基于生成对抗网络的配电网与多微网随机调度

随着可再生能源机组以多微网的形式接入配电网,其出力的不确定性会给配电网与多微网调度带来挑战。因此,如何对配电网与多微网中可再生能源的特性进行分析,准确把握可再生能源的出力特性,建立考虑可再生能源出力特性的配电网与多微网调度模型,成为目前亟待研究和解决的问题。本文提出了一种基于Wasserstein生成对抗网络的配电网与多微网日前随机调度方法。首先针对风电以及光伏日前预测的不确定性,采用基于Wasserstein生成对抗网络的数据驱动算法,对风电和光伏出力预测误差进行场景生成；对于生成的风光出力场景,基于K-mediods场景削减法得到风光典型场景；在配电网与多微网调度目标函数中综合考虑调度的经济性指标以及韧性指标,基于场景法模拟可再生能源出力的不确定性,建立配电网与多微网日前随机调度模型并求解。仿真结果表明,所提的配电网与多微网随机调度模型在可再生能源出力场景生成方面,相比于传统假定概率分布的生成方法,其生成的场景更接近实际场景。

Stochastic Scheduling of Distribution Network and Multi-microgrids Based on Generative Adversarial Networks

As renewable energy generations connect to the distribution network (DN) through multi-microgrids (MMGs), the uncertainty will bring challenges to the dispatch of the DN and MMGs. Therefore, how to analyze the characteristics of renewable energy in DN and MMGs,Saccurately grasp the output characteristics of renewable energy in the microgrid, and establish a scheduling model for the DN and MMGs considering the output characteristics of renewable energy, has become an urgent problem to be solved. This paper proposes a day-ahead stochastic scheduling method for DN and MMGs considering the output characteristics of renewable energy. First, aiming at the uncertainty of wind power and photovoltaic forecasts, the data-driven method based on Wasserstein generative adversarial network is used to generate scenarios for the wind power and photovoltaic output, and the typical scenarios are obtained based on the scenario reduction method. In the dispatching objective function of DN and MMGs, the economic index and toughness index of dispatching are considered comprehensively, and the uncertainty of renewable energy output is simulated by the scenario-based method. Finally, the day-ahead stochastic dispatching model of distribution network and multi-micro network is established and solved. The simulation results illustrate that in terms of the generation of renewable energy output scenarios, the proposed two-stage stochastic scheduling model of DN and MMGs generates scenarios that are closer to actual scenarios than the traditional scenario generation methods that assume the probability distribution of renewable energy output.