Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Multi-Objective Capacity-Constrained Optimization of Corn Planting Scheduling

Mingshi Cui, Kunting Qi, Byran Smucker, Durai Sundarmoorthi

TL;DR

The paper tackles capacity-constrained corn planting scheduling by linking weather-informed Growing Degree Units (GDUs) to a multiobjective optimization problem. It blends predictive GDUs—via simple averaging and LSTM—with NSGA-II to generate Pareto fronts for three model formulations that differ in how capacity deviations are penalized, and uses hypervolume to compare fronts before selecting a final plan with TOPSIS. The approach yields Pareto-dominant solutions that improve upon the 2021 Syngenta Challenge submissions, notably reducing the number of harvest weeks and waste while keeping harvest near capacity. By incorporating scenario-specific data and extending to a capacity-estimation variant (Scenario 2), the framework provides a replicable workflow for capacity-aware agricultural scheduling under weather-driven uncertainty.

Abstract

This article describes an improved set of solutions to the problems presented in the 2021 Syngenta Crop Challenge in Analytics \citep{Syngenta2021}. In particular, we use multiobjective optimization and predictive modeling methods to determine a corn planting schedule. The problem involves the following objectives: i. minimize the median and maximum absolute difference between weekly harvest quantity and the storage capacity, the number of nonzero harvest weeks, and the total amount of corn wasted. This is accomplished while respecting planting windows, expected harvest amounts, the growing degree units required to bring seeds to harvest, and historical weather data. We used a Long Short-Term Memory model to predict growing degree units for 2020 and 2021, based on historical data. Then, we used a genetic algorithm, and an extensive search of the tuning parameter space, to produce a Pareto front of solutions for three distinct optimization models related to the Challenge. We evaluate the quality of the Pareto fronts for each model, and use the results to choose a preferred model and final solution. We also provide comparisons between our final solutions, previous solutions submitted to the Challenge, and solutions from other groups.

A Multi-Objective Capacity-Constrained Optimization of Corn Planting Scheduling

TL;DR

The paper tackles capacity-constrained corn planting scheduling by linking weather-informed Growing Degree Units (GDUs) to a multiobjective optimization problem. It blends predictive GDUs—via simple averaging and LSTM—with NSGA-II to generate Pareto fronts for three model formulations that differ in how capacity deviations are penalized, and uses hypervolume to compare fronts before selecting a final plan with TOPSIS. The approach yields Pareto-dominant solutions that improve upon the 2021 Syngenta Challenge submissions, notably reducing the number of harvest weeks and waste while keeping harvest near capacity. By incorporating scenario-specific data and extending to a capacity-estimation variant (Scenario 2), the framework provides a replicable workflow for capacity-aware agricultural scheduling under weather-driven uncertainty.

Abstract

This article describes an improved set of solutions to the problems presented in the 2021 Syngenta Crop Challenge in Analytics \citep{Syngenta2021}. In particular, we use multiobjective optimization and predictive modeling methods to determine a corn planting schedule. The problem involves the following objectives: i. minimize the median and maximum absolute difference between weekly harvest quantity and the storage capacity, the number of nonzero harvest weeks, and the total amount of corn wasted. This is accomplished while respecting planting windows, expected harvest amounts, the growing degree units required to bring seeds to harvest, and historical weather data. We used a Long Short-Term Memory model to predict growing degree units for 2020 and 2021, based on historical data. Then, we used a genetic algorithm, and an extensive search of the tuning parameter space, to produce a Pareto front of solutions for three distinct optimization models related to the Challenge. We evaluate the quality of the Pareto fronts for each model, and use the results to choose a preferred model and final solution. We also provide comparisons between our final solutions, previous solutions submitted to the Challenge, and solutions from other groups.

Paper Structure

This paper contains 19 sections, 3 figures, 5 tables.

Table of Contents

  1. Introduction and Motivation
  2. Problem Description and Formulation
  3. Description
  4. Mathematical Formulation
  5. Model 1: Base
  6. Model 2: First Penalty Model
  7. Model 3: Second Penalty Model
  8. Solution Methodology
  9. Predictive Models for Daily GDUs
  10. Simple Averaging Model
  11. Long Short-Term Memory Model
  12. Predictive Model Results
  13. Multi-objective Optimization Methodology
  14. Solution Strategy
  15. Tuning Parameter Selection
  16. ...and 4 more sections

Figures (3)

  • Figure 1: Overview of our solution methodology.
  • Figure 2: For Site 0, real GDU values between 2009 and 2019, and predicted GDU values for 2020 and 2021, using the LSTM(2,80) model.
  • Figure 3: Comparison between the solution submitted to the Challenge and the solution optimized using the methods described in this paper.