一种软计算混合策略在多相催化剂建模与预测中的应用

为了减少发现新的碳一多相催化剂的时间、降低消耗,加速二甲醚合成工业化进程,提出一种新的软计算混合策略并应用于碳一多相催化剂建模与预测研究。支持向量回归机(Support Vector Regression,SVR)作为一种新的机器学习算法,能较好地解决小样本、高维、非线性和局部极小点等实际问题,在混合策略中被用于多相催化剂组分模型的开发。SVR模型的超参数选择采用启发式全局优化搜索算法——自适应混沌粒子群算法来提高SVR模型的预测精度和泛化能力。新策略的主要优势是在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。通过对两种不同建模方法、三种不同SVR超参数优化策略在Cu-Zn-Al-Zr合成二甲醚催化剂组分模型中的对比研究得出,新策略在多相催化剂建模与预测上是一个有前途的发展方向。

Modeling and Forecasting for Heterogeneous Catalysts Using Soft Computing Hybrid Strategy

This paper presents a new soft computing hybrid strategy of C1 heterogeneous catalyst modeling and forecasting methodology,for reducing both high temporal costs and financial costs,and accelerating the industrialization process of dimethyl ether(DME) synthesis.In this soft computing approach,a novel machine learning algorithm,namely support vector regression,can well solve the small sample,high dimension,nonlinear and local minimizing of practical problems,and has been utilized for developing catalytic kinetic models.The hyper-parameter selection of SVR model adopts heuristic global optimization algorithm,that is,chaotic particle swarm optimization algorithm(CPSO),to improve the SVR model in prediction precision and generalization ability.The main advantage of the new strategy is its ability to,for unknown or difficultly available reaction mechanism,model by a small history sample space,so as to avoid the blindness and chance in traditional trial-and-error catalyst research and development process.Through two different modeling methods and three different SVR super parameter optimization strategies in the comparative study on Cu-Zn-Al-Zr dimethyl ether catalyst component,the new strategy was found to be a promising development direction in the heterogeneous catalysts modeling and forecasting.