Abstract
The global agrochemical market is highly consolidated, with large multinational companies accounting for a major share of the market. Thus, even for a single agrochemical product, its global supply chain typically involves numerous paths connecting the raw material sources to the final customers. Besides structural complexity, agrochemical supply chains are also subject to seasonality and other unique uncertainties, thereby posing a need for risk management tools and strategies. In this study, we model and optimize an agrochemical supply chain by developing and solving a stochastic mixed-integer quadratic constrained program (MIQCP). We model and control the demand uncertainty in this scenario-based MIQCP using variance. We also apply perspective reformulation techniques to convert the MIQCP to a mixed-integer linear program (MILP). Computational experiment results from an illustrative example show that, successively introducing perspective cuts to the reformulated MILP not only leads to a tight approximation of the original MIQCP model, but is also more computationally efficient than directly solving the MIQCP.
Introduction
The global agrochemical market—comprising pesticides, herbicides, and fertilizers—is characterized by extreme consolidation. Currently, the top four conglomerates control over 60% of the market share. Managing these supply chains is a significant logistical challenge; they involve multi-stage networks that must bridge raw material sources with global customers while navigating the inherent volatility of seasonality, climate change, and unpredictable "black swan" events.
Existing optimization models for these networks often face two primary hurdles: risk quantification and computational complexity. Many traditional models rely on Conditional Value-at-Risk (CVaR), which can be overly sensitive to data outliers, often resulting in non-convex and unstable optimization problems. Furthermore, incorporating risk measures frequently transforms linear models into Mixed-Integer Quadratic Constrained Programs (MIQCPs), which are notoriously difficult to solve for large-scale global networks. This research addresses these challenges by proposing a robust, computationally efficient framework for supply chain design.
Methodology
Risk Metric: Variance vs. CVaR
Rather than using CVaR, this study employs the variance of demand loss across planning periods as the primary risk measure. Variance serves as a stabilizer, providing a more consistent characterization of demand unfulfillment risk. By focusing on variance, the model ensures the supply chain is not merely efficient on average but is resilient against the fluctuations and unpredictable outcomes that outliers can cause in portfolio-style optimization.
Perspective Reformulation
The introduction of variance as a risk metric creates quadratic (nonlinear) constraints, making the MIQCP difficult to solve. To overcome this, the authors utilize perspective reformulation. This technique involves generating "perspective cuts"—linear inequalities that approximate the convex envelope of the nonlinear constraints. By iteratively adding these cuts, the MIQCP is converted into a Mixed-Integer Linear Program (MILP). This allows the use of high-performance linear solvers to find tighter approximations more rapidly than solving the original quadratic formulation.
Illustrative Case Study
To validate the framework, the study utilizes an illustrative case study involving two Active Ingredient (AI) production plants, two warehouses, and two market regions. The model incorporates a wide array of operational parameters, including fixed and variable costs, holding costs, shipping capacities, and production durations.
The uncertainty is modeled through three distinct weekly demand scenarios for each market region, as detailed in the table below:
Key Results and Insights
The performance of the reformulated MILP was compared against the original MIQCP using a 600-second solve time limit. The results highlight the superiority of the linearization approach:
- Computational Efficiency: The reformulated MILP achieved an optimality gap of 8.72%, significantly outperforming the 12.29% gap of the original MIQCP.
- Cost Effectiveness: The MILP yielded an objective value of **$14,886**, which is **2.59% lower** than the MIQCP's $15,282.
- Operational Resilience: Interestingly, the MILP found a physically superior solution. While the original MIQCP solver became stuck in a solution allowing for 5 units of demand loss, the reformulated model successfully identified a configuration with zero demand loss.
These findings suggest that perspective cuts provide a much tighter approximation of the original problem than conventional branch-and-cut approaches, leading to better physical supply chain configurations.
Conclusion
This research demonstrates that variance is a viable and stable risk measure for agrochemical supply chain optimization. By applying perspective reformulation, complex quadratic models can be transformed into efficient linear programs that are easier for solvers to navigate. While future work is needed to test this framework on larger, truly global networks and non-convex Mixed-Integer Nonlinear Programs (MINLPs), this study provides a clear path toward more resilient and computationally manageable global supply chain designs.
Summary updated