基于深度学习直流闭锁判断的高风险连锁故障快速搜索

Fast Search for High-risk Cascading Failures Based on Deep Learning DC Blocking Judgment

大容量直流输电系统的快速发展及多馈入直流系统的出现,增加了系统发生连锁故障的风险。建立了基于深度学习的直流闭锁快速判断模型,提出一种高风险连锁故障快速搜索方法。直流闭锁快速判断模型以网络结构和故障位置相关的稳态特征量作为输入,利用堆叠降噪自动编码器(SDAE)提取输入的高阶特征,可以快速判断交流故障是否导致直流闭锁。根据交流故障后正常线路的负载率情况衡量该故障的影响程度,进而得到线路停运风险;搜索过程中采用深度优先策略,以高停运风险为搜索方向,优先获取故障概率较高的交直流连锁故障。实际电网仿真结果表明,所提方法能够快速给出高风险交直流连锁故障的传播路径和故障概率,可用于在线安全预警和防控决策。

Rapid development of large-capacity DC transmission systems and emergence of multi-infeed DC systems increase the risk of cascading failures.A fast judgment model for DC blocking based on deep learning is established and a fast search method for high-risk cascading failures is proposed.The steady-state features related to the network structure and the fault location are selected as inputs.The stacked denoising autoencoder(SDAE)is utilized to extract high-order features of the inputs.The influence of an AC failure on other lines is measured according to the load rate of normal lines after this failure and the line outage risk is defined based on the influence degree.The depth first search(DFS)strategy is adopted in the search process,which takes the high outage risk as the search direction.AC/DC cascading failures with high failure probabilities can be screened preferentially.Simulation results show that the proposed method can quickly provide the propagation paths and failure probabilities of high-risk AC/DC cascading failures,which can be used for the online security early warning and preventive control decisions.