基于时序生成对抗网络的月度风光发电功率场景分析方法

针对月度风光发电功率模拟面临的变量维度高、时空特征复杂等难题，提出一种基于时序生成对抗网络的月度风光发电功率场景分析方法。采用基于RV系数的聚类技术提取代表性日发电状态，基于Markov链刻画风光日发电状态转移规律；引入缩放点积注意力机制与时序卷积网络，构建时序生成对抗网络，模拟日内风光发电功率的时序性及空间相关性；提出月度风光发电功率场景的随机生成方法。考虑电网中长期分析需求，建立月度风光发电功率场景的优化削减方法。最后，采用我国东北地区6座风电场和6座光伏电站的历史发电功率数据，验证所提方法的有效性和正确性。

A Sequential Generative Adversarial Network Based Monthly Scenario Analysis Method for Wind and Photovoltaic Power

Monthly time series simulation for wind and photovoltaic(PV)power still faces the challenges of the high dimension of random variables and complicated spatial-temporal features.A sequential generative adversarial network(SGAN)based monthly scenario analysis method was proposed for wind and PV power.The representative daily generation states are exploited by a RV-coefficient based clustering technique.The transition regularity of daily generation states was described by Markov Chain.To capture the temporal and spatial features of daily wind and PV power time series,a SGAN was established by introducing scaled dot-product attention mechanism and temporal convolutional network(TCN).A stochastic scenario generation method for monthly wind and PV power time series was subsequently presented.Considering the requirements of medium/long-term power system analyses,a scenario reduction method for monthly wind and PV power time series was developed.Finally,the effectiveness and correctness of the proposed method were verified by the historical power output data of six wind farms and six PV plants located in Northeastern China.