I have measurements from a system over a long period of time. During this time, the system has been in different states (for example, different controller tuning configurations). Now, looking at the data alone (the time series), how can I identify the portion that corresponds to a certain (normal) state of the system (for example, optimal controller tuning)? OR: How to identify normal data from abnormal data in a time series?
You can research with google a bit with the subject Outlier detection of time-series data!!
Plot the whole sequence, check for the amplitude, depending upon the similarity in the amplitudes of the signal components, separate the data as frames. for every frame estimate the SNR and power. If the power and SNR remains same, then frames were obtained at the same time.
Hi, I guess hidden Markov chains can be applied here.
Do you have samples for "normal state", or you have to do it all in an unsupervised manner?
Hi,
If you want to find where the time series change (differes from the normal states), you may use "Structural Break Detection in Time Series" to find it.
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series