RECURRENCE PLOTS

Hello,

Following the python book of Martin Trauth, I'm trying to calculate the recurrence plots for time series measuring the displacement of the ground.
These times series are taken on broad areas in France and UK and I would like to test whether RP and CNN models can help me cluster/classify them to subsequently analyse them by groups.

Following the book and the section written by Norbert Marwan, I started with the assumption that the phase space is only one-dimensional (the displacement) from which I can calculate the symmetric distance matrix S and apply a threshold to create the RP.

Is this a first correct approach? meaning I don't need to use an embedding dimension m nor a time-delay tau ? or I am missing something ?

Any help would be appreciable,

Best Regards,
nicolas

It appears we are working on similar problems. The phase space is probably not one dimensional. I only say this because I have been working on reconstructing attractors from creepmeter data. The challenging part for me is detrending the entire signal from the creepmeter, rather than just a segment. It seems to influence FNN results (AMI not so much). I am finding that even after detrending the signal in any case, the first VT mode of the SVD breakdown of the Hankel matrix or embedded dimensions seems to "pull out" the information about the long range trend, while the second, third, and fourth SVD modes form a strange attractor.

Sample rate is 1/10 min