Efficient Surrogate-Based Optimization of Prefractionation Column Using Self-Adaptive Kriging Model with Modified Firefly Algorithm

The optimization of distillation columns is critically important due to their substantial contribution to operational costs in the petrochemical industry. This paper introduces a computationally efficient surrogate-based optimization framework designed explicitly for prefractionation columns. To address the challenges of high computational cost and model accuracy in model-based optimization, a self-adaptive Kriging model, which features automated hyperparameter tuning via Bayesian optimization, is implemented and trained using Latin hypercube sampling of historical process data. By integrating a self-adaptive Kriging model with a modified firefly algorithm, the framework efficiently identifies optimal operating conditions that maximize economic profit while adhering to operational constraints. Case studies demonstrate that the proposed framework achieves superior economic performance, increasing the average final profit by 0.17–0.31% compared to non-adaptive surrogate benchmarks. Furthermore, it is exceptionally stable, achieving a minimal relative standard deviation of only 0.037% in the final profit across 30 independent runs, significantly lower than the 0.266% and 0.237% achieved by the benchmark methods. This study provides a practical and efficient tool to optimize complex distillation columns with limited computational resources.