论文部分内容阅读
对复杂多变的热工对象建模是实现良好控制性能的难点,为此提出运用Kriging估计方法建立对象的自适应模型。该法是非参数回归的建模方法,无需确定模型结构和训练,就能实现对未知函数的无偏最优估计。通过对样本空间的实时调整还实现了一种自适应的Kriging模型。选取电站锅炉NOx排放作为建模对象,运用现场试验数据,比较了自适应Kriging模型和神经网络模型对NOx排放的内插和外推预测性能。5组结果显示,神经网络模型的平均预测误差为11.59%,而Kriging模型仅为3.49%。
Modeling complex and ever-changing thermal objects is difficult to achieve good control performance. Therefore, an adaptive model for establishing objects using Kriging method is proposed. The method is a non-parametric regression modeling method, without the need to determine the model structure and training, we can achieve unbiased optimal estimation of unknown function. Through the real-time adjustment of the sample space also achieved an adaptive Kriging model. NOx emissions from power plant boiler are selected as modeling objects. The field test data are used to compare the predictive performance of NOx emission by interpolation and extrapolation of adaptive Kriging model and neural network model. The results of the 5 groups show that the average prediction error of the neural network model is 11.59%, while the Kriging model is only 3.49%.