基于自适应神经模糊推理系统及随机分形搜索算法的黄酒发酵过程建模与优化

Modeling and optimization of rice wine fermentation process based on ANFIS and stochastic fractal search algorithm

黄酒酿造是多菌种混合发酵,具有产物多样的特点,已有的黄酒发酵过程模型是建立在主要生化反应基础上的发酵动力学模型,模型的精度和泛化能力尚不能满足工业需求。针对黄酒醪液中生成产物多样的特征,该文利用模糊系统的建模策略,将自适应神经模糊推理系统的单维度输出扩展到多维度输出,提出了多输出自适应神经模糊推理系统模型;然后针对该模型参数量大的特点,该文将莱维飞行和层次学习策略融入随机分形搜索算法,提出了层次学习随机分形搜索算法,用于模型参数的辨识与优化。仿真结果表明,该算法提升了模型的精度和泛化能力,实现了不同生产批次黄酒发酵状态的良好预测。

Chinese rice wine making is a mixed fermentation of multiple bacterial strains,which has the characteristics of diverse metabolites.The existed modeling approaches are fermentation kinetic models built upon the majority biochemical reactions in the fermentation process.However,these models cannot meet the industrial needs because they are weak in generality and low in accuracy.According to the diversity of products produced in the rice wine fermentation process,this paper proposes the multi-output adaptive network-based fuzzy inference system(MOANFIS)by using the modeling strategy of fuzzy system.MOANFIS extends the output of adaptive network-based fuzzy inference system(ANFIS)from one dimension to multiple dimensions.Then,in view of the large number of MOANFIS model parameters,this paper integrates Levy flight and level-based learning strategy into the stochastic fractal search(SFS)algorithm,and proposes the level-based learning stochastic fractal search(LLSFS)for the identification and optimization of model parameters.The experimental results showed that the LLSFS improved the accuracy and generalization ability of MOANFIS,rice wine fermentation status of different production batches can be accurately predicted.