基于深度特征融合的米粉原料指标含量预测问题研究

Study on the prediction of rice noodle raw material index content by deep feature fusion

米粉是中国南方地区的一种特色小吃,随着粮食行业的发展和生活水平的提高,选择合适的原料来生产优质的米粉是目前有待解决的问题之一。为此,在满足米粉各种特性的前提下,提出了一种深度特征融合技术,结合机器学习算法来实现米粉制品指标对原料指标的逆向预测。深度特征融合技术通过对米粉制品指标进行多层的加权特征融合,使用机器学习预测模型对原料指标预测,并使用粒子群算法对特征融合的权值与预测模型结构进行内外嵌套优化,达到提高预测精度的目的。实验表明,经优化后单项指标最高的决定系数(R2)可达0.987,单项指标最低的均方根误差(RMSE)可达0.030 2,其中水分、淀粉含量、蛋白质含量、膨润力、糊化温度等原料指标值的预测值与真实值之间的误差较小,具有很好的预测效果,可为生产优质米粉提供很好的原料选择参考。

Rice noodle is a special snack in southern China.With the development of the grain industry and the improvement of living standards,choosing the right raw materials to produce high-quality rice noodles is one of the problems to be solved at present.Therefore,on the premise of satisfying various characteristics of rice noodles,this paper proposes a deep feature fusion technology,combined with machine learning algorithm to achieve the backward prediction of rice noodles raw material index content.Deep feature fusion technology can improve the prediction accuracy by multi-layer weighted feature fusion of rice noodles product index,using machine learning prediction model to predict raw material index content,using PSO to optimize the weights of feature fusion and prediction model structure synchronously.Experimental results show that the highest R 2 of the single index of the prediction result can reach 0.987,and the RMSE of single index only reach 0.0302.The error between the predicted value and the real value of the indexes of water content,starch content,protein content,swelling force and gelatinization temperature is small,which has a good prediction effect,and can provide a good reference for the selection of raw materials for the production of high-quality rice noodles.