READ ME File For 'Dataset in support of the publication 'Segmented thermoelectric generator modelling and optimization using artificial neural networks by iterative training'

Dataset DOI: 10.5258/SOTON/D2454

Date that the file was created: July, 2023

-------------------
GENERAL INFORMATION
-------------------

ReadMe Author: Yuxiao Zhu, University of Southampton [OPTIONAL add ORCID ID]

Date of data collection: ADD IN COLLECTION DATES 2021 - 2022


--------------------------
SHARING/ACCESS INFORMATION
-------------------------- 

Licenses/restrictions placed on the data, or limitations of reuse:CC-BY


This dataset supports the publication:
AUTHORS:Zhu, Y., Newbrook, D., Dai, P., De Groot, K., Huang, R. And Liu, J.
TITLE:Segmented thermoelectric generator modelling and optimization using artificial neural networks by iterative training
JOURNAL:Energy and AI
PAPER DOI IF KNOWN:https://doi.org/10.1016/j.egyai.2022.100225

--------------------
DATA & FILE OVERVIEW
--------------------
The excel file contains raw data for the paper, ”Segmented thermoelectric generator modelling and optimization using artificial neural networks by iterative training”. The detailed description are below: 

Fig. 1. (a) Schematic of the singe-pair segmented thermoelectric generator modelled in this study. Temperature-dependent. (b) the ZT of the n-type and p-type thermoelectric materials selected for the STEG in this work.

Fig.2. The architecture of the forward modelling ANN for predicting the power density of the STEG model. The input layer contains design parameters (H_TE,FF,α_N,α_P) and operating conditions ( ρ_ct,ρ_cb,Q_in). The output layer contains power performance values (〖PD〗_max).

Fig. 3. Iterative ANN training process flowchart.

Fig. 4. Distribution of the design parameters (a) H_TE, (b) FF, (c) α_N, (d) α_P in the 1000 uniform datasets (red) and biased datasets (blue).

Fig. 5. Comparison of COMSOL output 〖PD〗_max  for two different ANNs (a) the histogram of the probability and average relative errors of two ANNs on the uniform test dataset, and (b) Uni4000, (c) Iter4000 on the uniform test dataset. Comparison of COMSOL output 〖PD〗_max  for two different ANNs, (d) the histogram of the probability and average relative errors of two ANNs on the biased test dataset, and (e) Uni4000, (f) Iter4000 on the biased test dataset.

Fig. 6. Genetic algorithm optimized STEG 〖PD〗_max using ANN Iter4000 (blue), uniform ANN Uni4000 (red), and COMSOL simulation as forwarding modellers (a), and the associated relative errors from Iter4000 (blue) and Uni4000 (red) compared with the COMSOL (b) as a function of different Q_in; Genetic algorithm optimized STEG 〖PD〗_max using ANN Iter4000 (blue), uniform ANN Uni4000 (red), and COMSOL simulation as forwarding modellers (c), and the associated relative errors from Iter4000 (blue) and Uni4000 (red) compared with the COMSOL (d) as a function of different ρ_c.

Fig. 7. 〖PD〗_max obtained from iterative ANN (line) and COMSOL simulation (triangles) as a function of (a) H_TE and FF, (b) Q_in, and FF.

Fig. 8. 〖PD〗_max of iterative ANN obtained by scanning  α_N and α_P at different heat flux (Q_in) conditions of (a) 500 mW/〖cm〗^2, (b) 500 mW/〖cm〗^2, (c) 500 mW/〖cm〗^2, respectively with other parameters fixed; 〖PD〗_max of  iterative ANN obtained by scanning  α_N and α_P at different electrical conductivity (ρ_ct=ρ_cb=10^(-9)) conditions (d) 10^(-9) Ω·m^2(e) 〖5×10〗^(-8) Ω·m^2 and (f) 10^(-7) Ω·m^2, respectively.