Reinforcement Learning Approach to Determine Effective Inventory PolicyTools Nurkasanah, Ika (2019) Reinforcement Learning Approach to Determine Effective Inventory Policy. [Dissertation (University of Nottingham only)] This is the latest version of this item.
AbstractSupply chain management (SCM) is believed to be a key factor in delivering competitive advantages for business. Aiming to regulate the flow of information, cash and products throughout all of a chain’s business processes to optimise total profitability, SCM is highly influenced by operational decisions made by stakeholders, especially in terms of inventory policy. Therefore, this topic has attracted a great deal of interest among researchers in recent decades as evidence suggests nearly 50% of supply chain costs are triggered by inventory-related costs. The previous mathematical approaches, such as Economic Order Quantity (EOQ), Periodic Order Quantity (POQ), continuous reorder quantity (s, Q), etc., remain popular due to their ease of use. Even so, considering some uncertain factors such as demand and lead time, these approaches can become inaccurate when defining an optimum inventory policy. To tackle such stochastic situation, machine learning is proposed because it offers superior abilities to explore hidden knowledge and more complex patterns within inventory-related information. Reinforcement learning (RL), an emerging machine learning algorithm, is used in this project due to its ability to not only considers an immediate payoff but also future value implications is used in this study. The effectiveness of such a method in minimising inventory costs is compared to previous traditional or mathematical approach. In addition, this project adds some constraints and adjustments to reflect more complex and real supply chain conditions that have not been considered before. The most striking finding is that through sufficient trial-error simulations, RL can perform better in minimising inventory costs by taking action more effectively (i.e. placing orders in appropriate quantities and at the right time). Results show that RL approach will be advantageous when applied by decision-makers with similar supply chain conditions as stated in this project. Since this study is limited only to single-item analysis in the manufacturing and supplier ordering process, it is suggested to combine RL with other machine learning algorithms for analysing items through end-to-end supply chain process for future work.
Available Versions of this Item
Actions (Archive Staff Only)
|
Tools
Tools
![[img]](/style/images/fileicons/application_pdf.png)
