基于优化VMD和BiLSTM的短期负荷预测  

Short-term Load Forecasting Based on Optimized VMD and BiLSTM Network

作　　者：谢国民[1] 陆子俊 XIE Guomin;LU Zijun(School of Electrical and Control Engineering,Liaoning Technical University,Huludao 125105,China)

机构地区：[1]辽宁工程技术大学电气与控制工程学院,葫芦岛125105

出　　处：《电力系统及其自动化学报》2025年第4期30-39,共10页Proceedings of the CSU-EPSA

基　　金：国家自然科学基金资助项目(51974151);辽宁省教育厅重点实验室基金资助项目(LJZS003)。

摘　　要：针对电力负荷数据周期性强、波动性高,预测效果不佳的问题,建立一种基于优化变分模态分解、改进沙猫群优化(improved sand cat swarm optimization,ISCSO)算法和双向长短时记忆(bidirectional long short-term memory,BiLSTM)网络的集成预测模型。首先,对原始电力负荷数据进行变分模态分解,降低数据复杂度,在变分模态分解中,引入白鲸算法对分解层数和惩罚因子寻优,优化分解效果。其次,采用Logistic混沌映射、螺旋搜索和麻雀思想引入的多策略改进方法,增加原始沙猫群优化算法的种群多样性,提升收敛精度和全局搜索能力,并用改进后的算法对BiLSTM中的超参数进行优化。然后,结合AdaBoost集成学习算法构建ISCSO-Bi LSTM-AdaBoost预测模型,将分解后的各分量输入模型预测。最后将各预测值叠加,得到最终预测结果。实验结果表明,本文建立的组合模型预测精度高,稳定性强。To solve the problems of strong periodicity,high volatility and poor prediction effect in power load data,an ensemble prediction model based on optimized variational mode decomposition(VMD),improved sand cat swarm opti-mization(ISCSO)algorithm and bidirectional long short-term memory(BiLSTM)network was established.First,the VMD of the original power load data was carried out to reduce the data complexity,in which the beluga whale optimiza-tion algorithm was introduced to optimize the number of decomposition layers and penalty factors to optimize the decom-position effect.Second,a multi-strategy improvement method introduced by Logistic chaotic mapping,spiral search and sparrow thought was used to increase the population diversity of the original sand cat swarm optimization algorithm and improve the convergence accuracy and global search capability,and the improved algorithm was used to optimize the hyperparameters in BiLSTM network.Third,combined with the AdaBoost ensemble learning algorithm,an ISCSO-BiLSTM-AdaBoost prediction model was constructed,and the decomposed components were input into the model for prediction.Finally,the predicted values were superimposed to obtain the final prediction result.Experimental results show that the combined model established in this paper has a high prediction accuracy and strong stability.

关 键 词：电力负荷预测 变分模态分解 双向长短期记忆网络 改进沙猫群优化算法 集成学习算法 

分 类 号：TM715[电气工程—电力系统及自动化]