风光出力场景生成的条件深度卷积生成对抗网络方法

Conditional Depth Convolution Generation of Confrontation Network Method for Scenery Output Scenario Generation

针对多能源电力系统中给可再生能源消纳和系统优化调度带来不利影响的风电和光伏发电功率的不确定性的问题,提出了一种基于改进条件深度卷积生成对抗网络的风光出力场景生成方法.首先,设计适用于风电和光伏出力场景生成的条件生成对抗网络的网络结构,并采用Wasserstein距离作为判别器损失函数.然后,通过条件生成对抗网络的博弈训练使生成器学习到随机噪声与真实历史数据训练集的映射关系,从而高效地生成与真实场景分布接近的场景.最后,利用我国西北某省的风光历史出力数据进行测试,并与基于Markov和Copula理论的场景生成方法进行对比验证,结果表明文中方法生成的场景能够准确地描述可再生能源出力的不确定性.

Aiming at the adverse impact of uncertainty of wind power and photovoltaic power generation on renewable energy consumption and system optimal scheduling in multi energy power system,this paper proposes a wind and solar power generation scenario generation method based on improved conditional deep convolution generation countermeasure network.Firstly,the network structure of countermeasure network is designed based on the conditions suitable for wind power and photovoltaic output scenario generation,and the Wasserstein distance is used as the loss function of discriminator.Then,the generator can learn the mapping relationship between the random noise and the real historical data training set unsupervised by the game training of the condition generation confrontation network,so as to generate the scene close to the real scene distribution efficiently.Finally,the historical output data of a province in Northwest China is used to test,and compared with the traditional scenario generation method based on Markov and copula theory,the results show that the scenario generated by this method can accurately describe the uncertainty of renewable energy output.