摘要
针对电力负荷时序变化非线性和多周期性特点,提出一种基于分解-预测-重构框架的CVMD-GRU-DenseNet短期负荷预测方法。分解阶段依据子序列间的相关熵确定VMD最佳分解数,提高负荷序列分解质量;预测阶段针对各子序列特点筛选输入特征,规律性强的低频分量采用GRU神经网络预测模型,强随机性的高频分量采用DenseNet神经网络预测模型;最后将各分量的预测结果重构为负荷预测曲线。湖北某市四季的实际负荷算例结果表明,该方法能有效提高短期负荷预测精度,并具有较强的泛化能力。
A CVMD-GRU-DenseNet model for short-term load forecasting based on a decomposition-prediction-reconstruction framework is proposed to aim at the nonlinear and multi-period characteristics of power load time series. In the decomposition stage, the optimal decomposition number of VMD is determined according to the correlation entropy between subsequences to improve the decomposition quality. In the prediction stage, the input features are selected according to the characteristics of each sub-sequence. The GRU neural network and the DenseNet model are employed to forecast the regular low-frequency and highly random high-frequency components, respectively. Finally, the prediction results for each element are reconstructed into a load prediction curve. The short-term load forecasting results of four seasons for a city in Hubei Province show that the proposed method can effectively improve forecasting accuracy and has strong generalization ability.
作者
章可
李丹
孙光帆
谭雅
贺帅
ZHANG Ke;LI Dan;SUN Guang-fan;TAN Ya;HE Shuai(College of Electrical Engineering and New Energy,China Three Gorges University,Yichang 443002,China;Hubei Provincial Collaborative Innovation Center for New Energy Microgrid,Yichang 443002,China)
出处
《水电能源科学》
北大核心
2023年第1期207-211,166,共6页
Water Resources and Power
基金
国家自然科学基金青年基金项目(51807109)。
关键词
短期负荷预测
变分模态分解
相关熵
门控循环单元
密集连接卷积网络
short-term load forecasting
variational modal decomposition
conrrentropy
gated recurrent units
densely connected convolution networks
