A Memetic NSGA-III for Green Flexible Production with Real-Time Energy Costs & Emissions

TL;DR

The paper tackles tri-objective green scheduling for flexible production under real-time energy costs and emissions by formulating a Flexible Job Shop Scheduling Problem with objectives $c^{max}$, $p^{sum}$, and $e^{sum}$ and solving it with a memetic NSGA-III. It extends a time-expanded MILP to include emissions and introduces a decoder-based genotype-phenotype representation with four gene strings, plus a greedy local refinement to reduce costs and emissions without increasing makespan. Experiments on Brandimarte benchmarks enriched with German energy-market data demonstrate that small increases in makespan can yield substantial energy-cost and emission savings, with energy-cost savings typically larger than emission reductions. The results justify production flexibility in dynamic grids and provide a framework for practitioners to balance makespan, energy costs, and emissions under real-time price signals, while suggesting avenues for uncertainty handling and real-production validation.

Abstract

The use of renewable energies strengthens decarbonization strategies. To integrate volatile renewable sources, energy systems require grid expansion, storage capabilities, or flexible consumption. This study focuses on industries that adapt production to real-time energy markets, offering flexible consumption to the grid. Flexible production considers not only traditional goals like minimizing production time, but also minimizing energy costs and emissions, thereby enhancing the sustainability of businesses. However, existing research focuses on single goals, neglects the combination of makespan, energy costs, and emissions, or assumes constant or periodic tariffs instead of a dynamic energy market. We present a novel memetic NSGA-III to minimize makespan, energy cost, and emissions, integrating real energy market data, and allowing manufacturers to adapt energy consumption to current grid conditions. Evaluating it with benchmark instances from literature and real energy market data, we explore the trade-offs between objectives, showcasing potential savings in energy costs and emissions on estimated Pareto fronts.

Figures (7)

• Figure 1: Example genotype for solution encoding
• Figure 2: Representation of the example genotype as a phenotype
• Figure 3: Example genotype for solution encoding based on burmeister2023memetic
• Figure 4: Representation of the example genotype as a phenotype based on burmeister2023memetic
• Figure 5: Greedy local refinement