Loading...

The Impact of Integrated Prediction and Optimization on the the Bullwhip Effect in supply chains

Salavatizadeh, Fatemeh Zahra | 2025

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 58440 (01)
  4. University: Sharif University of Technology
  5. Department: Industrial Engineering
  6. Advisor(s): Sedghi, Nafiseh
  7. Abstract:
  8. The Bullwhip effect is a phenomenon that refers to the amplification of demand variance as one moves upstream in a supply chain. This effect imposes additional costs on the supply chain and complicates inventory planning. Several factors contribute to the emergence of this phenomenon, one of which is demand forecasting. Since the demand faced by each level of the supply chain is uncertain, every level must first forecast demand in order to plan its inventory and meet customer requirements. However, this very process of forecasting contributes to the bullwhip effect. In this study, we investigate the impact of machine learning and deep learning forecasting methods, along with their hyperparameters, as well as traditional time series models—including LightGBM, Random Forest, Moving Average, and LSTM—on the intensity of the bullwhip effect. In addition to the bullwhip effect, the inventory cost is also taken into account. Instead of applying prediction and optimization separately—where only prediction accuracy is minimized in the prediction stage—this research adopts an integrated prediction-optimization framework. The goal is not to maximize demand forecasting accuracy, but rather to make forecasts in such a way that the bullwhip effect is minimized. The results show that models with strong time-series learning capabilities, such as LSTM and Moving Average, despite having lower prediction accuracy, generally lead to better control of the bullwhip effect. Furthermore, instead of selecting model hyperparameters based on prediction accuracy on the validation data, as in conventional approaches, we determine hyperparameters based on their performance in reducing the bullwhip effect after simulating the inventory system. The results show that this approach effectively reduces the bullwhip effect, although the reduction magnitude varies across different models. Moreover, the relationship between forecasting accuracy, the bullwhip effect, and inventory cost is highly dependent on the forecasting model, the nature of the dataset, the underlying demand patterns, and the magnitude of prediction errors. Finally, to integrate prediction and optimization, we introduce a novel loss function designed specifically to reduce both the bullwhip effect and the inventory cost
  9. Keywords:
  10. Bullwhip Effect ; Machine Learning ; Deep Learning ; Integrated Prediction and Optimization ; Inventory System Simulation ; Supply Chain

 Digital Object List

 Bookmark

No TOC