Machine Learning Test for Modulation Range of Ammonium Metatungstate Based Liquid Electrochromic Devices

In recent years, machine learning (ML) has been widely applied in material science for material synthesis and molecular structural prediction. However, the application of machine learning in the field of electrochromic devices (ECDs) is relatively limited and only involves traditional solid-state ECDs. In comparison to solid-state devices, liquid devices have simpler structures and better performance, making them a promising research direction for the future. In this study, we explore the effects of ferrous chloride and ferrous sulfate as additives on ammonium metatungstate liquid ECDs. Electrochromic solutions with different concentrations were synthesized using the hydrothermal method, to fabricate three-layer electrode / electrochromic liquid / electrode devices. The alternation of transmittance at different current were measured to calculate the modulation range. Using the measuring results as training data, seven different regression algorithms were used to construct the modulation range models of these two kinds of ECDs, and their generalization ability was compared. In addition, we used different models to predict the solution formulations of ECDs with optimal modulation range, then fabricated new ECDs based on these formulations to verify the predictions. It turns out that modulation range models using decision tree regression and kernel ridge regression have the best prediction performance. In addition, considering the model generalization ability and prediction accuracy for the optimal formulation, decision tree regression is the best ML algorithm for both ammonium metatungstate-ferrous chloride and ammonium metatungstate-ferrous sulfate based ECDs.