针对电站风机状态监测和故障预警问题，提出了一种基于密度峰聚类的多元状态估计方法。首先，利用密度峰聚类算法对风机正常运行工况下的历史数据进行分析，提取包含设备正常运行特征信息的数据，构建记忆矩阵；然后通过相关性原理分析观测向量与记忆矩阵之间的相似程度，使用多元状态估计技术对该观测向量进行估计。计算估计值与实测值之间的统计残差和相似度，确定风机的运行状态。最后，以南京某电厂一次风机为监测对象进行研究，建立动态故障预警模型，并结合故障实例分析验证。结果表明该新方法能够实时准确预测风机运行状态，提前发现故障征兆，指导设备运行和维护。

In the light of the problems relating to the condition monitoring and fault early warning of fans in power plants, proposed was a multi-variant state estimation method based on the density peak clustering. Firstly, an analysis was conducted of the historical data of a fan under the normal operating conditions by using the density peak clustering algorithm with the data containing the characteristic information of the equipment items under the normal operating conditions being extracted and a memory matrix being built. Afterwards, the correlation theory was used to analyze the similarity degree between the vector observed and the memory matrix and the multi-variant state estimation technology was employed to estimate the vector under observation. On this basis, the statistical residual error and similarity degree between the vector value estimated and one actually measured were calculated to determine the operation state of the fan. Finally, with a primary air fan in a power plant in Nanjing city serving as the object of study, a dynamic early warning model for faults was established and an analysis and verification were performed in combination with the real cases of faults. It has been found that the method in question can real-time and exactly predict the operation state of a fan, thus identifying in advance any sign of a fault and providing guidance for operation and maintenance of equipment items.

TP277