| 摘 要: 为了提高短期电力负荷预测的精度和运算效率,提出了一种基于聚类集合的经验模态分解法(Empirical Mode Decomposition,EMD)、卷积神经网络(Convolutional Neural Networks,CNN)、自注意力机制(Self Attention,SAM)及双向长短期记忆网络(Bi-directional LongShort-Term Memory,BiLSTM)的混合预测模型。该模型利用EMD算法和 K均值聚类算法将电力负荷数据分解与分组,并选取最优聚类分组数。随后,将各组数据送入CNN-BiLSTM 自注意力机制神经网络中进行预测并融合得到完整的负荷数据。实验结果显示,所提方法的平均绝对误差(MAE)、平均绝对百分比误差(MAPE)和均方根误差(RMSE)分别仅为3.436、1.049%和4.606,相较于传统算法,该模型在预测精度和效率上均有显著提升。 |
| 关键词: 短期负荷预测;经验模态分解;CNN-BiLSTM;自注意力机制 |
|
中图分类号: TP181
文献标识码: A
|
| 基金项目: 浙江省自然科学基金资助项目(LTGS23F030002);嘉兴市科技计划项目(2020AD10016) |
|
| Short-term Electricity Load Forecasting Based on Clustering Ensembles withEMD-CNN-BiLSTM Self-Attention Mechanism |
|
CHEN Yi, LIU Chunyuan
|
(School of Information Science and Engineering, Zhejiang Sc-i Tech University, Hangzhou 310019, China)
1289097138@qq.com; liuchunyuan_zjx@163.com
|
| Abstract: To improve the accuracy and computational efficiency of shor-t term electricity load forecasting, this paper proposes a hybrid forecasting model based on clustering ensembles, incorporating Empirical Mode Decomposition (EMD), Convolutional Neural Networks (CNN), Sel-f Attention Mechanism (SAM), and Bidirectional Long Shor-t Term Memory (BiLSTM). This model utilizes the EMD algorithm in conjunction with the K-means
clustering algorithm to decompose and group electricity load data, selecting the optimal number of clustering groups. Subsequently, the data from each group is fed into the CNN-BiLSTM Sel-f Attention mechanism neural network for prediction, and the results are fused to obtain the complete load data. Experimental results show that the proposed method achieves a Mean Absolute Error (MAE) of 3.436, a Mean Absolute Percentage Error (MAPE) of 1.049% , and a Root Mean Square Error (RMSE) of 4.606, indicating significant improvements in predicting accuracy and efficiency, compared to traditional algorithms. |
| Keywords: shor-t term load forecasting; Empirical Mode Decomposition ( EMD); CNN-BiLSTM; Sel-f
Attention mechanism |
