摘要
本文应用人工神经网络模拟了棉粕的挤压膨化脱酚工艺,建立了一个3层网络结构的BP神经网络模型用以预测游离棉酚的降解规律,采用十折交叉验证表明:选择隐藏层神经元数为8、网络训练函数为"traingdx",此网络参数条件下,网络预测准确度高,网络预测输出与实验结果的相关系数(R2)为0.9941、均方根误差为0.4971。基于神经网络模型利用遗传算法进行全局寻优的结果表明,棉粕挤压膨化脱酚的最佳工艺条件为膨化温度131℃、物料水分51%、螺杆转速158r/min、喂料速度136kg/h,在此条件下,游离棉酚的实际降解率为90.50%,与遗传算法优化预测结果的平均相对误差为1.38%,平均相对误差较小。本研究表明,神经网络模拟结合遗传算法对棉粕挤压膨化脱酚工艺具有较好的优化效果。
The artificial neural network（ANN）was used for the simulation of the degradation of free gossypol in cottonseed meal by the extrusion process.A three-layer back propagation neural network was optimized to predict the degradation of free gossypoI.The result of 10-fold cross validation showed that the model of back propagation neural network giving the smallest mean square error（ MSE） was the ANN with the training function as traingdx at hidden layer with 8 neurons.And ANN predicted results were very close to the experimental results with correlation coefficient（ R2）of 0.9941 and RMSE of 0.4971.A genetic algorithm （GA）based on an established neural network model was also used to optimizing de-gossypol process.The results of GA obtained showed that the optimal condition of de-gossypol by the extrusion process was temperature 131℃, water ratio 51%, rotational speed 158r/min,and feeding speed 136kg/h, and in this condition the degradation rate of free gossypol was 90.50%, which was close to the result of GA predicted with the small average relative error of 1.38%. These results suggested that the GA based on a neural network model might be an excellent tool for optimizing cottonseed meal de-gossypol process.
出处
《食品工业科技》
CAS
CSCD
北大核心
2015年第13期243-246,共4页
Science and Technology of Food Industry
基金
新疆农垦科学院引导项目(60YYD201308)
关键词
棉粕
挤压膨化
脱酚
神经网络
遗传算法
cottonseed meal
extrusion
de- gossypol
artificial neural network
genetic algorithm
