【摘要】 现代电网维度和不确定性的增加，使传统可靠性评价方法在计算性能上遇到瓶颈。该文提出一种基于卷积深度网络结合蒙特卡洛仿真的可靠性评价方法来提高含分布式电源复合电力系统的计算效率。该方法以分布式发电机组、输电容量和总负荷组成的样本作为输入，利用卷积特性对纳入电力系统的拓扑结构进行特征提取。采用重要性采样降低由分布式光伏与风电机组输出功率不确定性引发的可靠性评价过程波动。通过网络预测出采样系统状态下的负荷削减，以最小负荷削减量评估系统缺电概率和缺电频率组成的可靠性指数。在IEEE RTS可靠性测试系统上的实验结果表明，所提出的方法在计算综合系统可靠性指标方面是有效的。更多还原

【Abstract】 With the increase of dimensionality and uncertainty in modern power grids, the traditional reliability evaluation methods have bottlenecks in computational performance due to the need to repeatedly solve the optimal system tides. In this paper, a reliability evaluation method based on convolutional deep network combined with importance sampling is proposed to improve the computational efficiency of a composite power system containing distributed power sources. The method takes samples consisting of distributed generating units, transmission capacity and total load as inputs, and utilizes the convolutional property for feature extraction of the topology incorporated into the power system. Importance sampling is used to reduce the fluctuation of the reliability evaluation process triggered by the uncertainty in the output power of distributed PV and wind turbines. The convolutional neural network is used to predict load shedding in the sampled system states, and the reliability index consisting of the expected system power shortage supply, probability of power shortage and frequency of power shortage is evaluated with the minimum load shedding. The experimental results on the IEEE RTS reliability test system show that the proposed method is effective in calculating the integrated system reliability index.更多还原