Modeling thermoelectric figure of merit of BiCuSeO oxyselenide ceramic using extreme learning machine computational method

Abstract

BiCuSeO ceramic is an eco-friendly oxyselenide thermoelectric material that provides alternative directions for refrigeration and power generation through direct conversion between electrical and thermal energy. Aside from environmental friendliness of BiCuSeO ceramic, relatively cheap input materials, low toxicity, good thermal and chemical stability are among the unique features that contribute significantly to wider applicability of this material for energy conversion. The energy conversion efficiency of un-doped and pristine BiCuSeO thermoelectric material is very low while enhancement of its energy conversion efficiency has attracted serious attention from both experimental and theoretical perspectives thorough modulation doping, dual-vacancies creation and dual doping among others. This work models the thermoelectric figure of merit (energy conversion efficiency, ECE) of doped BiCuSeO based ceramics using extreme learning machine (ELM) with ionic radii and elemental concentrations predictors. ECE-SNELM model with sine activation function performs better than ECE-SGELM with sigmoid activation function using root mean square error (RMSE), correlation coefficient (CC) and mean absolute error (MAE) performance evaluation and assessment parameters with improvement of 20.61 %, 5.51 % and 5.77 %, respectively when computed using testing samples of BiCuSeO-based ceramics. Comparison of the developed ECE-SNELM and ECE-SGELM model with the existing machine learning (ML) based model shows that ECE-SNELM and ECE-SGELM model respectively outperforms ML (2022) model with improvement of 70.89 % and 35.8 % using RMSE metric. The developed ECE-T-SNELM model that includes temperature among its descriptors further outperforms the existing ML based model with improvement of 39.39 % on the basis of RMSE metric. The predictions of the developed models agree well with the measured values. The developed ECE-SNELM was further employed for investigating the behavior of BiCuSeO-based ceramics after dopant incorporation. The precision demonstrated by the developed ECE-SNELM and ECE-SGELM models in determining thermoelectric figure of merit of BiCuSeO-based ceramics is highly commendable and meritorious in strengthen energy conversion efficiency of the ceramics for addressing the global challenges of energy crisis.