基于动态关联图注意力网络的虚拟电厂居民短期负荷预测

Short-term Residential Load Forecasting Based on Dynamic Association Graph Attention Networks for Virtual Power Plant

居民短期负荷预测能够为虚拟电厂提供实时、灵活的电力需求信息,有助于虚拟电厂实现能源高效利用与优化电力市场交易。由于居民负荷相关性的日益凸显,传统预测方法仅基于单个居民历史负荷进行时序预测,无法满足规模化虚拟电厂对居民负荷关联性的综合需求。基于此,文中提出一种基于动态关联图注意力网络的虚拟电厂居民短期负荷预测方法。首先,提出了混合相关性分析方法来刻画居民负荷之间的线性和非线性关系,并进一步提出了权重剪枝阈值机制得到居民负荷混合相关性矩阵;然后,基于该矩阵构建动态关联图结构,进而提出时间图注意力网络机制以深入学习居民负荷的时空关联特性,并实现居民短期负荷预测目标;最后,以某地区实际居民负荷数据为例,验证了所提方法的有效性。

Short-term residential load forecasting can provide real-time and flexible power demand information for virtual power plants,which is helpful for virtual power plants to realize efficient utilization of energy and optimize electricity market transactions.With the increasing prominence of correlation among residential loads,traditional forecasting methods,which primarily rely on time-series forecasting based on individual residential historical load,fail to satisfy the comprehensive demands for load interconnectivity in large-scale virtual power plant.Based on this,a short-term residential load forecasting method based on dynamic association graph attention networks for the virtual power plant is proposed.Firstly,a hybrid correlation analysis method is proposed to describe the linear and nonlinear relationship between residential loads,and a weight pruning threshold mechanism is further proposed to derive the hybrid correlation matrix of residential loads.Secondly,a dynamic association graph structure is constructed based on the hybrid correlation matrix,and a temporal graph attention mechanism is proposed to deeply learn the spatial-temporal association characteristics of residential loads,achieving the objective of short-term residential load forecasting.Finally,the effectiveness of the proposed method is verified by actual residential load data from a specific region.