A data-based approach for the development of asynchronous and distributed optimization algorithms in complex engineering applications

Abstract The paper outlines a generic optimization approach with extensive capabilities to distribute the computational work in the optimization, also to explore interim data generated by the optimization algorithms without penalizing search times and effort. The approach postulates a network of autonomous computing nodes and data pools (data repositories); nodes and pools respectively generate and host interim solutions that are redistributed in the network with a parallel use of data analytics that constitute asynchronous tasks. The approach is demonstrated in complex problems of (single and multiphase) reactor network synthesis. The algorithm is implemented using an infrastructure of multiple CPU cores. Results confirm significant reductions of computing times due to parallelization, also the benefit to explore cost-free analytics over data generated internally. Future work includes a more systematic and extensive use of machine learning, also the use of dynamic networks of nodes and pools.