集成经验模态分解与深度学习的用户侧净负荷预测算法

随着用户侧分布式能源发电容量增长,配电网净负荷需求预测面临着更大困难.为此,提出一种改进的自适应噪声的完全集成经验模态分解(CEEMDAN)和深度信念网络(DBN)结合的用户侧净负荷预测方法.首先,通过CEEMDAN将原始净负荷数据分解为若干个频率、幅值不一的本征模态函数(IMF).然后,配合机器学习智能算法,使用DBN逐一对各个IMF分量进行特征提取和时序预测.最后,将多个目标预测结果累加得到最终用户侧短期净负荷预测结果.采用某地区实际数据进行算例分析,验证了所提CEEMDAN-DBN独立预测模型与直接预测相比,能够辨识各频率负荷分量特性,提高分布式能源与负荷耦合性增强背景下的负荷预测精度.

User-side Net Load Forecasting Method Integrating Empirical Mode Decomposition and Deep Learning

With the increase of generation capacity for distributed energy at the user side, the net load load forecasting of distribution network is facing greater difficulties. Therefore, an improved user-side net load forecasting method of complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and deep belief network (DBN) is proposed. First, the original net load data is decomposed by CEEMDAN into several intrinsic mode functions (IMFs) with different frequencies and amplitudes. Then, the DBN is used to perform the feature extraction and time-series forecasting for each IMF component one by one cooperating with the machine learning intelligent algorithm. Finally, the results of multi-target forecasting are accumulated to get the short-term net load forecasting results on the user side. The results of the case study show that the proposed CEEMDAN-DBN independent forecasting model can identify the characteristics of each frequency load component compared with the direct forecasting model, and improve the load forecasting accuracy under the background of enhanced coupling between distributed energy and load.