基于深度稀疏自编码网络和场景分类器的 电网气象故障预警方法

为保证电网安全稳定运行，提高电网防灾减灾和弹性水平，提出了一种基于深度稀疏自编码网络和场景分类器的电网气象故障预警方法。首先，采用主客观权重相结合的动态赋权方法，对气象因子进行初始赋权，以合理表征不同气象因子对电网故障的影响程度。然后，对传统的深度自编码网络增加稀疏性约束条件，以提高网络训练的收敛性，并在深度自编码网络的最后一层增加场景分类器，以提高气象因子与电网故障场景间关联关系的合理性。最后，将带权重的气象因子以及设备因子和环境因子作为深度稀疏自编码网络的输入，利用支持向量机构建多因素耦合的电网气象灾害故障预警模型。采用实际电网故障算例验证了所提方法的有效性。

Early warning method for a power grid fault caused by meteorology based on a deep sparse auto-encoder network and scene classifier

There are problems of complex operation and high overall cost caused by the low level of integration of current online monitoring equipment in a ring cabinet. Thus a monitoring terminal of the ring cabinet of the internet of things (IoT) is developed, integrated with micro-power sensor access, online monitoring of mechanical characteristics and of partial discharge and standardized communication interface. The structure of the sub-module and the software architecture based on the container architecture are introduced in detail, and the realization method of data zero drift adjustment, data trigger threshold selection and feature analysis related to the online monitoring of mechanical properties is emphasized. At the same time, an improved peak retention mechanism for local discharge sampling is introduced. In order to verify performance, an experimental platform is built and some typical mechanical characteristics faults are simulated. The results show that the monitoring data are accurate and consistent, and can meet the requirements of field applications. This work is supported by the National Key Research and Development Program of China (No. 2020YFE0200400).