关键词: 递推PLS, 模型, 性能评估, 滑动平均滤波器, 化学反应过程, 模拟

Abstract: In order to reduce the model updating frequency of recursive partial least squares (RPLS) modeling methods, a RPLS model based on the model performance assessment (MPA-RPLS) is developed. Firstly, a confidence limit of the model is generated automatically based on the initial behavior of a process. And a root mean squared error of prediction (RMSEP) is used as a performance index to evaluate the model. Base on the results of the model performance assessment, the model updating is selectively activated, in the meanwhile, the confidence limit is also updated. Subsequently, a moving average filter is integrated into the model to eliminate the noise embbeded in variables, and the effect of the noise on the model performance is then investigated. At last, the developed model is applied to a chemical reaction process, hydro-isomerization process of C₈-aromatics. Simulation is run based on a large number of industrial data. The simulation results show that the computational efficiency is improved greatly (model updating frequency is reduced greatly) by the developed model, while a minor loss of the prediction accuracy is found. The noise embedded in variables could be dealt with effectively by the moving average filter, hence the prediction accuracy is improved.

Key words: recursive partial least squares, model, performance assessment, moving average filter, chemical reaction process, simulation