基于多变量DSD-LSTM模型的有效波高预测  

PREDICTION OF SIGNIFICANT WAVE HEIGHT BASED ON MULTIVARIABLE DSD-LSTM MODEL

作　　者：庞军恒 黄炜楠 董胜[1] Pang Junheng;Huang Weinan;Dong Sheng(College of Engineering,Ocean University of China,Qingdao 266404,China)

机构地区：[1]中国海洋大学工程学院,青岛266404

出　　处：《太阳能学报》2024年第7期121-127,共7页Acta Energiae Solaris Sinica

基　　金：国家自然科学基金(52171284)。

摘　　要：利用改进的完全集合经验模态分解(ICEEMDAN)和递归量化分析方法设计一种新的信号分解算法(DSD),该算法将原始信号分解为确定性成分和随机性成分。考虑风速、风向对波高的影响前提下,将DSD算法与长短时记忆网络(LSTM)结合建立多变量混合模型DSD-LSTM-m进行有效波高的预测。该模型与单独的LSTM模型相比明显提高了预测精度,与单变量混合模型DSD-LSTM-u相比具有更好的预测效果。A new signal decomposition algorithm(DSD)is designed by using the ICEEMDAN and recursive quantification analysis method,which divides the original signal into deterministic and stochastic components.Considering the influence of wind speed and wind direction on wave height,a multi-variable DSD-LSTM model was established by combining DSD algorithm with Long and Short-Term Memory network(LSTM)to predict significant wave height.The proposed model significantly improved the prediction accuracy compared to the single LSTM model and has better prediction performance compared to the univariate hybrid model DSD-LSTM-u.

关 键 词：波浪能 波高预测 时间序列 信号处理 深度学习 长短时记忆网络 

分 类 号：P743.2[天文地球—海洋科学]