Counting cycles in time series (including period lengths)

[Cord Kaldemeyer]

Hi everybody,

is anyone familiar with cycle-counting algorithms that are applied to
storage systems (e.g. state of charge time series of stationary
electrical storages)?

I know of methods like rainflow-counting [1] that stems from fatigue
analysis and only deals with amplitudes and full/half cycles. Since I am
also interested in the period lengths (and not only the amplitudes) I
have thought about applying DFT methods [2] which do not seem to be the
right instrument since the time series are not really periodical.

Has anyone dealt with this topic before and some advice or keywords?
Aren't there methods to include both, the amplitudes and the period lengths?

Cheers
Cord

[1] https://en.wikipedia.org/wiki/Rainflow-counting_algorithm
[2] https://en.wikipedia.org/wiki/Discrete_Fourier_transform

--

Cord Kaldemeyer, M.Eng.

Wissenschaftlicher Mitarbeiter
Zentrum für nachhaltige Energiesysteme (ZNES)
Fachbereich Energie- und Biotechnologie
Hochschule Flensburg
Gebäude H, Raum H322
Kanzleistraße 91-93
D- 24943 Flensburg

T: +49 (0)461 805 1260
F: +49 (0)461 805 1300

www.znes-flensburg.de
www.hs-flensburg.de
cord.ka...@hs-flensburg.de

[Cord Kaldemeyer]

Hi everybody,

here's the summary of my answers:

1.) Discrete Fourier Transform [1] can and has been used to analyse the SOCs periodicy (shares) and get a rough estimation of the amplitudes. Nevertheless, cycles cannot be counted and one has to be careful with the interpretation of the amplitudes.

2.) Scaling the SOC timeseries to its overall mean (setting it zero), getting the zero crossing indices and calculating the cycle max/min and length values delivers both, the cycle amplitude, length and the cycle count. Nevertheless, this approach only detects longer/higher cycles correctly but omits the small "intermediate" cycles.

3.) The rainflow counting algorithm [2] can be used to count cycles and is a good standard since it is standardised according to ASTM. It (logically) counts more cycles than the zero-crossing-method. Nevertheless, it does not deliver the cycle length.

4.) A method developed by Jonny Dambrowski, Simon Pichlmaier and Andreas Jossen [3] is superior to the zero-crossing-method as it detects cycles exactly like rainflow counting algorithm and additionally delivers the cycle lengths. This is decribed in paper and has been tested by myself successfully. Nevertheless, it only does so with MATLABs peak detection algorithm and not with GNU OCTAVEs.

5.) Autocorrelation methods can be used to analyse a time series' periodicy but do not deliver information about the amplitude and single cycle lengths. In this case they seem to be rather useless compared to the other approaches.

So the selection of a suitable approach -as always- depends on the specific (research) question ;-)

Thanks for all answers and help!

Cheers
Cord

[1] https://en.wikipedia.org/wiki/Discrete_Fourier_transform
[2] https://en.wikipedia.org/wiki/Rainflow-counting_algorithm
[3] ASTM E 1049-85. (Reapproved 2005). "Standard practices for cycle counting in fatigue analysis". ASTM International.
[4] Jonny Dambrowski, Simon Pichlmaier, Andreas Jossen: Mathematical methods for classification of state-of-charge time series for cycle lifetime prediction, Advanced Automotive Battery Conference Europe, June 2012

-------- Weitergeleitete Nachricht --------

Betreff:	Counting cycles in time series (including period lengths)
Datum:	Wed, 1 Nov 2017 11:37:42 +0100
Von:	Cord Kaldemeyer <cord.ka...@hs-flensburg.de>
An:	openmod-i...@googlegroups.com