This repository containes the notebooks and datasets used during our talk. You can launch a readily configured environment by clicking the batch below:
Time series are an omnipresent data type. Data, generated by business processes or connected devices, are essentially time series. Sales, inventory in warehouses or the condition of a car battery measured over time, are other examples. Quickly the number of series growth such that a human can not observe them in an easy and economical way. There comes algorithmical methods and machine learning to the rescue to detect anomal behavoir and to forecast. We present and introduce modern ML algortithms, which get integrated into a data architecture. To make the algorithms practically usable for the audience we apply the algorithms on public dataset and provide ready to use notebooks.
During the talk we used two main demo-datasets.
In this dataset, you will find a refactorised version of the data from the London data store, that contains the energy consumption readings for a sample of 5,567 London Households that took part in the UK Power Networks led Low Carbon London project between November 2011 and February 2014. The data from the smart meters seems associated only to the electrical consumption.
A detailed data description can be found on Kaggle: Kaggle: Smart Meters in London.
We have a fleet of 100 aircraft engines of the same model. Each engine starts with different unknown degrees of initial wear and manufacturing variation which is considered normal. Each engine is operating at normal conditions at the beginning of the time series but degrades over time until a predefined, unknown failure threshold is reached. The objective is to predict from any point in time how long we have until we need to perform maintenance.
References: A. Saxena, K. Goebel, D. Simon, and N. Eklund, “Damage Propagation Modeling for Aircraft Engine Run-to-Failure Simulation”, in the Proceedings of the 1st International Conference on Prognostics and Health Management (PHM08), Denver CO, Oct 2008.
If you are interested to learn more about the methods, of if you have your own time series use-case, feel free to contact Zoi.
When you want to be one of the early adopters or give feedback we would be happy, if you contact us via [email protected]