基于FIA-ISE联用技术的锅炉水总碱度测量的LLE-SVR建模

LLE-SVR modeling to determine of the alkalinity in boiler water by FIA-ISE Method

工业锅炉水总碱度测量常用指示剂法测定,该方法费时,难于自动化。利用流动注射分析(FIA)技术的快速、准确等优点可以实现对工业锅炉水总碱度的快速测定,将流动注射分析与离子选择性电极联用(FIA-ISE)滴定酸碱溶液可以获取连续变化的电压信号。由于锅炉水中常存在未知物质的干扰,基于局部线性嵌入(Local Linear Embedded,LLE)和偏最小二乘支持向量回归机(Least Square Support Vector Regression Machine,LS-SVR)相结合的电压数据建模方法可以消除未知物质的干扰,同时可以降低电压数据的维数。LLE算法将高维电压信号映射到低维流形空间,实现高维非线性电压数据结构的特征提取,LS-SVR建模方法可以建立总碱度的对数和降维后电压信号之间的非线性回归模型。实际待测数据的计算结果表明:基于LLE-SVR方法获得的锅炉水总碱度回归模型,通过对16组测试样本的分析,得到测量的平均相对误差:NaOH为0.83%,Na_2CO_3为0.66%;对于LS-SVR回归建模,NaOH为0.98%,Na_2CO_3为0.86%。LLE-SVR的运算负担只有LS-SVR的3.3%且测量精度要高于常规的峰面积和总碱度之间的线性拟合方法。

The alkalinity of industrial boiler water has mainly been determined by using acid base titration, based on indicator method. It is time- consuming and not suitable for automation. Flow injection analysis （FIA）method can determine the alkalinity rapidly and accurately. The F1A, coupled with an ion selective electrode （ISE） to determine the alkalinity of the solution, can get the voltage continuously. The unknown constituents often coexist in boiler water. A local linear embedded （LLE）, coupled with least square support vector regression （LS-SVR） modeling method, is proposed for great voltage data to reduce interference from unknown substances and to decrease the to-be-modeled data. LLE is used to reduce the dimensions of voltage and the low-dimensional features are extracted. A nonlinear regression model is developed by LS-SVR method between the voltage and the alkalinity from the boiler water. Simulation results show the relative error of the prediction relative error of NaOH and Na2CO3 from LLE-SVR model was 0.83 and 0.48 percent respectively. Prediction relative error from LS-SVR model was 0.98 and 0.86 percent respectively. Computation load of the LLE-SVR was 3.3 percent of the SVR. Compared with the linear fitting between the peak area and the alkalinity, higher accuracy can be got from the LLE-SVR.