含高比例分布式电源接入的低感知度配电网电压控制方法

Voltage Control for Partially Visible Distribution Networks With High DG Penetration

大规模分布式电源的接入给中低压配电网的电压控制带来了新的挑战。考虑配电网中低四遥覆盖率和网络建模精度的限制,设计了一种无潮流模型依赖的低感知度配电网集中式电压控制方法。该方法以最小化节点电压与参考值的偏差为优化目标,并用一个深度神经网络来拟合可控节点注入功率与关键节点电压之间的函数关系,然后采用梯度下降的方法对优化模型进行求解实现系统电压控制。该方法对节点电压的拟合和控制梯度的计算都是基于深度神经网络实现,仅以系统历史运行数据作为训练数据,无需依赖系统的电气参数,因此可有效应用于低感知度配电网。最后以修改的IEEE33节点系统作为算例验证了所提方法的有效性。

The voltage control for distribution networks is challenged by increasing penetration of distributed generations(DGs).Considering the limitation of communication rates and modeling accuracy in distribution networks,a power flow model-free centralized voltage control method is proposed for partially visible distribution networks.Voltage control is redesigned by minimizing voltage deviation.A deep neural network taking visible nodal injection power as input is designed to approximate the key nodal voltage,and the optimization problem is solved with gradient descent method with the well-trained deep neural network.The control gradient is calculated using back propagation based on the deep neural network trained with historical operation data.So the method does not need accurately modeling the distribution system,making it possible to be applied to partially visible distribution networks.Effectiveness of the proposed model-free centralized voltage control method is verified with simulation of a modified IEEE 33 bus system.