两种记忆矩阵算法在送风机状态监测中的应用分析  

Application and Analysis of Two Algorithms for Memory Matrix in the Condition Monitoring of Blowers in Power Plants

作　　者：王博[1] 吴智群[1] WANG Bo WU Zhiqun(Xi＇ an Thermal Power Research Institute Co. Ltd. , Xi＇ an 710032, China)

机构地区：[1]西安热工研究院有限公司,陕西西安710032

出　　处：《电力科学与工程》2016年第12期16-20,共5页Electric Power Science and Engineering

摘　　要：为提高多元状态估计在送风机状态监测中的准确度,对比分析了两种不同的记忆矩阵构建算法特点及对预测结果的影响。以南京某电厂送风机为监测对象进行研究,分别用等距插值和快速聚类算法构造正常运行状态下的过程记忆矩阵;然后用多元状态估计技术对观测向量进行估计。计算估计值与实测值之间的残差,比较两种方法的预测效果。结果表明,等距插值算法在机组启停机过程或者负荷较低时预测精度更高,而快速聚类算法在机组其他正常工况下预测精度更高。该结果对于指导分段建模,从而提高模型预测精度有实用价值。In order to enhance the accuracy of the multivariate state estimation technique （MSET） in the condition monitoring of blowers in power plants, the characters and the impact on predicted results of two different algorithms which are used to create memory matrixes were compared and analyzed. Taking the air blower in a power plant in Nanjing as the monitoring object for study, the two algorithms, equidistant linear interpolation and k-means algorithm, were respectively applied to construct the memory matrixes in normal operating states. Then the current vectors were estimated by MSET. The residual errors between the estimated and measured values were calculated to compare the predicted effects of the two methods. The result shows that the predicted accuracy of equidistant linear interpolation is higher when the power generating unit is in the state of starting, stopping or in a low load condition, while the predicted accuracy of k-means algorithm precedes under other normal conditions. It is useful for us to guide the segmented modelling and improve the predictive accuracy.

关 键 词：多元状态估计 风机状态监测 记忆矩阵 等距插值 快速聚类 预测精度 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]