基于KF与ESN的糖厂澄清工段在线预测

糖厂澄清工段是甘蔗制糖的重要工艺环节之一,是一个复杂的物理、化学过程,具有非线性、大时滞、时变等特点.而且不同的榨季、甘蔗的品种、新技术的应用等情况,都可能导致过去良好的控制模型往往不能及时适应新情况的发生.基于大量离线、在线数据,结合回声状态网络(ESN)和Kalman滤波(KF)的特点,设计了应用于糖厂澄清工段的在线自适应预测方法.该方法将Kalman滤波应用于ESN的高维状态空间中,可以直接对网络的输出权值进行更新.将仿真结果与基于EKF的RBF网络相比较,说明了基于KF与ESN的糖厂澄清工段在线预测模型的优越性.

Online Prediction for Clarifying Process in Sugar Mill Based on KF and ESN

As a very important part,the clarification process in the cane sugar manufacturing is a complex physical and chemical process,with the characteristics of nonlinear,large time delay and time-varying.Under the conditions of different press quarter,cane sugar varieties,new technologies,and so on,the past good control model can not timely adjust to the new situation.According to a large number of offline and online data,a method of online adaptive prediction based on echo state network(ESN) and Kalman filter(KF) algorithm in the clarification process of the cane sugar manufacturing is designed.The KF is adopted to the high-dimension state space to directly update the output weights of the ESN.Compared with RBF network based on EKF,the simulation results illustrate the superiority of the algorithm based on KF and ESN applied to the clarification process of cane sugar manufacturing.