Optimization of coal gasification process based on a dynamic model management strategy

Abstract

Owing to the complex physical and chemical reactions, optimization of the coal gasification process requires a large computational cost. In chemical engineering design and optimization, the surrogate model is often used to assist the evolutionary algorithm (EA) in solving computationally expensive problems. In this paper, a dynamic model management strategy based on adaptive surrogate selection (ASS) is proposed to identify an appropriate surrogate model to assist the particle swarm optimization (PSO) algorithm faced with complex problems. In ASS, based on the given dataset, a right surrogate model was selected from the diversity model library by adopting the minimum root mean square error and K-fold cross-validation before assisting PSO to identify an optimal solution, and reselected constantly with the addition of a new sample. Based on the ASS, in addition to the adaptive global model, the local model is introduced and selected adaptively to refine the optimal solution more rapidly, then switched dynamically with the global model. The most uncertain sample was also considered for searching the unexplored region and escaping from the local optimum. Comparison of the specific surrogate EAs and three other state-of-the-art surrogate-assisted EAs revealed that the proposed strategy significantly improved the optimization performance of PSO and the effective syngas yield of the coal gasification process.