摘要
建立了协定法麦汁浊度、糖化力、黏度以及浸出率的BP神经网络预测模型,希望通过此模型能够预测在不同设定工艺变量条件下协定麦汁的主要理化指标。选取8组数据进行BP神经网络的训练仿真,并用2组未参加训练的数据进行验证。在均方差为0.001的条件下,网络于242次训练后收敛,模型训练的最大相对误差为2.58%,预测值的最大相对误差为10.08%,表明该模型具有良好的预测和仿真能力。
The BP Neutral Network predictive model of Congress wort turbidity, diastatic power, viscosity and extraction rate was developed. The model was expected to predict the main physical and chemical characteristics of Congress wort at the settings of different technical parameters. Eight groups of data were selected to train the BP Neutral Network and two groups of data to test the model. The Mean Square Error of Neutral Network dropped to 0.001 after 242 training epochs. The maximum relative error of simulated value and predictive value to the measured value were 2.58 % and 10.08 % respectively, indicating that the proposed model was in possession of good suitability of simulation and prediction.
出处
《酿酒科技》
北大核心
2008年第1期54-56,共3页
Liquor-Making Science & Technology
基金
国家"十一五"重点支持项目
关键词
啤酒酿造
制麦
理化指标
BP神经网络
beer brewing
wort preparation
analytic specification
BP neutral network
