摘要
受多种因素影响,电离层电子总含量(TEC)时间序列具有非线性、非平稳性特征,为提升长短期记忆(LSTM)神经网络模型在电离层TEC预报中的精度,本文在该神经网络模型的基础上,引入奇异谱分析(SSA)与粒子群优化(PSO)算法,构建了新的SSA-PSO-LSTM模型。一方面,利用了SSA对TEC时间序列进行数据预处理;另一方面,利用粒子群优化算法改进LSTM神经网络模型参数。选用欧洲地球参考框架(EUREF)提供的格网点电离层TEC时间序列数据进行实验,实验结果表明,在磁平静期与磁暴期,该组合模型的TEC预报结果均方根误差分别为0.28个总电子含量单位(TECu)、0.83个TECu,平均相对精度分别为96.35%、91.33%,均优于对比模型,验证了本文提出的组合预报模型的有效性与优越性。平均相对精度分别为96.35%、91.33%,均优于对比模型,验证了本文提出的组合预报模型的有效性与优越性。
Due to various factors,the time series of total electron content(TEC)in the ionosphere has nonlinear and nonstationary characteristics.To improve the accuracy of the long short-term memory(LSTM)neural network model in predicting TEC in the ionosphere,this paper introduced singular spectrum analysis(SSA)and particle swarm optimization(PSO)algorithms into the neural network model and constructed a new SSA-PSO-LSTM model.On the one hand,SSA was used for data preprocessing of the time series of TEC,and on the other hand,the PSO algorithm was used to improve the parameters of the LSTM neural network model.The experiment was conducted using the time series data of TEC in the ionosphere in grid points provided by the European Reference Organisation for Quality Assured Breast Screening and Diagnostic Services(EUREF).The experimental results show that during the magnetic calm period and the magnetic storm period,the root mean square errors of the TEC prediction results of the combined model are 0.28 TECu and 0.83 TECu,respectively,with an average relative accuracy of 96.35%and 91.33%,which are superior to those of the comparative model,verifying the effectiveness and superiority of the proposed combined prediction model in this paper.
作者
郑泽辰
黄志标
ZHENG Zechen;HUANG Zhibiao(Zhejiang Institute of Surveying and Mapping Science and Technology,Hangzhou,Zhejiang 310030,China)
出处
《北京测绘》
2024年第5期786-792,共7页
Beijing Surveying and Mapping
基金
国家自然科学基金(42261074)。
