Artificial High-Throughput Aided Design of High-Performance Liquid Electrochromic Devices Based on Multilayer Perceptron Neural Networks

Abstract

Electrochromic devices (ECDs), serving as thermal modulation and decorative components, are extensively employed across various domains. To meet diverse application demands in different contexts, the product parameters of ECDs continually vary, notably within the electrochromic layer. In this work, a multilayer perceptron neural network is leveraged to comprehend the transition from material parameter alterations to device performance parameter changes. Specifically, 56 different testing conditions were evaluated by varying the concentration of the electrochromic molecule and the counter species; each combination was then subject to 6 different testing voltage windows, for a total of 336 different testing conditions. Subsequently, a model comprising three hidden layers and three output neurons is developed, expanding the data set simulation to 2000 groups, yielding an R² over 0.9 on modulation amplitude. Through this model, accurate estimations of performance parameters of numerous unmanufactured liquid-based ECDs are achieved, with projected outcomes highlighting a peak modulation amplitude of 78.38%@620 nm. It is verified that the modulation amplitude prediction error is below 5% and demonstrates stable cyclic performance over 1000 cycles (97.93% at 1.45 V), thereby underscoring the feasibility of our approach. This research serves as a pertinent instance of the synergy between multilayer perceptron neural networks and ECDs, which can effectively provide guidance for more efficiently handling the complex and diverse parameters in the field of electrochromism.