Ultrasonic-Assisted Ethanol Extraction of Flavonoids from Supercritical CO2 Extracted Propolis Residues and Its Online UV-Vis Monitoring

Secondary extraction of flavonoids from propolis residue following supercritical CO₂ extraction was investigated, with concurrent development of an on-line and real-time flavonoid quantification monitoring method. Ultrasonic-assisted ethanol extraction (UAE) parameters were optimized by single-factor experiments and response surface methodology (RSM), with comparative assessment against traditional extraction without ultrasound assistance. Under optimal UAE conditions, an offline quantitative detection model for the content of flavonoids was established firstly by ultraviolet-visible (UV-Vis) spectroscopy coupled with machine learning algorithms. Secondly, an online detection model was obtained by calibrating the offline model with direct standardization (DS) algorithm. The optimal extraction conditions were determined to be: ultrasonic frequency of 28 kHz, power density of 136 W/L, extraction time of 15 min, temperature of 43°C, solid-liquid ratio of 1:10 (g/mL), and ethanol concentration of 75%. Under these optimized parameters, the extraction yield reached 9.41 mg/mL. Comparative analysis revealed that the ultrasonic method significantly outperformed both conventional extraction techniques, with low-speed agitation yielding 7.52 mg/mL and high-speed agitation yielding 8.26 mg/mL (p < 0.05). offline models, support vector regression (SVR) with normalized spectra and synergy interval partial least squares (Si-PLS) feature selection demonstrated superior performance (calibration: Rc=0.9914, RMSEC=0.3224; prediction: Rp=0.9926, RMSEP=0.3003)The statement "superior performance" refers specifically to the comparative results within this evaluation framework. The model built using Support Vector Regression (SVR) on normalized spectra with Synergy Interval Partial Least Squares (Si-PLS) for feature selection demonstrated superior predictive accuracy and robustness compared to the other modeling methods evaluated in the study, primarily Partial Least Squares (PLS) and Backpropagation Neural Network (BPNN). The DS-corrected online migration model achieved optimal real-time quantification (Rp=0.9660, RMSEP=0.7773). This study confirms UAE as an efficient low-temperature method for flavonoid recovery from propolis residue. Integrationof offline UV-Vis modeling with DS spectral correction enables robust online flavonoid monitoring during extraction.