基于支持向量机和相空间重构的多相催化建模

Heterogeneous catalysis modeling strategies integrating support vector machine and phase space reconstruction

为减少发现新的多相催化剂的时间和降低消耗,加速二甲醚合成工业化进程,提出一种基于支持向量机和相空间重构建立多相催化剂的成分模型和催化反应机理模型的新方法。支持向量回归作为一种新的机器学习算法,被用于多相催化动态模型的开发,SVM模型以催化剂的组分和制备条件作为输入数据,催化剂的性能指标(催化剂的选择性和转化率)作为输出数据。模型的输入空间使用相空间重构理论将一维时间序列映射到多维空间,提高了支持向量回归机的预测精度。支持向量机-相空间重构是一种多相催化建模的新策略,主要优势是在催化反应机理未知或难以获取的情况下,建模完全由历史进程的少量样本空间完成,避免了传统催化剂研发过程中"试错实验"的盲目性和偶然性。实验验证表明,本文提出的方法丰富了实验数据,预测的催化剂性能与实验获得的数据有很好的一致性,是一种有效的多相催化建模工具。

A new heterogeneous catalysis modeling methodology, namely support vector machine (SVM) and phase space reconstruction (PSR) was presented, for catalyst compositional models and catalytic reaction mechanism models, for reducing both high temporal costs and financial costs, and accelerating the process of industrialization synthesis of dimethyl ether (DME). In the SVM-PSR approach, a novel machine learning algorithms, namely support vector regression, was utilized for developing catalytic kinetic models. A SVM model was constructed for correlating process data comprising input variables of catalyst compositional, operating and output variables of performance of catalyst (catalyst selectivity and conversion). The input space of SVM model was alternated by phase space reconstruction (PSR), and one-dimensional time sequence was mapped into hyperspace for improving the forecasting precision. Finally, this algorithm was verified experimentally to be feasible. As a new strategy for heterogeneous catalysis modeling and optimization, the major advantage of the SVM-PSR is that modeling and optimization can be conducted exclusively from the historic small sample space data wherein the detailed knowledge of process phenomenology (reaction mechanism, rate constants, etc.) is not required and difficult to get. Another advantage is that it has avoided the blindness and contingency of the traditional catalyst 'trial and error' method. The predicted performance numbers of catalyst is not only rich of experimental data, also agreed well with the experimentally obtained data. All of these results clearly demonstrate that SVM-PSR strategy is an effective tool in the heterogeneous catalysis modeling.