Modeling thermoelectric performance of doped BiCuSeO oxychalcogenide ceramics using genetically hybridized support vector regression computational method

Abstract

Listen

Translate article

Oxychalcogenide BiCuSeO thermoelectric material has demonstrated significant potentials in addressing energy crisis through thermal to electrical energy conversion. Low toxicity, reduced cost of input materials and good stability when operated at high temperatures are unique features that promote the candidature of BiCuSeO ceramics. However, the thermoelectric figure of merit (FM) which influences the conversion (energy) efficiency of BiCuSeO material is characterized with small value due to low carrier mobility and concentration. Incorporation of various kind of dopants at different sites, vacancies and defects creation are among the experimental approaches of enhancing thermoelectric performance (usually, figure of merit). In order to circumvent laborious experimental procedures and to strengthen material design for sustainable applications, the maximum value of thermoelectric figure of merit of doped BiCuSeO ceramics is modeled in this work through genetic evolutionary algorithm hybridized support vector regression (HSVR) using ionic radii and dopant concentrations-based descriptors. FM-HSVR-G model developed with Gaussian kernel performs better than FM-HSVR-P that employs polynomial transformation kernel using different performance assessment parameters such as root mean square error (RMSE), correlation coefficient (CC) and mean absolute error (MAE). When validated on the testing samples of BiCuSeO ceramics, FM-HSVR-G model shows 45.40 % (for CC parameter), 16.87 % (for MAE parameter) and 14.00 % (for RMSE parameter) performance improvement over FM-HSVR-P model. Comparison of the present models with the existing machine learning (ML) model indicates that FM-HSVR-P and FM-HSVR-G model outperform ML (2022) model with enhancement of 22.05% and 72.04 %, respectively using metric based on RMSE. Ability of some elemental dopants to strengthen the thermoelectric performance of pristine BiCuSeO ceramic was investigated using the model developed while the obtained behavior agrees well with the experimental trends. The characteristic precision and accuracy of the developed models would definitely strengthen thermoelectric performance of BiCuSeO based compounds and further facilitate BiCuSeO ceramic material design for controlling energy crisis and other sustainable applications.