摘要
针对短期风电功率预测特征提取尺度单一问题,设计一种基于多尺度特征提取的Transformer短期风电功率预测模型(MTPNet)。首先,在Transformer构架的基础上,利用维数不变嵌入,设计多尺度特征提取网络挖掘风电功率序列本身时序特征,保证了特征提取时维数不被破坏;其次,利用融合自注意力机制的长短期记忆网络挖掘气象条件与功率之间的全局依赖关系;最后,融合风电功率序列本身时序特征和气象条件依赖关系,实现短期风电功率预测。实例仿真结果表明,MTPNet模型预测精度得到提升;消融实验证明了模型各模块的可靠性和有效性,具有一定的实用价值。
Upon addressing the issue of single-scale feature extraction for short-term wind power forecasting,a Transformer-based model known as“MTPNet”(multi-scale transformer power network)is developed.Firstly,based on the Transformer architecture,dimension-invariant embeddings is employed to design a multi-scale feature extraction network to mine the temporal characteristics within the wind power sequence,to ensure that the feature dimensions remain unchanged during feature extraction.Secondly,long short-term memory(LSTM)network with a fusion self-attention mechanism is used to capture the global dependencies between meteorological conditions and power output.Finally,the temporal characteristics of the wind power sequence and the dependencies on meteorological conditions are integrated to achieve short-term wind power prediction.Simulation results demonstrate that the prediction accuracy of MTPNet model is improved.Further ablation experiments confirm the reliability and effectiveness of each module in the model,highlighting its practical value.
作者
徐武
范鑫豪
沈智方
刘洋
Xu Wu;Fan Xinhao;Shen Zhifang;Liu Yang(School of Electrical Information Engineering,Yunnan Minzu University,Kunming 650500,China)
出处
《太阳能学报》
北大核心
2025年第2期640-648,共9页
Acta Energiae Solaris Sinica
基金
国家自然科学基金(U1802271)。
