Modeling Of Rice Wine Fermentation Process Based On Fuzzy System

Rice wine,a unique kind of wine in China,is one of the three ancient wines in the world,along with beer and wine.In recent years,with the continuous improvement of people’s living standards,the consumption market of rice wine has presented a broad prospect of development.The rice wine fermentation process is a nonlinear,multi-input multi-output,complex biochemical reaction process.However,the key biochemical parameters cannot be measured online in real-time,resulting in the automation level of rice wine being far lower than other industrial production processes.The modeling of fermentation process is the basis for optimizing and controlling rice wine production.Among various modeling methods,fuzzy system provides a reasonable framework for modeling by decomposing nonlinear systems into a set of local linear models.In this paper,we predict the state of rice wine fermentation process based on the fuzzy system modeling method,so as to ensure the controllability and stability of rice wine quality.The research contents of this paper are as follows:(1)Aiming at the characteristics of product diversity in the fermentation process of rice wine,this paper proposes a model called Multi-Output Adaptive Network-based Fuzzy Inference System(MOANFIS).MOANFIS extends the output of Adaptive Network-based Fuzzy Inference System(ANFIS)from one dimension to multiple dimensions,namely the output of each fuzzy rule is represented by a multi-dimensional vector instead of a scalar value.The posterior parameters of fuzzy rules in MOANFIS share a common ancestor parameter.To identify and optimize the model parameters of MOANFIS,this paper proposes a Level-Based Learning Stochastic Fractal Search(LLSFS)based on the Stochastic Fractal Search algorithm(SFS).Firstly,the Level-Based Learning strategy is used to divide the particle swarm in SFS,and the particles in low level learn from the particles in high level to update themselves.Secondly,the Levy-flight search method is used to avoid the particles falling into the local optimal solution.LLSFS has a strong global search ability when dealing with high-dimensional optimization problems,and it ensures the accuracy of MOANFIS model.(2)This paper proposes a Hierarchical ANFIS(Hierarchical ANFIS-ELM,H-ANFISELM).As the dimension of input data increases,MOANFIS faces the problem of fuzzy rule explosion,which makes the model complex and reduces its interpretability.To address this issue,this paper proposes two Dropout methods to reduce the number of fuzzy rules by discarding nodes with small weights,thus keeping the number of fuzzy rules within a reasonable range.Then,the Extreme Learning Machine(ELM)algorithm is used to expand the output dimension of ANFIS and solve the model parameters,forming the ANFIS-ELM module.Based on the ANFIS-ELM module,a two-layer H-ANFIS-ELM model is proposed,where the output of the first layer is used as the augmented input feature of the second layer after feature fusion,expanding the input information of the second layer.To reduce the influence of randomly initialized parameters in ELM on the model,H-ANFIS-ELM updates the posterior parameters of fuzzy rules during iterative cycles.The model achieves a good balance between interpretability,robustness,and accuracy.(3)This paper designs and develops a Web-based monitoring system for the rice wine pre-fermentation process.This system mainly has three functions.First,it realizes the information management of workshop staff.Second,it collects data from fermentation tanks and displays temperature curve of rice wine fermentation,enabling the monitoring of the fermentation process.Third,based on the models established in the previous two chapters,it predicts and alerts the fermentation status,providing data support for the optimization and control of the yellow rice wine fermentation process.