汽提塔塔底氨含量的软测量模型

Soft-sensor model for NH_3 concentration at stripper bottom

针对常规偏最小二乘算法(PLS)难以适用实际工业过程的时变特性、固定长度窗口PLS可能导致数据样本丢失信息问题,提出了基于变滑动时间窗口长度的PLS软测量建模算法。算法思想是设定时间窗口内样本数据长度不是固定的,建模的样本数据随新样本数据加入而逐步舍去。算法应用于煤气化废水处理流程中加压汽提塔塔底氨含量的工业软测量建模,仿真研究表明:相对于简单的PLS方法和固定滑动时间窗口长度的PLS方法,模型的预测平均相对误差分别减少约43%及13%,模型具有精度高、外推能力好等优点。另外,针对案例,讨论了最佳的窗口可变长度。

Soft-sensor model based on traditional partial least squares（PLS） method is difficult to be updated with the changes of industrial process,and moving windows PLS with fixed length are also subject to missing information in sample data.A modeling method based on the moving windows PLS with varied length is proposed in this paper.In the proposed algorithm,the length of the sample data in time window is not fixed,and the old sample data are discarded when new sample data are added.Application in predicting the NH3 concentration in the pressurized stripper for coal-gasification wastewater treatment demonstrates the attractiveness of the method.The results show that the model has high precision and strong extrapolation ability.In comparison with the simple PLS and the moving windows PLS with fixed length,the mean relative errors for prediction using new method are reduced by approximately 43% and 13% respectively.In addition,the optimized window length is discussed in this paper.