基于Bi-LSTM和改进残差学习的风电功率超短期预测方法  

Wind Power Ultra-short Term Forecasting Method Based on Bi-LSTM and Improved Residual Learning

作　　者：王进峰[1,2] 吴盛威 花广如 吴自高[1,2] WANG Jinfeng;WU Shengwei;HUA Guangru;WU Zigao(School of Energy Power and Mechanical Engineering,North China Electric Power University,Baoding 071003,China;Baoding Key Laboratory of Advanced Design and Smart Manufacturing,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,河北保定071003 [2]保定市先进设计与智能制造重点实验室,河北保定071003

出　　处：《华北电力大学学报(自然科学版)》2025年第1期56-65,共10页Journal of North China Electric Power University:Natural Science Edition

基　　金：国家自然科学基金资助项目(51301068);河北省自然科学基金资助项目(E2019502060).

摘　　要：现有的方法在以风电功率时间序列拟合功率曲线时,难以表达风电功率数据所包含的趋势性和周期性等时间信息而出现性能退化问题,从而导致预测精度下降。为了解决性能退化问题从而提高风电功率时间序列预测的精度,提出了基于双向长短时记忆(Bi-LSTM)和改进残差学习的风电功率预测方法。方法由两个部分组成,第一部分是以Bi-LSTM为主的多残差块上,结合稠密残差块网络(DenseNet)与多级残差网络(MRN)的残差连接方式,并且在残差连接上使用一维卷积神经网络(1D CNN)来提取风电功率值中时序的非线性特征部分。第二部分是Bi-LSTM与全连接层(Dense)组成的解码器,将多残差块提取到的功率值时序非线性特征映射为预测结果。方法在实际运行的风电功率数据上进行实验,并与常见的残差网络方法和时间序列预测方法进行对比。方法相比于其他模型方法有着更高的预测精度以及更好的泛化能力。When fitting the power curve using wind power time series,the existing method has difficulty expressing the temporal information,such as trend and periodicity,inherent in wind power data,which in turn leads to a decrease in prediction accuracy.To address this performance degradation and enhance the accuracy of wind power time series prediction,we propose a wind power prediction method based on bidirectional long short-term memory(Bi-LSTM)and improved residual learning.The method consists of two parts,the first part is integrating the residual connection method of dense residual block network(DenseNet)and multi-level residual network(MRN)on Bi-LSTM-dominated multi-residual block,and use one-dimensional convolutional neural network(1D CNN)on the residual connection to extract the nonlinear temporal features in wind power data.The second part is a decoder composed of Bi-LSTM and a fully connected layer(Dense),which maps the nonlinear features extracted by the multiple residual blocks into the prediction results.The proposed method is experimented on the actual operating wind power data and compared with the common residual network method and time series forecasting method.Compared with other model methods,the method has higher prediction accuracy and better generalization ability.

关 键 词：深度学习 残差网络 风电功率预测 双向长短时记忆 一维卷积神经网络 

分 类 号：TM614[电气工程—电力系统及自动化]