基于CEEMDAN模态分解和TCN-BiGRU的短期电力负荷预测

Short-term Electricity Load Forecasting Based on CEEMDAN Decomposition and TCN-BIGRU Model

唐竹 ¹肖宇航 ¹郭淳 ¹梅秋怡 ²王森 ³郑宽昀 ⁴毛新颖²

扫码查看

作者信息

1. 国网北京市电力公司,北京 100031
2. 南京农业大学人工智能学院,江苏南京 211800
3. 河海大学电气与动力工程学院,江苏南京 211100
4. 南瑞集团(国网电力科学研究院)有限公司,江苏南京 211106
折叠

摘要

为了深入分析复杂的短期电力负荷变化规律和其长期相关性,提出了一种基于自适应噪声完备集合经验模态分解(CEEMDAN)的时间卷积网络(TCN)与双向门控循环单元(BiGRU)的融合模型.采用CEEMDAN方法将完整的电力负荷数据分解成多个本征模态函数序列和残差项序列,通过时间卷积网络模型的扩展卷积策略提取时间序列的特征;结合双向门控循环单元的非线性拟合能力进一步提取TCN训练输出的非线性特征.实验结果表明,在短期电力负荷预测中,所提模型预测的决定系数为96.6%、平均绝对百分比误差为3.82%、平均绝对误差为45.5,各项指标均优于对比模型.

Abstract

In order to conduct an in-depth analysis of the complex short-term power load variation patterns and their long-term correlations,the paper proposes a fusion model combining the temporal convolutional network(TCN)based on the complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)and the bidirectional gated recurrent unit(BiGRU).The CEEMDAN method is employed to decompose the complete power load data into multiple intrinsic mode function sequences and a residual term sequence.The extended convolution strategy of the TCN model is utilized to extract the features of the time series.The nonlinear fitting ability of the BiGRU is combined to further extract the nonlinear features of the TCN training output.The experimental results demonstrate that,in the short-term power load forecasting,the determination coefficient of the proposed model for prediction is 96.6%,the mean absolute percentage error is 3.82%,and the mean absolute error is 45.5.All these indicators are superior to those of the contrast models.

关键词

短期电力负荷预测/模态分解/时间卷积网络/双向门控循环单元

Key words

Short-term electric load forecasting/Model decomposition/Temporal convolutional networks/Bidirectional gated loop cell

引用本文复制引用

出版年

2024

智慧电力

陕西省电力公司

智慧电力

CSTPCDCSCD北大核心

影响因子：0.831

ISSN：1673-7598

段落导航