Before asking a question, I have no background knowledge of engineering. So I have no idea about these issues like control or signal engineering. (Also, I'm not good at using English. sorry for that.)

I attached the picture related to my question below. It is time series data and I want to find the black circled points.

1) these data follows Gaussian distribution.

2) The most important thing in this question is that I can't know future data at specific time point. I know only past data at specific time point. I want to recognize black circled points around local max/min points or right after local max/min points.

3) There is no model with these data. e.g. There is no equations like relations of velocity, acceleration and positions. Only these data are the only measurements.

I've got some recommendations to use Kalman filter things, and It helps to make these lines differentiable. Now I'm studying Kalman filter but It seems I do need kinds of model.

Is it possible or useful to use Kalman filter? or, do I have to another methods to filter these data?


  • $\begingroup$ Have you evaluated low pass filters? $\endgroup$
    – user190081
    Commented Sep 14, 2018 at 14:13

1 Answer 1


The Kalman-filter is an observer, predicting the next system state based on an initial value. That prediction needs to be made based on some kind of model.

The closest solution that I can come up with is a learning-based adaptive control approach, to identify the model and later use the identified model to design an observer. This requires a rudimentary initial model. I never tried something like that for some arbitrary measured data.

If you can specify a bit where the data comes from, for example if you can influence the measured system in some way (does it have actuators that a controller could use?), I might be able to give a clearer answer.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.