Rapid temperature field prediction in natural product extraction

Abstract

Extraction process is critical in natural product manufacturing; precise temperature control during this process directly determines final product quality. However, existing technologies face two main challenges: first, temperature monitoring relies solely on discrete point measurements, precluding a comprehensive representation of the temperature distribution; second, real-time temperature prediction demands excessive computational resources, resulting in delayed control responses. To address these challenges, this study utilizes distributed temperature sensing (DTS) technology to detect the temperature field during the extraction process. Based on the detected temperature field data, a hybrid model approach (HMA) was developed by integrating a physical mechanism model with a residual neural network. The model enables rapid prediction of the temperature field. Experimental results demonstrate that the model's root mean square error (RMSE) ranges from 0.0971 to 0.5885, with 90.2 % of the temperature deviations confined within ±2 °C. These findings substantiate the model's effectiveness and accuracy in temperature prediction.