Joint Districting and Transportation Lot-Sizing Models with Machine Learning-Driven Yield Prediction for Sugarcane

The sugarcane management systems aim to procure raw materials and deliver them to production facilities on time while minimizing the total expected cost. According to industry interviews, sugar mills promote sugarcane cultivation within designated zones for providing lower-interest financing and agronomic support. To support this management strategy, we propose a joint districting and transportation lot-sizing framework that partitions the cultivation region into two districts: a promoted (inside) area and a non-promoted (outside) area prior to cultivation. Machine learning is proposed to predict the sugarcane yields in the cultivation period. Support vector machine regressors with different kernel functions enhance predictive accuracy, while decision tree regression offers improved interpretability and practical insights for decision-makers. A mathematical model and a hybrid metaheuristic algorithm (Combined Simulated Annealing Algorithm and Heuristic, CSAH) were developed to determine the transportation lot sizing for harvesting and transportation operations under uncertainty in sugarcane yields. The objective is to minimize the total expected cost, including cultivation, harvesting, holding, and transportation costs, while selecting the preferred solution by evaluating the workload balance among truck stations on the Pareto frontier. The performance of CSAH is compared with that of current operation (Separated Model, SM), in which districting and transportation lot-sizing decisions are solved independently. Computational experiments are conducted using three real-world datasets, including instances that capture uncertainty associated with first-, second-, and third-stubble sugarcane harvests. The results demonstrate that the CSAH model achieves greater efficiency than the SM by reducing the expected total cost and maintaining robust performance across different geographic distributions and yield scenarios.