Scholarlink

<p>Daqu is a solid-state starter used in the brewing of traditional Chinese liquor. The change in moisture content during Daqu fermentation is a key factor for measuring its quality. However, the traditional method for measuring the moisture content in Daqu is cumbersome and time-consuming, making the rapid detection of Daqu moisture during fermentation difficult. Therefore, this study aimed to propose an LSTM-Attention-LightGBM time-series prediction model to rapidly predict Daqu moisture using the environmental parameters of Daqu fermentation. Combining the characteristics of multi-point environmental parameters, the RF algorithm was used for feature point screening. Also, the Prophet model and expanding window technique were employed to extract features from Daqu moisture and the screened environmental data, respectively, so as to capture their change trends and fluctuations. The comparative analysis revealed that the LSTM-Attention-LightGBM model performed the best in predicting Daqu moisture content, its R² values reached 0.9235, 0.9337, and 0.9104 in the upper, middle, and lower layers, respectively, significantly outperforming the LSTM model (by 27.82%, 27.24%, and 22.38%), the LightGBM model (by 20.15%, 24.05%, and 22.43%), and the LSTM-LightGBM model (by 9.15%, 9.77%, and 7.16%). The experimental results indicated that the moisture content during Daqu fermentation could be predicted using environmental parameters, providing solid data support for regulating the Daqu fermentation process.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Application and analysis of a multi-factor feature fusion-based LSTM-Attention-LightGBM model for moisture prediction in Daqu fermentation process

  • 期刊论文

  • 发表日期：

    2026年1月14日

查看全文
  • Haili Yang,
  • Fuhao Han,
  • Sai Liu,
  • Xinjun Hu,
  • Liangliang Xie,
  • Manjiao Chen,
  • Xilong Liao,
  • Lei Fei

摘要

Daqu is a solid-state starter used in the brewing of traditional Chinese liquor. The change in moisture content during Daqu fermentation is a key factor for measuring its quality. However, the traditional method for measuring the moisture content in Daqu is cumbersome and time-consuming, making the rapid detection of Daqu moisture during fermentation difficult. Therefore, this study aimed to propose an LSTM-Attention-LightGBM time-series prediction model to rapidly predict Daqu moisture using the environmental parameters of Daqu fermentation. Combining the characteristics of multi-point environmental parameters, the RF algorithm was used for feature point screening. Also, the Prophet model and expanding window technique were employed to extract features from Daqu moisture and the screened environmental data, respectively, so as to capture their change trends and fluctuations. The comparative analysis revealed that the LSTM-Attention-LightGBM model performed the best in predicting Daqu moisture content, its R² values reached 0.9235, 0.9337, and 0.9104 in the upper, middle, and lower layers, respectively, significantly outperforming the LSTM model (by 27.82%, 27.24%, and 22.38%), the LightGBM model (by 20.15%, 24.05%, and 22.43%), and the LSTM-LightGBM model (by 9.15%, 9.77%, and 7.16%). The experimental results indicated that the moisture content during Daqu fermentation could be predicted using environmental parameters, providing solid data support for regulating the Daqu fermentation process.