Multi-objective Optimization for Plant Design via Tabu Search

Multi-objective optimization algorithms are employed in chemical process engineering to simultaneously model objectives related to profit, emissions, and safety. The challenge in generating trade-off curves for these problems comes from the nonlinearity and complexity of plant design models, so stochastic optimization techniques are considered in this work to compute Pareto-optimal surfaces. The purpose of this research is to investigate the efficacy and capabilities of the Tabu search algorithm for multi-objective optimization, specifically for plant design models. Traditional Tabu search algorithms have three key characteristics: local intensification, diversification, and the utilization of a Tabu list to apply adaptive memory to guide the search. Local intensification serves to ensure that search areas with several good solutions are searched thoroughly. Diversification allows access to wider regions of the search space. In this work, a Tabu Search algorithm is developed to solve plant design models for multiple objectives. An alkylation process was investigated in order to maximize the profit while simultaneously minimizing the byproduct founation. This is significant as energy requirements decrease with decreasing byproduct formation. Pareto-optimality curves were generated, and results show how tuning the parameters of the algorithm leads to more efficient determination of the Pareto surface, for highly nonlinear plant design models. While the Tabu search algorithm allows for the generation of Pareto-optimality curves, it is limited for larger problems by its lack of directionality in search. Therefore, the implemented multivariable Tabu search algorithm utilizes a greedy heuristic in determining the next region of the search space to investigate. A bounding component was introduced which progressively tightens of bounds, leading to smaller search regions and improving computational efficiency of the algorithm. Results show the efficacy of the novel approach.