基于CEEMDAN-AsyHyperBand-MultiTCN的短期风电功率预测
SHORT-TERM WIND POWER PREDICTION BASED ON CEEMDAN-AsyHyperBand-MultiTCN
刘凡 1李捍东 1覃涛1
作者信息
- 1. 贵州大学电气工程学院,贵阳 550025
- 折叠
摘要
为减少风电功率短期预测误差,提高风电利用效率,提出一种基于经验模态分解和异步超参数优化的多层时间卷积网络(CEEMDAN-AsyHyperBand-MultiTCN)的短期风电功率预测方法.首先,确定序列分量的数量,并使用自适应噪声完备集合经验模态分解(CEEMDAN)对原始风电功率进行分解,构成训练数据集.其次,使用深度残差级联(DRnet)构建多层的时间卷积网络(TCN),并使用AsyHyperband算法对序列分量模型进行超参数寻优.最后,对序列分量分别进行预测,重构预测结果得到预测值.实验表明,该文提出的方法相比于其他方法能有效降低风电功率预测误差.
Abstract
In order to improve the utilization efficiency and accuracy of short-term wind power,this paper proposes a method based on complete ensemble empirical mode decomposition with adaptive noise and multi-layer temporal convolution networks(CEEMDAN-AsyHyperBand-MultiTCN).Firstly,determine the number of sequence components,and then decompose the time series of wind power as training dataset using CEEMDAN.Secondly,apply the Deep Residual Cascade(DRnet)to build a multi-layer Temporal Convolutional Networks(TCN)model for each component,and the AsyHyperband algorithm is used to optimize the hyperparameters for the components model.Finally,the final prediction result is obtained after reconstructing the prediction results of each component model.The experimental results show that the proposed method can effectively reduce the wind power prediction error compared with other methods.
关键词
风电功率/预测/神经网络/多层/集成经验模态分解/超参数搜索Key words
wind power/forecasting/neural networks/multilayers/complete ensemble empirical mode decomposition with adaptive noise/hyperparameter search引用本文复制引用
出版年
2024
NETL
NSTL
维普
万方数据