针对目前国内电站锅炉燃烧建模存在煤质与负荷波动频繁、测点精度有限、设备运行组合多变等产生的问题,提出了电站锅炉燃烧的非对称神经网络建模方法.将锅炉模型的输入按照实际物理规律的关联关系设计网络结构组合,去掉关联性较弱的联系,从而使模型天然体现一定的锅炉燃烧规律,实现了不同燃烧器出力分配下的单一网络建模,提高了学习训练效率,并大幅降低了模型所需样本数量.将经典对称神经网络模型和非对称神经网络模型进行对比训练,结果表明本文提出的非对称神经网络建模方法检验正确率高.将其应用于某超临界660 MW机组的燃烧优化控制,锅炉效率平均可提高0.25％.%To solve the problems occurred during the combustion modeling of domestic power plants, such as frequent fluctuations of coal quality and unit load, limited measurement accuracy, various change in running equipments combination, an asymmetric neural network modeling method for utility boilers' combustion was proposed. In this method, the boiler model's structure is designed according to the actual physical relationship by removing the weak link. So the model naturally embodies the boilers' combustion laws and realizes single network modeling under conditions with different burner output distributions. Thus, this method greatly increases the training efficiency and sharply reduces the demand for the number of samples. Moreover, comparative training was carried out for classical symmetric neural network model and asymmetric neural network model. The results show that, the proposed method is more effective than the classical one. It was used to optimize the combustion control of an actual 660 MW unit, and the results show the boiler efficiency was increased by 0.25％.
展开▼
