Abstract

This paper presents a demand forecasting approach that automatically selects optimal article specific forecasting methods and optimizes the method parameters, using deterministic simulation and a Genetic Algorithm (GA). For an efficient demand forecast, choosing the best forecasting method based on the item-specific historical requirements time series is key. The optimization of the forecast parameters is also crucial for efficient demand planning. Both decisions lack digital method support, leading to suboptimal forecasts in practice and thus inefficient material requirements planning. This paper investigates the optimization potential of an automatically optimizing forecasting approach, featuring a simulation-based comparison of six standard forecasting methods, evaluated using a case-study from the capital goods industry. The methodological core of the optimization is a GA, which improves the underlying, method-specific forecast parameters. The simulation-based optimization provides a rolling-horizon demand forecast for each item, and is determined through the application of a rule-based heuristic. The results show a significantly improvement potential through this form of efficient item-specific demand planning.