Integrating Deep Learning Demand Forecasting with Multi-Objective Optimization for Circular Coffee Supply Chains: A Data-Driven Framework for Cost, Emissions, and Freshness Management

2026-06-06

AI Summary

This study addresses the fragmented treatment of demand forecasting, optimization, and traceability in coffee supply chains, which often impedes simultaneous consideration of cost, carbon emissions, and product freshness. The authors propose a two-stage data-driven framework: first, a CNN-LSTM hybrid model achieves high-accuracy demand forecasting (MAE = 22.87, R² = 0.90); second, these forecasts are integrated into a closed-loop, multi-period, intermodal network formulated as a tri-objective mixed-integer linear programming model—uniquely combining deep learning-based prediction with circular economy–oriented multi-objective optimization. An inventory-age–based exponential decay model for freshness is innovatively incorporated. Using the ε-constraint method, 25 Pareto-optimal solutions are generated to support policy trade-offs; notably, a balanced solution increases costs by only 9.9% while reducing carbon emissions by 22.4% and maintaining near-optimal freshness.

Abstract

The coffee supply chain is one of the most complex agri-food networks, marked by geographically dispersed production, multi-tier coordination, and high sensitivity to quality and freshness. While sustainability and digitalization have gained attention, demand forecasting, optimization, and traceability are often treated separately. This study presents a two-phase integrated framework. First, a hybrid CNN-LSTM model is used for demand forecasting. On the public Coffee Chain Sales dataset with chronological 70/15/15 splitting, the model achieves MAE of 22.87 and R^2 of 0.90, outperforming the best deep learning benchmark by ~12% and classical methods by over 30%. In the second phase, the forecasted demand feeds a tri-objective mixed-integer linear programming (MILP) model that jointly minimizes cost, minimizes carbon emissions, and maximizes product freshness in a multi-period, multimodal, closed-loop supply chain with circular recovery. Freshness is modeled via exponential decay based on inventory age. Using the epsilon-constraint method, 25 Pareto solutions are obtained. Sensitivity and policy analyses show that balanced sustainability policies can reduce emissions by 22.4% with only a 9.9% cost increase while maintaining near-optimal freshness. Keywords: Coffee supply chain; Deep learning; Demand forecasting; Multi-objective optimization; Circular economy; CNN-LSTM; Mixed-integer linear programming.

Problem

• Coffee supply chain
• Demand forecasting
• Multi-objective optimization
• Circular economy
• Freshness management

Innovation

• CNN-LSTM
• Multi-objective optimization
• Circular supply chain
• Demand forecasting
• Mixed-integer linear programming