基于数据驱动与非线性灵敏度函数的配电网光伏承载力研究

Voltage calculations are critical for assessing photovoltaic hosting capacity; however, acquiring precise parameters and the topology of the medium voltage distribution networks poses a significant challenge, thereby rendering traditional power flow computational methods ineffective. To address this issue, this paper introduces a hybrid method that utilizes a data-driven approach in conjunction with nonlinear functions to determine node voltages. Firstly, a deep neural network model for distribution network's power flow and voltage-power sensitivity analysis is established using historical data. This model captures the data-driven error, which reduces time consumption and increases accuracy. Secondly, a fourth-order Taylor expansion of power to voltage is derived based on the power flow mathematical equation to extrapolate voltage. This is necessary because when photovoltaic generators are connected to the nodes, the load data often exceeds the historical data range, rendering neural networks inapplicable. Finally, the sparrow search algorithm is employed to determine the hosting capacity. The proposed methods are validated using IEEE 33 and IEEE 69 case systems, demonstrating that the data-driven approach, combined with nonlinear functions, can ensure the accuracy in obtaining node voltage and the hosting capacity.