A New Parallel Tempering Algorithm for Global Optimization: Applications to Bioprocess Optimization

• Published in: Computer Aided Chemical Engineering Vol. 26; pp. 513 - 518
• Main Authors: Ochoa, Silvia, Repke, Jens-Uwe, Wozny, Günter
• Format: Book Chapter
• Language: English
• Published: 2009
• Subjects: Bioprocess Dynamic Optimization; Bioprocess Optimization; Parallel Tempering; Parameter Identification; Parallel Tempering; Parameter Identification; Bioprocess Dynamic Optimization; Bioprocess Optimization
• ISBN: 0444534334; 9780444534330
• ISSN: 1570-7946
• DOI: 10.1016/S1570-7946(09)70086-0

A new optimization algorithm inspired in the behavior of molecules in solution is presented. The molecules are subject to different forces: intermolecular repulsion, stochastic and friction. Molecules are classified according to their objective function into explorers and refiners. Refiners are subject to higher friction values, forcing the search to a narrow region around their current position (local optimization). Explorers are strongly influenced by repulsion forces resulting in the displacement towards low molecular density zones, allowing the evaluation of unexplored regions (global optimization). The performance of the new molecular-inspired parallel tempering (MIPT) algorithm is compared to other stochastic and gradient-based methods for solving two case studies in bioprocess optimization. It is demonstrated that the MIPT algorithm is very efficient, reaches the best optimal values, explores a wider region of the solution space without requiring much computational effort and it is capable of finding different global and local optima simultaneously.