Title: Thermal efficiency predictive modelling of dedicated hybrid engines based on an optimal multi-network structure

Authors: Chengqing Wen; Ji Li; Quan Zhou; Guoxiang Lu; Hongming Xu

Addresses: Department of Mechanical Engineering, School of Engineering, University of Birmingham, Birmingham, UK ' Department of Mechanical Engineering, School of Engineering, University of Birmingham, Birmingham, UK ' Department of Mechanical Engineering, School of Engineering, University of Birmingham, Birmingham, UK ' BYD Auto Co Ltd., Guangzhou City, China ' Department of Mechanical Engineering, School of Engineering, University of Birmingham, Birmingham, UK

Abstract: This paper presents an evolutionary data-driven modelling approach for dedicated hybrid engine thermal efficiency (TE) prediction, in which a multi-network structure is developed to further improve the prediction accuracy. This structure enables adaptively segmenting input channels in order to reduce the nonlinearity of data representation in each channel so that the sub-networks can be trained efficiently. In the context, the grey wolf optimisation (GWO) algorithm is applied to find the breakpoints of segmentation for building an optimal multi-network structure. The multilayer perceptron (MLP) is introduced as the basic network due to its simple structure with only two hidden layers. Validated by the experimental data, the accuracy of the multi-network prediction model incorporating GWO improves from 82% to 89%. Also, GWO converges to the optimal solution with 21 iterations compared to 26 for particle swarm optimisation and 31 for the gravitational search algorithm, which demonstrates that GWO has a better performance in this study.

Keywords: optimal network structure; engine thermal efficiency; grey wolf optimisation algorithm; data-driven modelling.

DOI: 10.1504/IJPT.2024.137994

International Journal of Powertrains, 2024 Vol.13 No.1, pp.54 - 74

Received: 25 Apr 2022
Accepted: 05 Nov 2022
Published online: 16 Apr 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article