Abstract:

The production process of corn deep processing to produce fructose has problems of outdated control and lack of refinement in production and processing. However, due to the complexity of the process, it is difficult to establish and optimize mechanism based models. Big data technology offers an effective solution by utilizing a substantial volume of production data to uncover process insights and identify key points. Initially, crucial target variables of this process were selected. Using big data technology, the original production data underwent preprocessing steps such as handling missing values, addressing outliers, noise reduction, and dimensionality reduction. Subsequently, three machine learning models—random forest (RF), extreme gradient boosting (XGBoost), and artificial neural network (ANN) were constructed, all achieving R² values exceeding 0.90. Lastly, the SHAP method is used to explain different machine learning models, validate the credibility of the models, obtain the contribution levels of different features to the prediction results, and integrate the results of explanations from different models. This process generates a ranking of the importance of different points in the production process. Combining this with production experience, a mechanistic analysis is conducted to obtain the final key point table.

Key words: data mining, food processing, systems engineering, explainability, neural network

摘要：

玉米深加工制果糖生产流程存在着控制落后，生产加工缺乏精细化的问题，然而由于其流程复杂，基于机理的模型难以建立与优化。大数据技术提供了有效解决方案，通过大量的生产数据来挖掘流程中知识、筛选出关键位点。首先，选取了该工艺流程的关键目标变量并利用大数据技术对生产流程中的原始数据进行缺失值处理、异常值处理、降噪和降维等预处理，然后构建了随机森林（RF）、极端梯度提升树（XGBoost）和人工神经网络（ANN）三种机器学习模型，其模型的R²均达到0.90以上，最后利用SHAP方法对不同的机器学习模型进行解释，验证模型的可信性，得到不同模型的特征对预测结果的贡献程度，并综合不同模型解释的结果，得到生产流程中不同位点的重要程度排序，结合生产经验进行机理分析，得到最终的关键位点表。

关键词: 数据挖掘, 食品加工, 系统工程, 可解释性, 神经网络