基于动态模式分解的精馏吸收状态变量非设计条件下的重构与预测

Reconstruction and prediction of distillation or absorption state variables underoff-design conditions based on dynamic mode decomposition

精馏和吸收作为典型的非线性过程,其操作过程中存在大量描述系统特征的状态变量。为了对这些状态变量进行重构和预测,实现精馏吸收过程的实时数字孪生,通过动态模式分解算法(DMD)获取非线性系统的近似线性化模型,用于快速获取精馏吸收过程中各级浓度、流量、温度和持料量等状态变量。在此基础上,应用Kalman滤波器对DMD生成的线性模型进行实时校正,使得在非设计和有限测量条件下,也可以有效地预测吸收或精馏的状态变量,而无须重新训练模型。

Distillation and absorption,as typical nonlinear processes,have a large number of state variables that describe the characteristics of the system during their operation.In order to reconstruct and predict these state variables and realize the real-time digital twin of the distillation and absorption process,this paper obtains an approximate linearized model of the nonlinear system by the dynamic mode decomposition method(DMD),which is used to quickly obtain the state variables such as concentration,flow rate,temperature and holding capacity at each tray of the distillation and absorption process.Based on this,a Kalman filter is applied to correct the linear model generated by DMD in real time,which makes it possible to predict the state variables of absorption or distillation effectively without retraining the model even under off-design and limited measurement conditions.