基于改进强化学习算法的主动配电网在线等值建模

Online Equivalent Modeling of Active Distribution Network Based on Improved Reinforcement Learning Algorithm

随着高比例可再生能源的渗透,主动配电网的建模对电网调度和规划非常重要。为了准确描述分布式发电的随机性和负荷的时变性,提出一种基于改进强化学习算法的主动配电网等值模型。首先,考虑主动配电网中可再生能源及负荷的不确定性,对风电机组、光伏发电和负荷分别建模;其次,提出一种权重衰减策略,对采样样本进行评估;最后,基于改进强化学习算法对模型参数进行在线辨识,得到主动配电网等值模型。为了验证该等值模型的精度,在IEEE 30节点输电网和IEEE 33节点配电网的联合系统中进行仿真测试,结果表明:引入自适应学习速率以及增加搜索方向的强化学习算法能更准确辨识环境状态改变下的样本,得到的边界节点电压和视在功率误差更小;基于权重衰减策略获得样本,可以更好地区分样本对等值模型辨识的影响,提高了等值模型的精度。

With the penetration of high proportion of renewable energy,the modeling of the active distribution network is very important for network dispatching and planning.In order to accurately describe the uncertainties of distributed generations and time-varying characteristics of loads,this paper proposes an equivalent model of active distribution networks based on the improved reinforcement learning algorithm.Firstly,considering the uncertainties of renewable energy and loads in the active distribution network,it proceeds to model the wind turbine,photovoltaic power generation and load respectively.Secondly,it proposes a weight decay strategy for evaluating the samples,and obtains the equivalent model of the active distribution network by identifying model parameters online based on the improved reinforcement learning algorithm.To verify the precision of the equivalent model,the paper carries out simulation tests in the combined system based on IEEE 30-bus transmission network and IEEE 33-bus distribution network.The results show that the improved reinforcement learning algorithm introducing self-adaptive learning rate and the increase of search direction can accurately identify the samples under the change of ambient state,and obtain smaller voltage and apparent power errors of the boundary bus.Meanwhile,it is able to better distinguish the effects of the samples on identifying the equivalent model by using the weight decay strategy to obtain the samples,which improves the precision of the equivalent model.