针对火电机组设备工况复杂、报警系统效率低下及“报警泛滥”现象频发等问题，提出了基于AHCGP混合模型的火电机组报警数据过滤方法，消除冗余性报警。首先，采用近邻传播算法(AP)结合类间类内划分指标(BWP)确定最佳聚类数目，再使用凝聚式层次聚类算法(AHC)进行聚类，区分各类复杂工况。其次，利用高斯过程模型(GP)结合后验报警概率估计值实现机组冗余性报警数据的准确过滤。最后，采用某电厂1 000 MW机组在3种典型故障下的实际主蒸汽温度、主蒸汽压力等报警数据作为实验数据集，验证所提方法有效性。结果表明：AHCGP混合模型相较于单一高斯过程模型，冗余性报警数据的过滤准确率提高了10.7%，误判率降低了50.1%，证明了模型的有效性；与支持向量机和梯度提升决策树等成熟算法相比，漏检率、误判率均较低，具有良好的报警数据过滤性能，可准确定位冗余性报警数据，减少“报警泛滥”问题的发生。

To address the problems of complex equipment conditions, low efficiency of alarm system and frequent occurrence of "alarm flood" in thermal power units, an alarm data filtering method for thermal power units based on AHCGP hybrid model was proposed to eliminate redundant alarms. Firstly, the optimal number of clusters was determined by using the nearest neighbor propagation algorithm (AP) combined with the interclass and intraclass partitioning index (BWP), and then the agglomerative hierarchical clustering algorithm (AHC) was used for clustering to distinguish various complex operating conditions. Secondly, a Gaussian process model (GP) combined with a posteriori alarm probability estimate was used to achieve accurate filtering of unit redundancy alarm data. Finally, the actual main steam temperature and main steam pressure alarm data of a 1 000 MW unit in a power plant under three typical faults were used as the experimental data set to verify the effectiveness of the proposed method. The results show that the AHCGP hybrid model improves the filtering accuracy of redundancy alarm data by 10.7% and reduces the misclassification rate by 50.1% compared with the single Gaussian process model, which proves the effectiveness of the model. Meanwhile, compared with the mature algorithms such as support vector machine and gradient boosting decision tree, it has a lower omission rate and misclassification rate as well as a better performance in filtering alarm data, which can accurately locate redundant alarm data and reduce the occurrence of "alarm flood" problem.

TH81，TP181

河北省省级科技计划资助(22567643H);中央高校基本科研业务费专项资金(2018QN096);河北省自然科学基金(E2018502111)