【摘要】 针对基于批量数据的传统偏最小二乘(PLS)模型无法随生产过程的变化而更新的问题,提出基于块式递推PLS的限定记忆法.新的自适应算法保持定长数据块队列,根据最新采集的数据块更新模型,并按数据块的时间先后运用遗忘因子赋予其相应的可信度,从而确保模型跟踪过程变化.结合实际工业过程的应用要求,给出了方法的实施步骤,并运用该方法建立了气相聚乙烯工业生产过程质量指标的自适应软测量模型.与固定PLS模型相比,该模型能更好地跟踪过程变化,具有更高的预测精度.更多还原

【Abstract】 Batch-wise data based traditional partial least squares (PLS) model cannot be updated with the process changes. By combining block-wise recursive PLS with finite memory method, a new adaptive algorithm was proposed to build adaptive soft-sensor. To adapt to the process changes in time, the method holds a predetermined length of the modelling data block queue, incorporates a forgetting factor in the queue and updates the model in terms of both the new data and the old model. According to the demands of practical industrial application, the procedure for implementing the algorithm was presented, and the foregoing scheme was applied to construct an adaptive soft-sensor for predicting the quality index of an industrial fluidized bed reactor. The results show that the adaptive model has stronger tracking ability and higher precision than the traditional PLS model and its effectiveness is verified.更多还原