多级负荷聚类和解耦机制的配电网短期负荷预测方法

Short-term Load Forecasting Method for Distribution Network Based on Multi-level Load Clustering and Decoupling Mechanism

为解决配电网供电分区负荷特性因用电结构与用户用电习惯差异呈现多样性,导致泛化的预测模型难以提供满意计算精度,以及新投运配变由于缺乏历史数据积累,无法为机器学习提供大量训练样本的问题,提出了一种多级负荷聚类和解耦机制的短期负荷预测方法。首先,进行基于变电站用电量以及台区用户用电特性差异的多级负荷特性聚类。随后,对不同聚类配变构建基于脉冲神经网络的短期负荷预测模型,并采用负荷标幺曲线和基准值分开预测的解耦机制应对新投运配变的小样本问题。最后,综合分类预测结果得到日负荷预测曲线。实例证明该方法能实现负荷预测的精细化,并减小新投运配变的预测误差影响,改善了综合预测结果。

The general forecasting models are not accurate enough for distribution network supply areas due to the diversity of load characteristics,which result from the differences in power supply structures and power consumption habit of users.Meanwhile,the new distribution transformers that are put into operation lack historical data accumulation,so they cannot provide massive training samples for machine learning.To solve these problems,a short-term load forecasting method based on multi-level load clustering and decoupling mechanism is proposed in this paper.First,the clustering of multi-level load characteristics is carried out based on the differences in the electricity demand of substations and users’power consumption characteristics.Then,short-term load forecasting models are constructed based on the spiking neural network(SNN)for different clusters of distribution transformer,and the small sample problem of new distribution transformers is dealt with based on the decoupling mechanism,which divides the load forecasting into a per-unit load curve and reference values.Finally,the daily load forecasting curve can be obtained by synthesizing the classified forecasting results.Simulation results of an example show that the proposed method can realize the refinement of load forecasting and reduce the prediction errors of new distribution transformers,thus improving the comprehensive forecasting results.