计及复杂气象耦合特性的模块化去噪变分自编码器多源–荷联合场景生成

The Joint Scenario Generation of Multi Source-load by Modular Denoising Variational Autoencoder Considering the Complex Coupling Characteristics of Meteorology

气象因素的强随机性与强波动性直接影响新能源出力与用户用电行为。针对基于整体历史数据生成多源-荷联合场景集时难以体现特定气象下的多源-荷概率分布特性的不足,提出一种计及气象因素差异的模块化去噪变分自编码器(modular denoising variational autoencoder,MDVAE)多源-荷联合场景生成模型。首先,分析风速、辐照、负荷等与气象因素相关性,确定源-荷气象耦合特征集;在此基础上,针对历史气象数据集进行聚类,获得具有不同气象特点的聚类结果;之后,以类内所含日期中风速、辐照、负荷历史数据,构建基于数据驱动的MDVAE联合场景生成模型;最后,通过将生成的风速、辐照转化为风-光出力,构建多源-荷场景。实测数据分析表明,新方法生成场景集能体现不同气象条件下差异性,并能有效提高生成场景集与实测数据间概率分布的相似性。

The meteorological randomness and volatility directly affect renewable energy outputs and electricity consumption behavior. To overcome the shortcomings of generating multi source-load scenarios based on total historical data, a model of multi-source/load scenario generation using modular denoising variational autoencoder(MDVAE) was proposed, which takes into account the differences of meteorological factors. Firstly, the correlation between wind speed, radiation, load and meteorological factors were analyzed, and the source-load meteorological coupling features were determined. On this basis, the historical meteorological data was clustered to obtain the clustering results with different meteorological characteristics. Then, the data-driven MDVAE joint scene generation model was constructed with wind speed, irradiation and load history data of various clusters. Finally, multi-source-load scenarios were constructed by converting the generated wind speed and radiation into wind-solar output. The measured data analysis shows that the new method can generate the scenes to reflect the difference under different meteorological conditions, and effectively improve the similarity of probability distribution between the generated scenes and the measured data.