结构分离和参数分离的灰箱建模

为解决复杂系统的建模问题,以石化工业中广泛应用的连续搅拌釜式反应器(CSTR)为背景,提出以结构分离和参数分离为基础的灰箱建模方法.建模的思路是充分利用系统的先验知识,最大可能的保留系统的已知结构和参数,以保证建立的模型从内部到外部逼近原型系统.在对非线性系统进行结构分解后,建模步骤分两步进行:对反应速率中的Arrhenius方程进行对数变换,可使用便捷、有效的最小二乘算法拟合其中的未知参数；仅对线性部分的未知参数进行辨识,不仅大大降低模型的训练时间,更是提高了模型的灰箱性,从而保证模型的可靠性和泛化能力.通过与常规使用的全参数辨识的建模方法对比,证明了该方法的有效性.

Hybrid modeling based on structural separation and parameter separation

For the purpose of modeling complex systems,a new hybrid modeling method based on structural separation and parameter separation was presented in consideration of the continuous-flow stirred tank reactor(CSTR) which is widely used in the petrochemical industry.The concept behind this method is making full use of the system’s prior knowledge,and maintaining the known structure and parameter of the system to the utmost so that the hybrid model can approximate the prototype system from the inside to the outside.After the structure of the nonlinear system was decomposed,the modeling was divided into two steps.First,the Arrhenius function in a reaction rate equation was converted into a linear function through a logarithm transform,and then the unknown parameters could be fitted by the least square method,which is convenient and effective.Here only the unknown parameters in the linear part were identified;in this way not only the training time can be shortened,but also the reliability and generalization of the models can be guaranteed by enhancing the model’s grey-box-property.The new methods were compared with the conventional all-parameters identification approach in this paper;the results show the effectiveness and feasibility of the presented methods.