基于神经网络内部模型的非线性偏最小二乘法用于火电厂烟气光谱定量分析

针对火电厂烟气光谱数据的非线性特性，采用了基于神经网络内部模型的非线性偏最小二乘定量分析方法。该方法进行偏最小二乘（PLS）回归后，将自变量和因变量的隐变量作为神经网络的输入和输出进行训练，即可得到非线性内部模型。将PLS、基于向后传递神经网络内部模型的非线性PLS（BP-NPLS）、基于径向基函数神经网络内部模型的非线性PLS（RBF-NPLS）和基于自适应模糊推理系统内部模型的非线性PLS（ANFIS-NPLS）对火电厂烟气多组分进行测定后比较，BP-NPLS、RBF-NPLS和ANFIS-NPLS较之PLS，将二氧化硫预测模型的预测均方根误差（RMSEP）分别降低了16.96%，16.60%和19.55%;将一氧化氮预测模型的RMSEP分别降低了8.60%，8.47%和10.09%;将二氧化氮预测模型的RMSEP分别降低了2.11%，3.91%和3.97%。实验表明，非线性PLS较PLS更适用于火电厂烟气定量分析。通过神经网络对非线性函数的高度逼近特性，基于本文所提及内部模型的非线性偏最小二乘方法有较好的预测能力和稳健性，在一定程度上解决了基于多项式和样条函数等其他内部模型的非线性偏最小二乘方法的自身局限性。其中，ANFIS-NPLS的效果最好，自适应模糊推理系统的学习能力能够有效降低残差，使模型具有较好的泛化性，是一种比较准确实用的火电厂烟气定量分析方法。

Spectral Quantitative Analysis by Nonlinear Partial Least Squares Based on Neural Network Internal Model for Flue Gas of Thermal Power Plant

To deal with nonlinear characteristics of spectra data for the thermal power plant flue, a nonlinear partial least square (PLS) analysis method with internal model based on neural network is adopted in the paper. The latent variables of the independent variables and the dependent variables are extracted by PLS regression firstly, and then they are used as the inputs and outputs of neural network respectively to build the nonlinear internal model by train process. For spectra data of flue gases of the thermal power plant, PLS, the nonlinear PLS with the internal model of back propagation neural network (BP-NPLS), the nonlinear PLS with the internal model of radial basis function neural network (RBF-NPLS) and the nonlinear PLS with the internal model of adaptive fuzzy inference system (ANFIS-NPLS) are compared. The root mean square error of prediction (RMSEP) of sulfur dioxide of BP-NPLS, RBF-NPLS and ANFIS-NPLS are reduced by 16.96%, 16.60% and 19.55% than that of PLS, respectively. The RMSEP of nitric oxide of BP-NPLS, RBF-NPLS and ANFIS-NPLS are reduced by 8.60%, 8.47% and 10.09% than that of PLS, respectively. The RMSEP of nitrogen dioxide of BP-NPLS, RBF-NPLS and ANFIS-NPLS are reduced by 2.11%, 3.91% and 3.97% than that of PLS, respectively. Experimental results show that the nonlinear PLS is more suitable for the quantitative analysis of glue gas than PLS. Moreover, by using neural network function which can realize high approximation of nonlinear characteristics, the nonlinear partial least squares method with internal model mentioned in this paper have well predictive capabilities and robustness，and could deal with the limitations of nonlinear partial least squares method with other internal model such as polynomial and spline functions themselves under a certain extent. ANFIS-NPLS has the best performance with the internal model of adaptive fuzzy inference system having ability to learn more and reduce the residuals effectively. Hence, ANFIS-NPLS is an accurate and useful quantitative thermal power plant flue gas analysis method.