[R] Modeling unusual cyclical trends using dlm

[rsteckel]

I'm trying to use the dlm package to model some solar radiation data. There's
both a cyclical and a seasonal pattern to the data. The seasonal portion
follows a regular pattern and dlm can easily handle this. However, the
cyclical pattern is more difficult. Since this is solar radiation data, some
period of the day is 0 (when the sun has set). I am having a difficult time
specifying this pattern in the state space model. (A qqnorm plot of the
residuals looks awful). 

The measurements are taken every 10 minutes. So, the pattern repeats every
144 (6 * 24) minutes. I'm thinking I should use dlmModTrig, but not sure how
since it does not follow a typical sine-like pattern. 

Does anyone have any suggestions?

Here's an example of a few days worth of data:

http://r.789695.n4.nabble.com/file/n3754110/CyclicalTrend.jpg 



--
View this message in context: http://r.789695.n4.nabble.com/Modeling-unusual-cyclical-trends-using-dlm-tp3754110p3754110.html
Sent from the R help mailing list archive at Nabble.com.