Automatically Generated Problem-Tailored Genetic Algorithms for the Optimization of Chemical Processes

The performance of chemical processes is often determined by the selectivity and activity of catalysts used in them. Optimizing a catalyst accordingly leads to a high-dimensional constraint optimization task for both continuous and discrete variables. To solve that task, genetic algorithms are most frequently used in catalysis, though their routine use is still hindered by a lack of appropriate implementations. Generic implementations of this method do not address all features of the optimization task, and use low-level coding of input variables, which is unacceptable for chemists. On the other hand, specific algorithms developed directly for the optimization of catalytic materials are usable only for a narrow spectrum of problems. This paper presents an approach the main idea of which is to automatically generate problem-tailored genetic algorithms from requirements concerning the optimized materials. For a specification of those requirements, a formal description language has been developed. To automatically generate corresponding algorithms from the formal descriptions, a program generator is needed. In this paper, the requirements expressible with the description language are reviewed and an overall scheme of the approach is outlined. Finally, a first prototype of a program generator for algorithms generated from that language is sketched. Key-Words: Computer applications in chemistry, Optimization methods, Empirical objective function, Genetic algorithms, Problem-tailoring, Formal description language, Program generator