Optimal Inventory Management for New Tools Using Model-Based Deep Reinforcement Learning: A Solution for Short Product Life Cycles

Abstract

This research examines the optimum inventory management issue for new instruments, serving as a pertinent illustration of a supply chain characterised by a short product life cycle. Establishing the correct inventory level minimises wasted opportunities and faulty stock, which is crucial for enhancing profitability. Mathematical optimisation and reinforcement learning methodologies have been suggested for inventory management; nonetheless, the majority of these methodologies concentrate on things that are consistently offered over an extended duration. Consequently, when the objective is a new product, optimising inventory management from the date of its launch is challenging owing to an insufficiency of data for analysis. We address this issue by concentrating on model-based deep reinforcement learning characterised by high sample efficiency and present an inventory management strategy for new items that integrates model learning in an offline setting with planning in an online context. Simulations using authentic historical sales datasets indicate that the suggested strategy surpasses current methodologies for profitability, efficiency, and customer happiness. The suggested strategy enhances overall incentives and inventory turnover by under 5% compared to the trust area policy optimisation method, while preserving the same stock-out rate. Furthermore, the findings indicate that the suggested strategy can sustain consistent inventory management for multiproduct and multistore supply chains.