Combine times series with different length by interpolation

Hey Community,

I have the following problem: in order to use the machine learning toolbox, I want to create a single table which contains the measurements of the hydraulic system pressure from 7 flights. Since those flights vary in their length, the number of rows of the tables from each flight varies as well.

In order to compare the flights, I want to "manipulate" the data and bring all flights to the same length without loosing the trend and rashes in the data. I think I need to interpolate my flights by their time (so x-axis) but I only found the option to interpolate the values itsself and not the timestamp.

Is there a possibility and if so, could some hero provide me with help/information/examples?

Thanks and best, Richard

Answer 1

[Guillaume]

I would convert your tables to timetables and synchronize those using whichever time basis or time base is appropriate.

[Richard Vergin]

Hey,

I synchronized them and now have a row of flights. But unfortunately the toolboxes for machine learning are not compatible with time series. Is there a possibility to transfer the time series to an array afterwards?

Best Richard

Answer 2

[Sara Nadeau]

If you have Simulink, the Simulation Data Inspector has a robust comparison algorithm that can handle this for you. This topic explains the algorithm: How the Simulation Data Inspector Compares Data. If your data isn't coming from a simulation, you can import it into the Simulation Data Inspector: View Data from the Workspace or a File.

I hope this helps! The Simulation Data Inspector also has a programmatic interface you can use to harness the comparison algorithm. Inspect and Compare Data Programmatically

[Richard Vergin]

Hey, I'm trying to use the Simulation Data Inspector but it is impossible to import my time series. Everytime I try to import one or multiple time series, either from workspace from a file, the data does not appear in the data inspector alltough a popup with "complete" appears.

Do you happen to have an idea why this could be the case?

Best Richard

[Sara Nadeau]

Hey Richard,

What is the format of your data and what version of MATLAB/Simulink are you working with?

One requirement for using the Simulation Data Inspector is that your data has to be in a format that pairs time and data values, like the timeseries or timetable format.

[Richard Vergin]

Hey Sara,

I'm working with R2018a. I'm trying to implement a time series but maybe the format of my time series is wrong, since it says its a "1x1 timetable timeseries". How can i create a time series out of a timetable, table or array then? I tried it with: newTS = timeseries(NameOfTimeTable)...

thanks a lot and best, Richard

[Sara Nadeau]

The Simulation Data Inspector needs the time vector to be in seconds. Is your timetable using datetime? If so, this answers post may help you do a quick conversion to seconds.

I did a quick test and it looks like you can import a timetable into the Simulation Data Inspector if the time is in units of seconds.

[Sara Nadeau]

Oh! Forgot to include the link to the answers post: https://www.mathworks.com/matlabcentral/answers/131216-convert-date-and-time-to-seconds

Guillaume edit: made link live

[Richard Vergin]

Not sure if I did it right but unfortunately its not working. I created a timetable with the time column in format seconds and the time table is not recognized from the Simulink Data Inspector. Alltough it is in my active workspace, there is no file I could select to import. Seems like I can not import timetables to this tool, doesn't matter if the time column is in 'ss' or in 'yyyy:MM:dd-HH:mm:ss'. The tool does only recognize time series but I don't know how to form a proper time series out of a timetable, table or array...

Is there any other idea?

Thanks and best,

Richard

[Peter Perkins]

Richard, you said two things that make me thing you are creating the wrong kind of thing:

1) "it says its a '1x1 timetable timeseries'" - that's can't be what it says, or if it does, it can't be what you intended. Perhaps you could show a short example of what you are creating and how you are creating it.

2) I think Sara was saying that the row times vector has to be a duration, but "doesn't matter if the time column is in 'ss' or in 'yyyy:MM:dd-HH:mm:ss'" makes me think you are changing the display format of a timetable that has datetime row times. Perhaps if you converted those to duration using something like

tt.Time = tt.Time - tt.Time(1)

Sara linked to an old post that did the analogous thing using datenums; I recommend that you stick with datetime/duration if possible.