Population-distributed stochastic optimization for distillation processes: Implementation and distribution strategy

Stochastic optimization is inefficient, although it shows robustness against local optimum and can guarantee high-quality solutions. Parallel computation can be a promising way to improve the efficiency of stochastic optimization. However, the common environments do not support calling multiple simulators through the win32com interface, which hinders parallel computation. As a countermeasure, this study proposes a population-distributed differential evolution (DDE) framework, which combines multiple optimizers through the shared message passing medium. The framework distributes the population into groups (subpopulations) on different threads by a pool model, which can make full use of a multi-core CPU and significantly accelerate the computation. Moreover, we considered both the synchronously and asynchronously distributed differential evolution. Three case studies (benzene/toluene/xylene conventional distillation, acetone/methanol/water extractive distillation, and heat pump assisted dividing-wall column separating benzene/toluene/xylene) are optimized to show the superior performance of the DDEs. The parallel framework can reduce the computing time by ∼70% on a 4-core CPU, which is a significant improvement. DDEs cause some parallel efficiency loss, which is 5–10% and 10–20% for ADDE and SDDE, respectively. Further, based on time consumption analysis, we explain the reasons for the efficiency loss.