Optimizing Agricultural Order Fulfillment Systems: A Hybrid Tree Search Approach

TL;DR

The paper tackles seed order fulfillment under stochastic arrivals and tight deadlines by modeling the wave scheduling as a Markov Decision Process and solving it with an Adaptive Hybrid Tree Search that augments Monte Carlo Tree Search with problem-specific action-space reduction. It leverages historical data to build time-varying arrival models and uses a dual-criteria candidate set and a dynamic peak-reducing factor to prune actions, enabling long-horizon planning in a large, uncertain domain. Empirical results from a high-fidelity warehouse simulator show that the proposed method significantly outperforms greedy, rolling-horizon, and vanilla MCTS baselines in on-time fulfillment and average delays, across multiple sensitivity analyses. The work demonstrates the practical viability of online, domain-informed planning for complex agricultural fulfillment systems and suggests avenues for real-time adaptation and robustness enhancements.

Abstract

Efficient order fulfillment is vital in the agricultural industry, particularly due to the seasonal nature of seed supply chains. This paper addresses the challenge of optimizing seed orders fulfillment in a centralized warehouse where orders are processed in waves, taking into account the unpredictable arrival of seed stocks and strict order deadlines. We model the wave scheduling problem as a Markov decision process and propose an adaptive hybrid tree search algorithm that combines Monte Carlo tree search with domain-specific knowledge to efficiently navigate the complex, dynamic environment of seed distribution. By leveraging historical data and stochastic modeling, our method enables forecast-informed scheduling decisions that balance immediate requirements with long-term operational efficiency. The key idea is that we can augment Monte Carlo tree search algorithm with problem-specific side information that dynamically reduces the number of candidate actions at each decision step to handle the large state and action spaces that render traditional solution methods computationally intractable. Extensive simulations with realistic parameters, including a diverse range of products, a high volume of orders, and authentic seasonal durations, demonstrate that the proposed approach significantly outperforms existing industry standard methods.

Figures (7)

• Figure 1: Overview of the centralized seed fulfillment process. The process begins with the arrival of seed stocks from multiple sites with stochastic, a priori unknown arrival distributions and ends with the fulfillment of orders with different deadlines and quantities. Our proposed adaptive hybrid tree search approach provides an efficient solution to the wave scheduling problem, optimizing the process of order fulfillment.
• Figure 2: Illustration of the order fulfillment process in the warehouse with (a) incoming product quantities, (b) items placed in containers, (c) orders fulfilled by the system.
• Figure 3: Illustration of the time-varying Markov chain prediction model estimated from historical data: (a) historical data depicting the harvest progress of corn in Iowa; (b) ten instances of simulated harvest progress using the model.
• Figure 4: Illustration of the MDP formulation for the wave scheduling problem. The state space consists of the current inventory levels of all products and the time step. The action space consists of all feasible waves of orders. The transition probability function models the stochastic replenishment of inventory and the time taken to execute the wave.
• Figure 5: Illustrative example of the arrival distribution of two different products over the season. (left) The product has significant inventory available at the beginning of the season when compared to the production in the current season. (right) The product has significantly more inventory arriving in the current season when compared to the inventory available at the beginning of the season.