Title: Analysis and optimisation of RON loss via compound variable selection and BP neural network

Authors: Yunshu Dai; Jianwei Fei; Fei Gu; Chengsheng Yuan

Addresses: School of Computer and Software, Nanjing University of Information Science and Technology, Nanjing, 210044, China ' School of Computer and Software, Nanjing University of Information Science and Technology, Nanjing, 210044, China ' School of Computer and Software, Nanjing University of Information Science and Technology, Nanjing, 210044, China ' School of Computer and Software, Nanjing University of Information Science and Technology, Nanjing, 210044, China

Abstract: The loss of octane in the gasoline refining process can cause huge economic losses. However, the analysis and optimisation of octane loss is a high-dimensional nonlinear programming problem. In this work, we propose a compound variable selection scheme. Based on the results of independent variables by outlier and high correlation filtering, the representative operations are selected by random forest and grey correlation analysis, and the octane loss is then predicted by the BP neural network and XGBoost. To optimise the octane loss, an operation optimisation scheme based on fast gradient modification (FGM) is proposed. Experiments show that the composite variable selection scheme proposed in this paper can effectively screen independent and representative variables and has high prediction accuracy for octane loss. The proposed optimisation method also has sufficient feasibility and meets the needs of real scenes.

Keywords: RON loss optimisation; variable selection; XGBoost; BP neural network; correlation analysis; fast gradient modification; nonlinear programming.

DOI: 10.1504/IJAACS.2024.138149

International Journal of Autonomous and Adaptive Communications Systems, 2024 Vol.17 No.3, pp.201 - 214

Received: 18 Jan 2022
Accepted: 06 Feb 2022
Published online: 30 Apr 2024 *

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