面向直流受端新型电力系统暂态电压稳定的紧急控制策略

在直流受端新型电力系统中，新能源电源及直流的接入导致同步机开机减少，动态无功功率相对紧张，暂态电压失稳风险显著增大。为高效快速获取紧急控制策略，即直流电流控制方案，并使其适应不同的电网运行方式和故障场景，尤其是电网拓扑结构的变化，基于图卷积神经网络（GCN）对常规深度强化学习模型深度确定性决策梯度（DDPG）的网络结构进行改造，构建了GCN-DDPG融合模型。在此基础上，引入双评价网络机制和评价网络与动作网络非同步更新策略以提升算法效果。进一步地，在应用方面，基于GCN-DDPG融合模型构建紧急控制模型并将其下达至安控主站，安控主站将依据电网实际运行方式和故障等信息，对紧急控制策略进行在线量化计算，并发送至直流控保系统执行。最后，通过改造的IEEE 14节点系统验证所提方法的有效性和优越性。

Emergency control for transient voltage stability of HVDC receiving-end new power system

In the high voltage direct current(HVDC) receiving-end new power system, the connection of renewable energy and HVDC leads to the reduction of synchronous generators and a relatively tensive state of dynamic reactive power, the risk of transient voltage instability significantly increases. Therefore, in order to efficiently obtain the emergency control strategy, i. e. DC current control scheme, and adapt it to different operation modes and fault scenarios, especially the changes in the grid topology, the networks of deep deterministic policy gradient(DDPG) are modified by graph convolutional network(GCN) and a GCN-DDPG integrated model is constructed. On this basis, the dual evaluation network mechanism and a non-synchronous update strategy between critic network and actor network are adopted to improve the effectiveness of the algorithm. Further, based on the GCN-DDPG integrated model, an emergency control model is constructed and sent to the security control master station, the security control master station will quantitatively calculate the corresponding emergency control strategy online according to the actual operation mode and fault and send it to the DC control and protection system for implementation. Finally, the effectiveness and superiority of the proposed method are verified by a modified IEEE 14-bus system.