some sort of non-linear optimization but I am not a mathematician at all so I'm a bit lost among all those theories and concepts.
<strong>EDIT</strong>:
I have a lot of different time series to analyse, I'd like to know if there is a standard/good enough technique (library would be better) for calculating the smoothing parameters that I should give to the Holt-Winters algorithm.