Optimising inventory control and demand forecast accuracy though multi-objective optimisation

Overview

In today’s competitive business environment, effective inventory management and accurate demand forecasting are critical for minimising costs and maximising profitability. This project aims to address these two challenges simultaneously by applying a multi-objective optimisation approach. The primary objectives are to improve demand forecast accuracy while optimising inventory control decisions, balancing trade-offs between conflicting business goals such as minimising stockouts, reducing excess inventory, and maintaining customer service levels.

Traditional approaches to inventory management and demand forecasting often treat these processes separately, which can result in suboptimal outcomes. A highly accurate demand forecast, for example, does not always guarantee efficient inventory control. Furthermore, optimising inventory control in isolation may lead to increased costs due to either excess inventory or stockouts. To address these limitations, this project integrates the two aspects by using multi-objective optimisation to consider competing objectives simultaneously, leading to a more holistic and robust solution.

Research activities

You will work with and apply a combination of predictive modelling and optimisation techniques:

• Predictive models for demand forecasting
  • A combination of statistical and machine learning models (e.g. ES, XGBoost, and LSTM) will be employed to generate highly accurate demand forecasts. These models leverage historical sales data and external factors such as market trends and seasonality.
• Inventory control models
  • Mathematical models such as Order Up To Level (OUT) and stochastic inventory models (Newsvendor) will be employed to determine optimal inventory levels based on forecasted demand and lead times.
• Multi-objective optimisation
  • A combination of evolutionary algorithms (e.g. NSGA-II) or Pareto optimization methods is used to balance the trade-offs between forecast accuracy and inventory management performance. The optimization seeks to minimize multiple conflicting objectives like inventory holding costs and forecast error simultaneously.

Outcomes

The project focuses on solving the following key problems:

• Demand forecast accuracy
  • Improving the precision of demand forecasts using advanced machine learning models and time-series analysis. Accurate demand forecasting is essential for aligning inventory levels with customer demand, minimizing costs, and avoiding stockouts or overstocks.
• Inventory control optimisation
  • Implementing optimisation models to make better inventory decisions by controlling the levels of safety stock, reordering points, and stock replenishment quantities. The objective is to reduce holding costs, stockouts, and late deliveries while maintaining service levels.
• Multi-objective optimisation
  • The core methodology involves formulating the problem as a multi-objective optimization model, where the trade-offs between demand forecast accuracy (measured in terms such as Mean Absolute Scaled Error or MASE) and inventory performance metrics (e.g. stockout rate, holding cost) are optimised simultaneously. This enables businesses to strike an optimal balance between competing priorities.

Skills and experience

• Fluent in R or Python.
• Familiar with optimisation models.

Keywords

• forecasting
• machine learning
• optimisation
• operations research

Contact

Contact the supervisor for more information.