Numberical Method to ID changing slope in poor data

[elfman]

I have a set of data and overall it tells me what I want to know. However, there are areas in the curve that the data jumps up and down. My algorithm looks at the slope from point to point, and gets confused when the general trend is negative, but between two or three points it goes positive. In my data, I expect the slope to change several times from start to end, but numerically I can't figure out how to identify the trends vs. the localized slope changes. Any one have some ideas one numerically analyzing the data?

Thanks in advance.

 

[GregLocock]

Do a polynomial fit and throw out the higher terms.

Analyse the moving average instead of the individual data points.

FFT it, throw the high frequency stuff away, then IFFT it (may be more complex than that)

There are many solutions.


Cheers

Greg Locock

 

[sreid]

I would start with a three or five point moving average.

 

[unclesyd]

Here is a crutch.

http://www.systat.com/products/TableCurve2D//

 

[m777182]

There are proceedures to filter row data that do not lose peaks in derivatives.Search for proceedure "LOESS" or "Savitzky-Golay" or commercialy available "Peak finder".These routines are heavilly used in modern instrumentation algorithms like thermoanalysis,spectrometry etc and are not difficult to program.
m777182