How do I compute 3 minute moving average in timeseries?

I have a time series object with two columns : Date,time (dd-mm-yyyy HH:MM:SS format) and Value. The data is sampled every 2 seconds. The total data is available is for around 10 days. How do I compute a timeseries with 3-minute moving average values?

Answer 1

[Image Analyst]

3 minutes at every 2 seconds would be about 91 elements, so use movmean():

movingAverageSignal = movmean(signal, 91);

[Chris Turnes]

If the times are stored as datetimes, you don't even need to figure out that it's about 91 elements; you can do

y = movmean(x, minutes(3), 'SamplePoints', timestamps)

Answer 2

[Peter Perkins]

If you mean, "means for each 3-minute time window", consider using a timetable and the retime function. If you mean, "means at each 2-sec step over the previous 3 minutes", you can also do that on a timetable, using smoothdata.

[nlm]

How to calculate a 10 day average when the timeseries has duplicate dates. I do not want to average the dupliate dates priorly. I want to include them in the 10 day average. When I use movmean, I get this error, any suggestions ?

Error using builtin
'SamplePoints' value contains duplicates.

Error in datetime/movmean (line 105)
    y = builtin('_movmeandt', args{:});

[Peter Perkins]

Use retime.

Answer 3

[Andrei Bobrov]

Let ts - your timeseries object.

k = 3*60/2; % the number of elements corresponding to three minutes in your case
ts.Data(:,2) = movmean(ts.Data,[0, k - 1]);

Answer 4

[sudharshini]

preprocds = transform(letterds, @scale)