提出了基于局部加权偏最小二乘回归算法的污垢预测算法,通过在训练集的污垢数据局部模型内对新测得的数据进行偏最小二乘回归分析,并应用自适应算法对模型参数、各模型之间的加权系数进行自动优化调整。算法能很好地解决新旧数据相互影响问题,以适应冷凝器水质及工况参数的动态变化,具有学习速度快、泛化能力强及鲁棒性强的特点。通过与各种工况下的污垢预测值比较,实验结果说明基于局部加权偏最小二乘回归学习算法的污垢模型预测精度比神经网络模型、渐近污垢模型有显著提高。
The paper proposes a locally weighted partial least squares regression algorithm for the prediction of condenser fouling,which fits the new measurement data by means of partial least squares regression analysis in a number of local models of old measured dada,then applies adaptive algorithm to adjust optimization model parameters and weighted coefficients of the models.The algorithm is a very good solution to the mutual impact of the new and old measurement data,and adapts the condenser water quality changes and device parameter dynamic changes.The algorithm features fast learning speed,strong generalization ability and robustness characteristics.Experimental results show that the prediction accuracy of the locally weighted based partial least-squares regression algorithm is remarkably higher than that of neural network fouling model and asymptotic fouling model.