基于海量广域时序数据分析的电力系统暂态电压稳定评估方法研究

With the reinforcement of network structures and the growth of the scale and capacity in Chinese power grids, the angle/frequency stability is gradually enhanced. However, in terms of metropolitan areas with heavy loads, due to increasing amounts of air conditioners and substantial multi-infeed DC transmissions, the transient voltage stability problem caused by insufficient supports of local reactive power is becoming more and more pronounced. Faced with the practical criterion’s shortage of adaptability and the difficulty of its threshold settings, as well as the troublesome analytics of transient voltage stability, and based on in-depth learning of massive wide-area time series data, this proposal develops a novel systematic architecture and methodology for regional transient voltage stability assessment in large-scale power grids. In particular, the maximum Lyapunov exponent-based time series approach, stemming from nonlinear theories, is combined with semi-supervised cluster learning for fundamental stability/instability evaluation. From the perspective of the receiving end, its critical features of regional transient voltage instability are modeled via the incorporation of time-varying information and spatial distributing trends, and efficiently extracted by an enhanced time series shapelet classification method, aiming at speed-up of feature extraction from high-dimension time series. Besides, the distinct sensitivities of false dismissals and false alarms are taken into account for successive classification learning, and evolutionary incremental learning is designed for the establishment of adaptive and robust online assessment models and criteria. Finally, the whole methodology is tested on a digital/physical hybrid experimental platform for possible improvement, and it’s expected to be applied to part of real-world power grids for further demonstration, so as to tightly relate theoretical innovation to practical application.

中国电网随着结构加强和规模增大，其功角/频率稳定性逐渐增强，但对于大负荷中心地区，由于空调比例不断提高、多回直流密集馈入等原因，局部无功不足造成的暂态电压稳定问题日益突出。针对现有工程判据阈值设定困难并缺乏适应性，以及暂态电压稳定机理性分析的难题，本项目从海量广域时序数据的深度学习出发，研究适用于复杂电网的区域暂态电压稳定评估新体系和新方法：以非线性系统最大Lyapunov指数的时序数据计算和半监督聚类学习方法为评估基础，对受端电网暂态电压失稳动态的“时间变化-空间分布”特性进行建模，从中提取高维时间序列的shapelet关键动态特征，大幅提升其搜索效率，在考虑漏判/误判代价敏感性基础上，通过在线增量学习建立不断进化的区域暂态电压稳定评估模型和标准。最后，开发相应软件系统，在数字/物理混合仿真实验平台上进行测试改进，并争取在实际电网中工程示范，实现理论创新和实际应用的紧密结合。

对于中国电网大负荷中心地区，由于空调比例不断提高、多回直流密集馈入等原因，局部无功不足造成的暂态电压稳定问题日益突出。针对现有工程判据阈值设定困难并缺乏适应性，以及暂态电压稳定机理性分析的难题，本项目从海量广域时序数据的深度学习出发，研究了适用于复杂电网的区域暂态电压稳定评估新体系和新方法：以非线性系统最大Lyapunov指数的时序数据计算和半监督聚类学习方法为评估基础，对受端电网暂态电压失稳动态的“时间变化-空间分布”特性进行建模，从中提取高维时间序列的shapelet关键动态特征，大幅提升其搜索效率，在考虑漏判/误判代价敏感性基础上，通过在线增量学习建立不断进化的区域暂态电压稳定评估模型和标准。最后，开发了相应软件系统，并在南方电网和国家电网进行了示范应用，实现了理论创新和实际应用的紧密结合。