Study on optimization and real-time monitoring of extraction process parameters of spice essential oil based on deep neural network

Due to its wide application in fields such as food, medicine, and cosmetics, spice essential oils require higher efficiency and stability in extraction processes. The traditional extraction process has problems such as low efficiency and unstable parameter control, while deep neural networks (DNN), especially long short-term memory networks (LSTM), have significant advantages in processing complex process and time series data. In this study, three spices, ginger, garlic and pepper, were taken as the objects, and the key process parameters affecting the extraction of essential oil were determined through experiments, including distillation temperature, time, pressure and material-liquid ratio, and these parameters were modeled and optimized by LSTM model. The results showed that the optimized process parameters significantly improved the output of essential oil: the output of essential oil of ginger increased from 12.5 ml/kg to 15.4 ml/kg (by 23.2%), that of garlic increased from 10.2 ml/kg to 12.0 ml/kg (by 17.6%), and that of pepper increased from 8.7 ml/kg to 10.4 ml/kg (by 19.5%). At the same time, the real-time monitoring system realizes the dynamic adjustment and stable operation of the extraction process through data acquisition, preprocessing, model prediction and feedback control modules. The prediction accuracy of LSTM model on the verification data set reaches 92%, showing good generalization ability. This study shows its great application potential in the field of spice essential oil extraction, which provides theoretical basis and technical support for the development of intelligent extraction technology in the future.