A Hybrid Modeling Framework for Soft‐Sensing Prediction of Key Aroma Compounds in Xiaoqu Light‐Aroma Baijiu Fermentation

ABSTRACT Solid‐state fermentation (SSF) in Xiaoqu light‐aroma Baijiu production involves complex and nonlinear interactions among environmental conditions, physicochemical variables, and microbial metabolism, making accurate prediction of flavor evolution highly challenging. To address this challenge, this study proposes a hybrid intelligent soft‐sensing framework that integrates multi‐sensor environmental monitoring, feature screening, and ensemble learning to enable high‐precision prediction of key aroma compounds during Baijiu fermentation. Environmental parameters—including temperature, humidity, pH, and CO 2 concentration—were continuously monitored using IoT‐based sensors, while physicochemical indicators were measured throughout the fermentation process. The concentrations of 11 volatile compounds were determined via gas chromatography–flame ionization detection (GC–FID) as offline reference measurements for model training and validation. Aroma‐active compounds were first screened using odor activity value (OAV) analysis, and principal component analysis (PCA) was subsequently applied to identify dominant variation patterns in aroma profiles. Four key aroma indicators—ethyl acetate, ethyl lactate, isobutanol, and 3‐methyl‐1‐butanol—were identified as representative predictors of aroma quality. Within the proposed framework, a Marine Predators Algorithm‐optimized backpropagation neural network was employed as the base learner, and its integration with an AdaBoost ensemble strategy improved prediction robustness and generalization performance. Comparative experiments showed that the proposed framework achieved lower MAE and RMSE and higher R 2 values than BP–AdaBoost and PSO–BP–AdaBoost models. After model calibration, the developed framework enables near real‐time soft‐sensing prediction of aroma evolution using only multi‐sensor environmental and physicochemical data, providing a data‐driven basis for intelligent monitoring and digital transformation of traditional Baijiu fermentation processes.