Optimization of BTEX Emissions and Water Content in TEG Dehydration Using Hybrid Multi-Objective Evolutionary Methods

Abstract

This work introduces a hybrid multi-objective optimization method aimed at improving the triethylene glycol dehydration process, specifically targeting the optimization of emissions from benzene, toluene, ethylbenzene, and xylene, as well as the water content in dry gas. The methodology incorporates various sophisticated optimization frameworks, including the Modified Multi-Objective Grey Wolf Optimizer, Multi-Objective Lichtenberg Algorithm, and Multi-Objective Particle Swarm Optimization, in conjunction with machine learning models such as Gated Recurrent Unit, Multi-Layer Perceptron, Random Forest, and Support Vector Regression, to enhance prediction accuracy and stability. The hybrid model utilizes a multi-stage initialization approach and dynamic parameter modifications to equilibrate exploration and exploitation, thus improving the overall optimization procedure. The study's results illustrate the efficacy of the proposed method, with numerical data indicating a 15% reduction in mean squared error relative to conventional methods. The average inverted generational distance for the ZDT4 test function was 0.032, demonstrating the hybrid model's higher performance compared to solo optimization strategies. The methodology was utilized to predict China's economic cycle, showcasing the resilience of the hybrid approach despite uncertainty shocks. The results underscore the practical significance of employing hybrid optimization methods in gas dehydration processes to reduce environmental emissions while maintaining operational efficiency. The research provides significant insights for enhancing predictive modeling in industrial applications and intricate optimization challenges.