TFS: The Time Series Forecaster of ESFA
The containerised application contains TFS, a Bayesian Optimization of the state-of-the-art AI, N-Beats. The TFS is one of the components of ESFA's AI. TFS Docker Image has been adapted to work with any kind of time series, from weather to Bitcoin prices or industrial machine's performance.
Categories
What is the challenge that is being addressed?
Forecasting time series is a complex task, classical machine learning (Artificial Neural Networks, Long Short Term Memory, etc) might not be enough for most users, who usually expect higher accuracy from AI algorithms. TSF is the next step in terms of forecasting time series, addressing both, the skill problem and the AI optimization problem with an automated manner
What is the AI solution the project has implemented?
ESFA is an AI developed by Geoskop under AI4Copernicus framework to solve the problem of the low accuracy of Seasonal Forecasts Systems (SFS). ESFA consists of a complex Artificial Intelligence effectively combining different state-of-the-art AIs to achieve the most skilled time-series forecast with minimal human effort. One of the components of ESFA is the TFS, a N-Beats AI optimizer (working on top N-Beats theoretical optimization). TFS might be used by ESFA if ESFA decides to do so.
This AI Asset has been adapted to work with any kind of time series, be it univariable or multivariable
Who helped implement the AI solution?
This solution is implemented in the context of ESFA, a winning project from the AI4Copernicus 3rd Open Call, by the Geoskop company. The results of the project are available here.
