Multi-modal soft measurement modeling method based on hidden Markov modal division fermentation process

By combining Hidden Markov Models (HMMs) and LightGBM, multimodal soft measurement modeling of the fermentation process was achieved. This solved the problems of difficulty in real-time measurement of key parameters during fermentation and poor adaptability of traditional models, improved prediction accuracy and model transparency, and provided a theoretical basis for process optimization.

CN122157825APending Publication Date: 2026-06-05BEIJING UNIV OF CHEM TECH
0 Cites 0 Cited by

Patent Information

Application Number
CN202610256240.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-03-04
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

Key parameters during fermentation are difficult to measure accurately in real time. Traditional single-modal soft measurement models have poor adaptability, and cluster analysis methods lack time-series analysis capabilities, resulting in discontinuous modality division results that are difficult to conform to the gradual change pattern of the fermentation process.

Method used

A multimodal soft measurement modeling method based on Hidden Markov Models is adopted. Through data preprocessing, Hidden Markov Model parameter configuration and training, Viterbi algorithm decoding and LightGBM model construction, the modal division and stage-specific prediction of the fermentation process are realized. The interpretability of the model is improved by combining the SHAP analysis method.

Benefits of technology

It significantly improved the overall prediction accuracy and model adaptability of key parameters, enhanced the interpretability of the model, and provided reliable data-driven support for process optimization of fermentation.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention discloses a fermentation process multi-modal soft measurement modeling method based on hidden Markov modal division. The method is used for solving the problems that in the fermentation process, key parameters are difficult to measure on line, and the adaptability of a single model is insufficient. The method comprises the following steps: firstly, preprocessing historical data, performing modal division on time series data by using a hidden Markov model, and identifying different fermentation stages; based on a division result, respectively training LightGBM soft measurement sub-models of multiple stage specificity so as to accurately describe a non-linear relationship among parameters of each stage; and finally, an SHAP interpretability analysis method is introduced, the contribution degree of each process variable to a prediction result is quantified, and the model interpretability is enhanced. According to the method, through multi-modal modeling and stage self-adaptive prediction, the online prediction precision of the key parameters is improved, and the method has good engineering application value.
Need to check novelty before this filing date? Find Prior Art