Hmt: Hybrid Mechanism Transformer for Bio-Fabrication Prediction Under Complex Environmental Conditions

作     者：Song, Yichen Xu, Hu Li, Changdi He, Qunshan Tian, Zijian Liu, Xinggao 

作者机构：State Key Laboratory of Industry Control Technology College of Control Science & Engineering Zhejiang University Hangzhou310027 China 

出 版 物：《SSRN》 

年 卷 期：2023年

核心收录：

主　　题：Fabrication 

摘      要：Hybrid models that integrate mechanism-based and data-driven approaches have not gained widespread adoption in the field of bio-fabrication, primarily due to the inherent complexities of the fabrication process and the limited quality of available data for bio-fabrication. Addressing these limitations, this study presents a novel hybrid model based on the mechanism feature network and the Transformer-based data-driven model (HMT) that effectively overcomes these challenges. The HMT combines mechanism modeling with data-driven techniques to accurately predict the target sequence variables in the bio-fabrication process. Initially, the HMT explores the feature relationships between the time domain and the space domain of the target sequence variables through a mechanism model, facilitating the fusion of these variables across both domains. Subsequently, the fused target variables are modeled and predicted using the Transformer-based model. To achieve high-precision predictions, HMT incorporates feature dimension enhancement, attention weighting, and feature disassembly consecutively. Comparative analysis with various data-driven models, particularly those employing artificial neural network method in the field of bio-fabrication, reveals that the hybrid model outperforms others in terms of mean square error, mean absolute error, mean absolute percentage error, robustness performance, and Theil s U statistic. Moreover,The validity of the feature network based on the mechanism model and the feature fusion method in HMT are proved. The state-of-the-art performance of the HMT in long sequence prediction, periodic prediction, and short sequence prediction confirms its feasibility and potential for industrial applications. © 2023, The Authors. All rights reserved.