基于记忆模式的NO_x支持向量回归预测研究  

N0_X Prediction Based on Memory Mode Support Vector Regression

作　　者：黄景涛[1] 罗威[1] 任志伟[1] 茅建波[2] 

机构地区：[1]河南科技大学电子信息工程学院,河南洛阳471003 [2]浙江省电力试验研究院,浙江杭州310014

出　　处：《控制工程》2012年第4期704-708,共5页Control Engineering of China

基　　金：国家自然科学基金项目(60904058);河南省教育厅自然科学研究计划(2011A510018)

摘　　要：低NO_x排放是电站锅炉燃烧优化的主要目标之一,影响燃煤锅炉NO_x排放因素众多且复杂,对锅炉燃烧过程NO_x浓度进行准确预测是低NO_x燃烧优化的基础。机组全工况运行时表现出强时变性,静态预测模型难以保证预测精度,考虑到观测样本的时效性,模拟记忆模式对观测数据进行重采样,进而基于支持向量回归算法构建NO_x排放预测模型,构造一种基于记忆模式的支持向量回归算法。以某机组热态试验数据为基础,对算法进行了仿真分析,结果表明,该算法在保证回归建模精度的同时,在训练速度、稳定性以及泛化性能等方面较传统支持向量回归算法更有优势。Low NOx emissions is one of the main objectives of the boiler combustion optimization, the impact factors of NOx emissions in coalired boiler are numerous and complex. Accurate prediction of the NOx concentration during the boiler combustion process is the key basis of low NO~ combustion optimization. The unit shows strong time varying characteristics at full working conditions, and the stat- ic prediction model is difficult to guarantee the prediction accuracy. The timeliness of the observation samples is taken into account. The observation data is resampled by imitating the memory mode, and then the NOx emission prediction model is built using support vector regression （SVR） algorithm, so a memory model based on support vector regression algorithm is constructed. The algorithm is tested on the data sampled from a unit. The analysis results show that the proposed algorithm can not only ensure the regression modelling accuracy, but also has advantages in term of training speed, stability and generalization performance compared to the traditional support vector regression algorithm.

关 键 词：锅炉燃烧 记忆模式 重采样 支持向量回归 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]