Energy storage efficiency modeling of high-entropy dielectric capacitors using extreme learning machine and swarm-based hybrid support vector regression computational methods

Abstract

• Models were developed for predicting energy storage efficiency of dielectric capacitors • SG-ELM model outperforms SI-ELM and PS-SVR models for energy storage efficiency prediction • Dependency of storage efficiency and coersive energy was established using SG-ELM • Dependency on concentration of dopant in A and B-sites was established using SG-ELM Lead-free dielectric energy storage BaTiO 3 ceramics are eco-friendly and high-performance energy storage materials with excellent storage stability, high power density, ultra-fast discharge and charge speed, excellent mechanical strength and long service life. These materials have potential applications in many modern power electronics industries and other technological areas where energy consumption plays a significant role. However, low energy storage efficiency of BaTiO 3 -based ceramics hinders their applications especially along the global trend of miniaturization. This work proposes a computational method of addressing low energy storage efficiency of BaTiO 3 -based ceramics. Crystallographic modification of BaTiO 3 ceramics to high-entropy perovskite materials with relaxor ferro-electric, diffused ferro-electric and para-electric features potentially leads to enhanced energy storage efficiency with various substitutions at A and B-sites of the perovskite. The dependence of energy storage efficiency of high-entropy ceramics on dopants substitutions, concentrations and coercive energy is experimentally tedious, costly and time consuming hence the need for computational and theoretical approaches to model the energy storage efficiency of these materials. This work employs single hidden layer extreme learning machine (ELM) algorithm and hybrid particle swarm optimization-based support vector regression (PS-SVR) for determining energy storage efficiency of high-entropy ceramics. The developed sigmoid (SG) activation function-based ELM (SG-ELM) shows performance improvement over sine (SI) function-based ELM (SI-ELM) model and PS-SVR model with an improvement of 79.25 % and 89.4 % using root mean square error (RMSE) performance measuring parameter. The dependency of energy storage efficiency on coercive energy and concentration of dopants in A and B-sites of the perovskite was established using the developed SG-ELM model. The precision and accuracy demonstrated by the developed intelligent models facilitate efficiency enhancement in high-entropy ceramic systems for many energy storage applications with the ultimate goal of addressing the global energy challenges.