This work addresses the hybrid modeling problem of key variables in chemical production processes. Considering the difficulty in uniformly expressing and utilizing empirical knowledge from different representation forms in hybrid modeling to enhance model reliability and accuracy, the authors propose a method named “Dictionary Learning-Based Knowledge Fusion ESN (Echo State Network) Hybrid Modeling” for variables prediction. The proposed method utilizes a dictionary matrix capable of characterizing the deep features of the data as its foundation, employs different dictionary matrices to represent various operating conditions, and designs a new data operating condition classification algorithm based on this framework. Furthermore, the authors consider empirical knowledge as constraints in modeling. These constraints are then applied to the transformation process of the dictionary D, elevating the dimensionality of the dictionary to construct a reservoir matrix W_res that incorporates diverse knowledge. The constructed W_res was applied in the training process of the ESN, ensuring that the ESN training process is guided by knowledge and data information. The proposed method is validated on a simulated continuous stirred tank reactor and real-world data from a hydrogenation distillation unit. The results demonstrate that the proposed method achieves stable predictions, avoids abnormal prediction values, and outperforms other comparative methods on the real system, improving prediction accuracy. The results indicate that this method has significant potential for practical applications.

中文翻译：

---

基于字典学习的知识融合 ESN 混合建模：一种整合实证知识的化学关键变量建模方法

这项工作解决了化工生产过程中关键变量的混合建模问题。考虑到混合建模中难以统一表达和利用来自不同表示形式的经验知识以提高模型的可靠性和准确性，作者提出了一种名为“基于字典学习的知识融合 ESN（Echo State Network）混合建模”的变量预测方法。所提出的方法以能够表征数据深层特征的字典矩阵为基础，采用不同的字典矩阵来表示各种运行条件，并基于该框架设计了一种新的数据运行条件分类算法。此外，作者将经验知识视为建模的约束。然后将这些约束应用于字典 D 的转换过程，提升字典的维数，以构建一个包含不同知识的储层矩阵 W_res。构建的 W_res 应用于 ESN 的训练过程，确保 ESN 训练过程以知识和数据信息为指导。所提出的方法在模拟的连续搅拌罐式反应器和来自加氢蒸馏装置的真实数据上进行了验证。结果表明，所提方法实现了稳定的预测，避免了异常的预测值，在真实系统上优于其他比较方法，提高了预测精度。结果表明，该方法具有巨大的实际应用潜力。