Harnessing a Novel Machine-Learning-Assisted Evolutionary Algorithm to Co-optimize Three Characteristics of an Electrospun Oil Sorbent.

Abstract

The optimization of materials is challenging as it often involves simultaneous manipulation of an assembly of condition parameters, which generates an enormous combinational space. Thus, optimization models and algorithms are widely adopted to accelerate material design and optimization. However, most optimization strategies can poorly handle multiple parameters simultaneously with limited prior knowledge. Herein, we describe a novel systematic optimization strategy, namely, machine-learning-assisted differential evolution, which combines machine learning and the evolutionary algorithm together, for zero-prior-data, rapid, and simultaneous optimization of multiple objectives. The strategy enables the evolutionary algorithm to "learn" so as to accelerate the optimization process, and also to identify quantitative interactions between the condition parameters and functional characteristics of the material. The performance of the strategy is verified by in silico simulations, as well as an application on simultaneously optimizing three characteristics, namely, water contact angle, oil absorption capacity, and mechanical strength, of an electrospun polystyrene/polyacrylonitrile (PS/PAN) material as a potential sorbent for a marine oil spill. With only 50 tests, the optimal fabrication parameters were successfully located from a combinatorial space of 50 000 possibilities. The presented platform technique offers a universal enabling technology to identify the optimal conditions rapidly from a daunting parameter space to synthesize materials with multiple desired functionalities.