LSTM与Transformer融合模型在时间序列预测中的应用研究  

Research on the application of LSTM and Transformer fusion model in time series prediction

作　　者：王首晨 王利民 Wang Shouchen;Wang Limin(School of Information Engineering,Hebei University of Architecture,Zhangjiakou 075000,China;Faculty of Science,Hebei University of Architecture,Zhangjiakou 075000,China)

机构地区：[1]河北建筑工程学院信息工程学院,河北张家口075000 [2]河北建筑工程学院理学院,河北张家口075000

出　　处：《网络安全与数据治理》2025年第3期27-31,38,共6页CYBER SECURITY AND DATA GOVERNANCE

摘　　要：随着可再生能源的快速发展,风电功率预测对于电网稳定运行和能源管理具有重要意义。风电功率预测是一个复杂的非线性问题,涉及多种气象因素和环境条件。提出了一种基于长短期记忆网络(LSTM)、自适应稀疏自注意力机制(ASSA)和Transformer的融合模型,用于发电功率的时间序列预测。该模型结合了LSTM在捕捉时间序列长期依赖性方面的优势、ASSA在处理局部特征交互稀疏性方面的高效性以及Transformer在捕捉全局依赖性方面的强大并行处理能力。通过实验验证,该模型在发电功率预测任务中表现出色,尤其是在极端波动或拐点处的预测精度上有所提高。与传统方法相比,该模型能够更准确地捕捉风电功率变化的复杂性和动态性,为风电场的运营管理提供了有力的决策支持。With the rapid development of renewable energy,wind power prediction is of great significance for the stable operation of the power grid and energy management.Wind power prediction is a complex nonlinear problem that involves multiple meteorological factors and environmental conditions.This article proposes a fusion model based on Long Short-Term Memory Network(LSTM),Adaptive Sparse Self-Attention Mechanism(ASSA),and Transformer for time series prediction of power generation.This model combines the advantages of LSTM in capturing long-term dependencies of time series,the efficiency of ASSA in handling local feature interaction sparsity,and the powerful parallel processing capability of Transformer in capturing global dependencies.Through experimental verification,the model performs well in power generation prediction tasks,especially in improving prediction accuracy at extreme fluctuations or inflection points.Compared with traditional methods,this model can more accurately capture the complexity and dynamics of wind power changes,providing strong decision support for the operation and management of wind farms.

关 键 词：自适应稀疏自注意力机制 LSTM TRANSFORMER 时间序列 功率预测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]