Skip to main content

I’m currently writing my Master thesis in collaboration with Aize (represented by V. Flovik), where I’m trying to use Generative adversarial Networks (GAN) to generate synthetic time series data. To do this, I’m using data from Cognite OID. I’ve explored the data in the previous semester, and I figured that one of my main challenges is to find longer periods of data that are “Normal”. Meaning periods without to large irregularities and without too many missing data points. Because even when I aggregate the samples over 30 seconds, there are still samples missing in some periods. Some of these periods are hours long and can’t really be filled in without messing up the temporal development of the data. I can’t find any information about the cause of these periods either.

The time series I’ve been looking at are:

I’m no expert in process engineering, and I only chose these 9 sensors because another master thesis from 2019 had used these sensors in their project. I therefore reach out here to ask for some guidance on which time series are well suited for the purposes mentioned above. Ideally, I would need at least 3 different time series, as I try to generate data for a whole system of sensors at once. Any tips about previous research similar to what I’m doing, is also appreciated.

Hi @Fredrik Sveen,

Develop a script with the capability to analyze and determine appropriate time series data. Execute this script across all time series incorporated in the project to ensure comprehensive coverage. You can find our documentation here.   


Reply