Increasingly precise forecasting has become a decisive factor in the success-oriented planning of companies. The growing capacity of recent IT-systems enables the preparation of forecasts based on algorithms of growing complexity and effort. In doing so, artificial neural networks range among the most promising procedures. The present thesis applies feedforward networks to the creation of forecasts with respect to a real process within RHI AG. The training process is governed by a backpropagation algorithm. The use of artificial neural networks is a natural consequence of the availability of RHI AG's internal data material that contains both, time series as well as cross-sectional data. We also present an analysis of the network's performance as well as of various parameter settings and procedures which we involved in order to obtain accurate forecasts. Finally, the results of the present thesis are compared to those delivered by RHI AG's internal algortihmic forecasts.
|Translated title of the contribution||Production Volume Forecast using Neural Networks|
|Award date||14 Dec 2012|
|Publication status||Published - 2012|
Bibliographical noteembargoed until null
- neural networks
- production volume