2423. Model Building and Forecasting with Recurrent Neural Networks
Invited abstract in session WE-6: Predictive Analytics: Forecasting II, stream Analytics, Data Science, and Forecasting.
Wednesday, 16:30-18:00Room: H9
Authors (first author is the speaker)
| 1. | Hans Georg Zimmermann
|
| Corporate Technology CT RTC BAM, Siemens AG | |
| 2. | Nico Beck
|
| Fraunhofer Institut für Integrierte Schaltungen IIS | |
| 3. | Julian Stengl
|
| Fraunhofer IIS |
Abstract
Time series modeling is often seen as a process in which one starts with a data set, identifying features of the data (e.g. periodicity) and go on to the model building. In this talk we want to follow another guideline:
First, think about a reasonable mathematical frame for your modelling. Here we will insist not only on universal approximation but will work out the specifics of Historical Consistent Neural Networks. Second, ensure that the chosen model class can be adapted in a best possible way to observations (Typically, in this frame we must work with overparameterized models). Third, think about the relevant observations for your modelling. Without step one and two it is irrelevant to discuss the importance of data (a crazy counterexample would be: if you believe in linear models than the best model of a sinus wave would be the x axis). Thus, the mathematical frame and its handling has to be done before you focus on the data.
Finally one has to discuss the uncertainty of the forecast including concepts for this uncertainty and their representation.
Keywords
- Forecasting algorithms
- Predictive Analytics
- Time Series Analysis
Status: accepted
Back to the list of papers