By Andrew Mohn and Adriele Pradi
Editor’s Note: The SCM thesis Leveraging Simulation-Based Optimization to Generate Optimal Transportation Plans in the Real World was authored by Andrew Mohn and Adriele Pradi and supervised by Dr. Ilya Jackson (ilyajack@mit.edu). For more information on the research, please contact the thesis supervisor.
Transportation planning for the furniture and home goods industry presents a complex puzzle. Online retailers, like the company we’re about to uncover, face the challenge of efficiently moving products from warehouses to customers’ doorsteps. However, traditional planning methods often fail to account for the inherent randomness and variation in real-world operations.
In our research, we explore the transformative journey of a large online retailer as it embraces the power of probability in middle-mile transportation planning. Their approach relied on point forecasts, not including the unpredictability of real-world operations. Consequently, disruptions, suboptimal outcomes, and increased costs were part of their logistics network, challenging their promise of timely deliveries.
Our focus was on developing a solution that leveraged probability distributions and stochastic inputs to create robust transportation plans. The key challenge was to incorporate these methods effectively, considering the complexity of implementation and the need for computational resources. To tackle this problem, we developed a simulation model within a scenario-based framework. This approach involved sampling uncertain inputs’ distributions and repeatedly solving the optimization problem with realized values. By doing so, we aimed to evaluate the model’s performance, identify blind spots, and assess improvements in plan optimization.
Embracing the unpredictable for optimal delivery
The company’s current system relies on point forecasts, which do not adequately consider the randomness and variation inherent in real-world operations. The research was motivated by the need to develop a solution that more accurately accounts for the probability distributions of inputs. Techniques such as stochastic and robust optimization, simulation and scenario testing, and reinforcement learning have been shown to address this issue effectively. However, implementing these methods at scale poses challenges in terms of expertise and computational resources.
To address these challenges, we focused on developing a simulation model using a scenario-based framework. The goal was to provide a practical approach to incorporating probability distributions into transportation planning. By leveraging this approach, our project aims to analyze and mitigate the impact of uncertainties in two primary sources of variation: facility operations and driving time between facilities.
The objective of our project is to improve plan performance and minimize disruptions by evaluating the simulation model’s effectiveness in comparison to the existing approach. The central metric of evaluation is the company’s ability to fulfill the customer promise, which measures its consistency in meeting delivery timelines.
Uncovering logistics uncertainty through data and simulation
Simulation results demonstrated differences between the current transportation plan outcomes and those generated by the scenario-based model. Certain lanes showed higher yard dwell times in the simulation compared to the service level agreement-based plan, while others exhibited the opposite trend. This suggests that service level agreements may be overly optimistic or conservative for specific lanes. Our research showcased the capability of stochastic optimization models to capture the inherent variability in transportation time and dwell, enabling more robust transportation plans that better reflect real-world conditions.
While our study highlighted the benefits of stochastic optimization, it also identified certain limitations. The accuracy of input data, specifically estimated distributions of transportation time and dwell, plays a crucial role in determining result quality. To address this, future research should focus on improving these estimates by incorporating real-time data and leveraging advanced analytics techniques. Moreover, the computational complexity associated with stochastic optimization models may pose challenges in their practical implementation for large-scale transportation networks. Exploring avenues to enhance the efficiency of these models would render them more accessible and feasible for widespread adoption.
Every year, approximately 80 students in the MIT Center for Transportation & Logistics’ (MIT CTL) Master of Supply Chain Management (SCM) program complete approximately 45 one-year research projects.
These students are early-career business professionals from multiple countries, with two to 10 years of experience in the industry. Most of the research projects are chosen, sponsored by, and carried out in collaboration with multinational corporations. Joint teams that include MIT SCM students and MIT CTL faculty work on real-world problems. In this series, they summarize a selection of the latest SCM research.