基于ViT和LSTM的风速多步预测  

MULTI-STEP WIND SPEED FORECASTING BASED ON VIT AND LSTM

作　　者：向玲[1] 陈锦鹏 付晓孟婷 姚青陶 Xiang Ling;Chen Jinpeng;Fu Xiaomengting;Yao Qingtao(School of Energy,Power and Mechanical Engineering,North China Electric Power University,Baoding 071003,China)

机构地区：[1]华北电力大学能源动力与机械工程学院,保定071003

出　　处：《太阳能学报》2024年第9期525-533,共9页Acta Energiae Solaris Sinica

基　　金：国家自然科学基金(52075170,52175092)。

摘　　要：精确的风速预测对风力发电具有指导作用,据此提出一种多维时间序列下Vision Transformer(ViT)和长短时记忆网络(LSTM)的风速预测方法,实现对风速的超前一步和超前多步预测。结合斯皮尔曼系数(Spearman)和变分模态分解将风速分解为多维时间序列,多维时间序列能更好地表征原始风速的周期性和波动性;采用ViT提取多维时间序列中的特征以及隐藏信息,在保持ViT模型自注意力机制优势的同时,用LSTM进一步建立所提取特征和风速之间的关系,从而提高ViT-LSTM模型的泛化性和预测准确性。使用内蒙古某风场记录的数据进行试验分析,在超前1步、超前6步、超前12步和超前24步预测时,该方法的平均绝对误差分别比LSTM模型减少了15.15%、34.41%、68.32%和81.71%,结果表明该模型在超前多步风速预测方面具有较好的效果。Accurate and dependable wind speed forecasting plays a guiding role in wind power generation.A novel wind speed forecasting method combined with Vision Transformer(ViT)and long short-term memory(LSTM)is proposed,which is applied in multidimensional time series for achieving one-step ahead and multi-step ahead wind speed forecasting.Firstly,the wind speed is decomposed into multidimensional time series by combining Spearman's coefficient and variational mode decomposition(VMD)in order to enhance the ability of forecasting model to capture the periodicity and volatility of the wind speed sequence.And then,ViT is used to extract features and hidden information from multi-dimensional time series.Finally,while maintaining the self-attention mechanism advantage of ViT model,the LSTM component is employed to further establish the relationship between extracted features and wind speed.This approach is particularly effective in handling the time-dependent nature of the data and capturing the correlations between variables,so as to improve the generalization and forecasting accuracy of ViT-LSTM model.The experimental analysis is conducted using data recorded from a wind farm in Inner Mongolia.The proposed method achieved a reduction of 15.15%,34.41%,68.32%,and 81.71%in mean absolute error compared to the LSTM model at one-step ahead,six-step ahead,twelve-step ahead and twenty-four-step ahead forecasting,which indicates that the proposed model performs well in multi-step wind speed forecasting.

关 键 词：风速 预测 长短时记忆网络 变分模态分解 VIT 

分 类 号：TM614[电气工程—电力系统及自动化]