Optimizing Production Planning with Neural Networks and Genetic Algorithms: A Case Study in Forecasting and Buffer Management

Abstract

This study proposes a two-stage hybrid methodology that integrates neural networks (NN) for demand forecasting with genetic algorithms (GA) for buffers-related cost optimization in demand-driven material requirements planning (DDMRP) systems. The approach addresses a key gap in production planning research by systematically combining data-driven demand prediction with combinatorial optimization to reduce production costs using decoupled lead time (DLT), a core DDMRP metric. Forecast-derived demand intensities are incorporated into DLT weighting, enabling prioritized buffer placement at high-demand, long-lead-time workstations. Applied to an automotive sun-visor production line with eight sequential workstations, the method yielded an optimal buffer configuration that demonstrates the effectiveness of coupling NN pattern-recognition capabilities with GA optimization performance. The resulting framework provides production managers with a flexible and quantifiable decision-support tool for buffer positioning in dynamic manufacturing environments. Future research should extend the method to multi-product systems, explore advanced long short-term memory (LSTM) architectures to enhance forecasting accuracy, and conduct cross-industry evaluations to assess generalizability.