基于双重分解和双向长短时记忆网络的中长期负荷预测模型  

Medium-and Long-term Load Forecasting Model Based on Double Decomposition and BiLSTM

作　　者：王继东[1] 于俊源 孔祥玉[1] WANG Jidong;YU Junyuan;KONG Xiangyu(Key Laboratory for Smart Grid(Tianjin University),Ministry of Education,Nankai District,Tianjin 300072,China)

机构地区：[1]智能电网教育部重点实验室(天津大学),天津市南开区300072

出　　处：《电网技术》2024年第8期3418-3426,I0121-I0126,共15页Power System Technology

基　　金：国家重点研发计划项目(2022YFB4200703)。

摘　　要：针对中长期电力负荷序列噪声含量高、难以直接提取序列周期规律从而影响预测精度的问题,提出了一种基于完全自适应噪声集合经验模态分解(complete ensemble empirical mode decomposition with adaptive noise,CEEMDAN)和奇异谱分析(singular spectrum analysis,SSA)双重分解的双向长短时记忆网络(bidirectional long and short time memory,BiLSTM)预测模型。首先,采用CEEMDAN对历史负荷进行分解,以得到若干个周期规律更为清晰的子序列;再利用多尺度熵(multiscale entropy,MSE)计算所有子序列的复杂程度,根据不同时间尺度上的样本熵值将相似的子序列重构聚合;然后,利用SSA去噪的功能,对高度复杂的新序列进行二次分解,去除序列中的噪声并提取更为主要的规律,从而进一步提高中长序列预测精度;再将得到的最终一组子序列输入BiLSTM进行预测;最后,考虑到天气、节假日等外部因素对电力负荷的影响,提出了一种误差修正技术。选取了巴拿马某地区的用电负荷进行实验,实验结果表明,经过双重分解可以将均方根误差降低87.4%;预测未来一年的负荷序列时,采用的BiLSTM模型将拟合系数最高提高2.5%;所提出的误差修正技术可将均方根误差降低9.7%。A bidirectional long and short-time memory network forecasting model based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise and Singular Spectrum Analysis(CEEMDAN-SSA)double decomposition is proposed in this paper,aiming at the problem that the noise content of medium and long term power load sequence is high and it is difficult to directly extract the main laws of the sequence,which affects the prediction accuracy.First,CEEMDAN was used to decompose the historical load to obtain several sub-sequences with clearer cycle rules.Multiscale entropy(MSE)calculates the complexity of all subsequences,and similar subsequences are reconstructed according to the complexity on different time scales.Then,using the SSA denoising function,the highly complex new sequences are decomposed twice to remove the noise in the sequences and extract more important rules,further improving the prediction accuracy of medium and long sequences.The final set of subsequences is input into Bidirectional Long Short-Term Memory(BiLSTM)for prediction.Finally,an error correction technique is proposed considering the influence of external factors such as weather and holidays on the power load.The experimental results show that the RMSE can be reduced by 87.4%after double decomposition.When predicting the load series of the next year,the BiLSTM model adopted can increase the fitting coefficient by 2.5%.The error correction technique mentioned in this paper can reduce the RMSE by 9.7%.

关 键 词：中长期负荷预测 二次分解 多尺度熵 奇异谱分析 双向长短时记忆网络 长序列处理 

分 类 号：TM715[电气工程—电力系统及自动化]