A Distributionally Robust Optimization Model for Cracking Furnace Scheduling under Product Price Uncertainty

The optimization of cycle scheduling for cracking furnace system is pivotal in enhancing the production efficiency of ethylene plants. However, fluctuations in product prices significantly affect the production profitability and optimization direction of scheduling decisions in ethylene plants. To address this challenge, this study constructs a novel data-driven two-stage distributionallly robust optimization framework, leveraging the advantages of distributionally robust optimization in uncertainty management. First, historical data are collected to design a data-driven ambiguity set with the Wasserstein distance, which encapsulates the core characteristics of uncertain probability distributions. The isolation forest algorithm is then implemented for outlier detection, enabling a more precise data-driven support set. Based on this foundation, a data-driven two-stage distributionally robust optimization model adapted to cracking furnace system scheduling is developed and reformulated into a computationally tractable robust counterpart. Industrial scheduling case validation demonstrates that the proposed strategy effectively utilizes uncertain data features, achieving a significant enhancement in daily profit.