换流阀是直流输电工程的核心设备，其价值约占换流站成套设备总价的22%～25%，其运行状态直接影响直流输电系统的可靠性。文章针对换流站阀区现有故障人工定位效率低、耗时长、严重依赖运行人员水平等问题，提出了基于度量学习和知识推理的换流站阀区故障定位方法。文中基于PSCAD(power system computer aided design)仿真分析软件构建了阀区故障仿真模型，生成阀区典型故障录波数据，通过时频域变换获取前100根最大谱线；构建基于度量学习的数据降维网络，通过最大化类间距离、最小化类内距离实现录波电气量特征提取；形成电气量-开关量-故障类型三元组，以知识图谱形式存储故障定位知识，设计知识图谱推理算法实现基于录波数据分析的故障定位。实验结果表明，该方法基于换流站阀区录波数据实现故障定位检出率在92%以上，将有效提升运行人员故障定位效率。
英文摘要:
The converter valve is the core equipment of DC transmission projects, and its value accounts for about 22%-25% of the total price of converter station equipment. Its operating status directly affects the reliability of the DC transmission system. This paper proposes a fault location method for the valve area of a converter station based on metric learning and knowledge reasoning, which addresses the problems of low efficiency, long time consumption, and heavy dependence on the level of operating personnel in manual fault location. A valve area fault simulation model is constructed based on power system computer aided design (PSCAD), generating typical fault waveform data for the valve area. The top 100 maximum spectral lines are obtained through time-frequency domain transformation. A data dimensionality reduction network based on metric learning is constructed to achieve feature extraction of recorded electrical quantities by maximizing inter class distance and minimizing intra class distance. A triplet of electrical quantity-switch quantity-fault type is formed, which stores fault location knowledge in the form of a knowledge graph, and designs a knowledge graph inference algorithm to achieve fault location based on waveform data analysis. The experimental results show that the proposed method achieves a fault location detection rate of over 92% based on the recorded waveform data of the converter station valve area, which will effectively improve the efficiency of fault location for operators.