Applying frequency based forecasting for resource allocation

Abstract

Soaring complexity in supply chains with more fluctuations and ever increasing uncertainty in demand puts an increased focus on flexibility and changeability in manufacturing. Thus, it is increasingly important to determine the right change type, such as changes in the number of employees or overtime, at the right time in order to be able to react appropriately and sustainably to changes in demand. The developed approach uses frequency analysis to predict future changes in demand in different frequency ranges in order to assign appropriate change types to them and optimize the change intensity for each change type and time step. The foundation of the related algorithm is a discrete Fourier analysis that extracts relevant frequencies and assigns change types using generative algorithms to enable cost-minimizing production. The algorithm is validated against LSTM and ARIMA forecasting in a use case with seasonal time series including different noise levels.