Ingredient Planning for Copper Industry: A Deep Reinforcement Learning-Based $\epsilon$-Constrained Multi-Objective Optimization Framework

Effective ingredient plan plays a crucial role in the copper industry as it directly impacts both economic interest and resource utilization of the enterprise. Aims at minimizing both production cost and loss simultaneously, a multi-objective ingredient optimization model considering practical constraints is established in this study. In order to effectively address the model, a deep reinforcement learning-based
$\varepsilon$
-constrained multi-objective optimization framework (DRL-ECMO) is designed. In detail, a dynamic assistant population is introduced, in which the involved solutions are infeasible but potentially valuable for giving further search directions. Then a new
$\varepsilon$
-constrained method is correspondingly developed for the assistant population, where
$\varepsilon$
value is adaptively adjusted based on deep reinforcement learning (DRL). Finally, the experimental results based on real world data demonstrate the effectiveness and superiority of the proposed framework comparing with other state-of-the-art algorithms and methods for ingredient planning optimization in copper industry.