Research On Modeling And Optimization Of Luzhou-Flavor Liquor Processing

Liquor is a traditional drink in China,which involves a high output in the food industry.In-fected by the long fermentation period,it's always time-expensive to take experiment-based pa-rameter optimizations.Even the costly optimized parameters can be affected by fluctuation in fermentation conditions,environmental microorganisms,material,and daqu characteristics,mak-ing the yield and taste declining in process.To maintain the production stability,it's necessary to model the process and design an optimization scheme of process parameters further.Firstly,a continuous optimization scheme is designed to promote the craft parameters of Luzhou-flavour liquor based on prediction models.It consists of a production predicting module that can learn the time-varying trend incrementally and an optimization module that can solve the optimal parameters under the given material characteristics.The prediction module can be further decom-posed into a fermentation model and a distillation model to handle related predictions separately.And the optimization module can be divided into the parameters optimization part and the material characteristics back-stepping part.Secondly,a fermentation model is built based on related craft,principles,and modeling expe-rience.The kinetic part of this model considers the influences of the substrate,product,and oxygen concentration,plus temperature and microbe aging,and fits properly on the traditional strong-flavor fermentation sample.Then a statistical regression part is combined to improve predicting accuracy based on gradient boosting principle and incremental learning pattern.With the custom validation construction rules,the impact of the time-varying trend can be traced and the potential knowledge in processing data can be discovered continuously.Thirdly,a distillation model is built based on the related craft which can be abstracted as a special packed column distillation procedure.The model is mainly constructed by the equilibrium stage assumption,hydrodynamic equation,evaporation efficiency hypothesis,and the UNIFAC activity coefficient method.This mechanism-based non-stable model fits properly on the tempera-ture distribution samples measured by self-made equipment and the liquor concentration samples.Additionally,it is possible to predict the final productions and vapor height in distillation correctly.Finally,a model-based optimization method is designed to optimize the craft parameters in lieu of an experiment-based one,reducing the most of overhead in experiments and approximating the parameter points towards the optimal points as much as possible.Aiming to conjecture the drift of material characteristics that cannot be measured by the automatic control system,a material characteristics back-stepping algorithm is designed referring to the concept of soft-sensor,which can trigger a manual measurement request and a re-optimization procedure with precise parameters.