MVSG-GS: A metaheuristic virtual sample generation method for soft sensor modeling based on guidelines sharing

• Published in: Expert systems with applications Vol. 290; p. 128427
• Main Authors: Peng, Yu, Li, Erchao
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 25.09.2025
• Subjects: Metaheuristic; Neuroevolution; Soft sensors; Virtual sample generation; Wastewater treatment process; Virtual sample generation; Metaheuristic; Soft sensors; Wastewater treatment process; Neuroevolution
• ISSN: 0957-4174
• DOI: 10.1016/j.eswa.2025.128427

In the modern chemical industry, obtaining a sufficient number of samples for developing soft sensors can be challenging due to physical limitations and the high cost of measurements. The scarcity and uneven distribution of modeling data significantly hinder the widespread application of data-driven methods in intelligent optimization. Taking inspiration from generative adversarial networks and neuroevolution, a metaheuristic virtual sample generation model based on guidelines sharing (MVSG-GS) is proposed to enhance the data quality and diversity. MVSG-GS learns the overall distribution of original data based on metaheuristic optimization framework, and improves the consistency with structured low-entropy data. Further, a sparse knowledge transfer mechanism is designed to reveal the implicit relevance of multivariate data in biochemical processes, enabling multi-task parallel modeling of complex reaction processes. Experiments based on different datasets show that the proposed model outperforms the baseline approach in terms of accuracy and effectiveness. Based on the results, virtual samples generated by the proposed method show a closer resemblance to the real samples when compared to the other seven competitors. Furthermore, as an open framework, MVSG-GS introduces an effective approach to enhancing limited training data, contributing to improving the prediction accuracy and broader model generalization.