基于暂态录波与多元状态估计的阀冷系统故障预警及识别

Fault warning and identification of valve cooling system based on transient recording and multivariate state estimation

阀冷系统是直流输电工程的重要设备之一,提高其故障预警及识别水平具有重要实用价值。利用主泵周期切换触发的双频暂态录波数据,提取系统在切泵扰动下的特征量,从而提高运算效率;提出基于聚类算法和切换工况相结合的动态过程记忆矩阵构建方法,根据多元状态估计得出估计向量,计算估计向量与观测向量的余弦相似度,采用3σ离群检测值作为预警阈值,然后根据预警观测向量与切换工况的耦合程度,以及3种故障类型对误差的贡献率实现故障识别。最后,通过案例分析验证了所提方法的有效性。

Valve cooling system is one of the important equipment in DC transmission engineering,and improving the level of its fault warning and identification has important practical value.The dual frequency transient recording data triggered by the main pump cycle switching is used to extract the system’s characteristic quantities under pump switching disturbance,thereby improving operational efficiency.A dynamic process memory matrix construction method based on a combination of clustering algorithm and switching conditions is proposed.An estimation vector based on multivariate state estimation is obtained.The cosine similarity between the estimation vector and the ob-servation vector are calculated.3σoutlier detection is used as the warning threshold,and then fault identification is achieved based on the coupling degree between the warning observation vector and the switching condition,as well as the contribution of the three fault types to the error.Finally,the effectiveness of the proposed method is verified through case analysis.