基于奇异谱分析和长短期记忆神经网络的叶绿素a浓度短时预测研究  

Study on short-term prediction of chlorophyll-a concentration based on singular spectrum analysis and LSTM Neural Network

作　　者：易洋 何先波[2] 王淳睿 YI Yang;HE Xianbo;WANG Chunrui(College of Electronics and Information Engineering,China West Normal University,Nanchong Sichuan 637009,China;College of Computer,China West Normal University,Nanchong Sichuan 637009,China)

机构地区：[1]西华师范大学电子信息工程学院,四川南充637009 [2]西华师范大学计算机学院,四川南充637009

出　　处：《智能计算机与应用》2022年第11期134-137,共4页Intelligent Computer and Applications

基　　金：西华师范大学英才科研基金项目(17YC149)。

摘　　要：有害藻华(Harmful Algal Blooms,HABs)近年来在全球频繁发生,实时预报水体藻华的出现时间和区域,可为环保监督管理部门提供有效的参考依据。为了提高水华预测的准确性,本文提出了一种基于奇异谱分析(Singular spectral analysis,SSA)和长短期记忆神经网络LSTM(Long Short-Term Memory,LSTM)的SSA-LSTM模型,将BYK站点的叶绿素a浓度时间序列分解重构为趋势特征和周期特征,并对其变化的趋势进行预测。分析对比了单个LSTM、时序神经网络(Temporal Convolutional Network,TCN)、卷积神经网络(Convolutional Neural Network,CNN)的实验结果。验证了SSA-LSTM在叶绿素a短时预测上有更好的表现,模型的RMSE、MAE和MAPE分别为0.67、0.38和0.09。Harmful Algal Blooms(HABs)occur frequently all over the world in recent years.Real-time prediction of the occurrence time and region of algal blooms in water bodies can provide effective reference for environmental protection supervision and management departments.In order to improve the accuracy of blooms prediction,a SSA-LSTM model based on singular spectrum analysis(SSA)and long short-term memory neural network(LSTM)is proposed in this paper.The time series of chlorophyll-a concentration at BYK site is decomposed and reconstructed into trend characteristics and periodic characteristics,and the changing trend is predicted.The experimental results of single LSTM,time series neural network(Temporal Convolutional Network,TCN)and convolution neural network(Convolutional Neural Network,CNN)are analyzed and compared.It is verified that SSA-LSTM had better performance in short-term prediction of chlorophyll-a,and the RMSE,MAE and MAPE of the model are 0.67,0.38 and 0.09,respectively.

关 键 词：叶绿素A LSTM 短时预测 奇异谱分析 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]