领域知识监督下基于行波暂态波形图像认知的输电线电气扰动检测与定位

The new fault characteristics presented by AC/DC hybrid bulk power grids pose serious challenges to traditional relay protection and stability control. It is an urgent need for faster, more sensitive and more reliable grid electrical disturbance detection, event sourcing and fault location. The time-domain travelling wave transients composed of the initial traveling wave and its following multiple round-trips contain rich information about the disturbances and faults. Deep data mining in the AC transmission network is expected to improve the results of disturbance detection, fault identification and location. This project attempts to image the travelling wave transient data, and establish the analysis framework based on the time domain waveform image cognition from the perspective of machine vision. Characteristics of travelling wave propagation and its expression in waveform will be investigated. Detection and feature extraction of travelling wave transient waveform from macroscopic overview perception to local key features will be studied. Multi-point data association based on joint detection and matching of waveform image features will be studied, to actively detect disturbance events and retrospective events. Case-based reasoning fault location based on waveform image morphology cognition will be proposed. The ability of waveform cognition will be enhanced by introducing historical cases and domain knowledge supervision to imitate experts’ experience, at the same time, the powerful storage, image feature recognition, retrieval and matching ability of computers will be fully utilized, so the method and conclusions can be reused. The intelligence of fault disturbance analysis and fault location algorithms and the visualization and interpretability of the results will be improved. The performance is expected to continue to improve with the accumulation of historical samples.

交直流混联大电网所呈现的故障新特性给传统继电保护和安稳控制带来了严重挑战，迫切需要更快速、灵敏和可靠的电网电气扰动检测、溯源和定位。初始行波及其后多次往返构成的时域行波暂态量蕴含着关于电气扰动和故障丰富信息，在交流输电网中深入发掘有望提升扰动和故障感知与定位效果。本项目尝试将行波暂态量数据图像化，从机器视觉的视角建立基于时域波形图像认知的分析框架。研究行波传变特性及传播规律于波形图上的体现；研究行波暂态量波形从宏观概貌整理感知到局部关键特征的检测与提炼；研究基于波形图像特征联合检测匹配的多测点数据关联，主动发现、追溯扰动事件；研究基于波形图像形态认知的案例推理式故障测距，引入历史案例提示和领域知识监督来模拟专家经验强化波形认知，发挥计算机强大的存储能力和图像特征识别、检索和匹配能力，实现方法和结论的复用。增强故障扰动分析和定位的智能化、结果可视化和可解释性，效果有望随历史样本积累提升。

交直流混联大电网迫切需要灵敏和可靠的扰动检测与故障定位。初始行波及其后多次往返构成的暂态量蕴含丰富故障信息，但因其受干扰严重，基于行波暂态数据分析与测距可靠性差。本项目通过对行波暂态数据图像化，结合数字图像处理技术和历史案例提示，将领域知识映射为几何约束，形成基于波形图像认知的输电线行波数据分析与测距框架。首先解析故障行波并获时域形式化表达，引入合理的幅值和时间约束，提出面向电网的故障行波波过程快速计算方法，具有计算步长自适应、波过程表达清晰、对复杂大电网无需等值、计算效率高、耗时短等优点，能够分析故障行波于复杂电网传播的广度和深度，提供幅值、极性、到达时差等关键参照特征。进而由杜阿梅尔积分和伸缩平移变换阐明了去标度波形图像为分析带来的观测冗余性增强、基准自适应、历史案例强化复用、对局部缺损和突变不敏感、宏观特征洞察稳健等优势。发现了故障行波灵敏角和非轴对称性特征，提出基于变角度投影的单端多次故障行波可靠辨识方法；发现了故障行波同源性和波形图像兴趣区域特征，提出利用兴趣区域匹配的波头时差直接标定；基于波形图像的形态特征提出行波存续性的判别以及故障行波复杂程度量化分类，实现较规整行波和复杂行波的区分，增强事件发现与分类的可采信性。提出了行波测距方程的统一构造与自动解析方法，实现了多测点测距公式的自动推导；发现了近端故障波形独有特征以及与线路全无关之特性，提出基于领域知识规则与历史样本结合的近端故障辨识与定位；发现了单端行波的近似构造及跨线迁移规律，提出基于波形变尺度匹配的历史故障位置折算式故障测距新原理，充分借鉴历史样本先验信息，克服现有方法故障点反射波识别标定困难；发现了数字仿真波形与实测波形主导特征一致性，提出基于最近邻仿真逻辑推断的单端测距。基于波形图像认知的行波分析与测距将领域知识和历史数据有效结合，直观、抗干扰能力强，适用场景广泛。大量仿真和现场实测数据表明可靠、有效。