基于张量SOM和VAE的多风电时空功率日场景生成

A daily scenario generation for spatio-temporal power of multi-wind power based on tensor SOM and VAE

现有场景生成方法往往忽略时空功率相关性的多样性,且无法准确反映原始场景中功率时空分布关系。针对该问题,文章首先以功率日场景的时空二阶张量距离为依据,采用自组织映射神经网络将具有相似时空相关性的日场景历史样本聚合;然后分别构建各簇日场景的变分自编码器编码解码网络,编码得到各簇场景隐含特征,对其按比例进行独立抽样;解码后再聚合,获得随机模拟新场景集合。实际算例结果表明,文章所提出的方法能有效生成符合真实多风电场功率时空相关性和概率分布规律的多风季和少风季风电功率时空场景数据。

Existing scene generation methods often ignore the diversity of spatio-temporal power correlation,and cannot accurately reflect the spatio-temporal distribution relationship of power in the original scene.To solve this problem,this paper based on the spatio-temporal second-order tensor distance of the power daily scenes,it uses the self-organizing map neural network to aggregate the historical samples of the daily scenes with similar spatio-temporal correlation,then builds the variational auto-encoder encoding and decoding network for each cluster of daily scenes respectively,encoding the hidden features of each cluster of scenes to independently sample them in proportion,and finally decoding and aggregating them to obtain a set of random simulated new scenes.The actualexample results show that the proposed method can effectively generate spatio-temporal scene data of wind power in the windy and less windy seasons in accordance with the spatio-temporal correlation and probability distribution of real multi-wind farm power.