摘要
为了更精确地预测GNSS高程时间序列,解决传统GNSS高程时间序列预测方法特征选取不完善、稳定性差的问题,本文提出一种基于STL-XGBoost组合模型的预测方法。该方法应用STL分解对GNSS数据进行分离得到趋势项,将其从原始GNSS数据中去除后的非趋势项作为以时间为特征的XGBoost模型的输入数据,利用其强大的非线性建模能力进行GNSS高程时间序列的预测。同时,利用STL分解对趋势项进行预测,将非线性预测值和趋势项预测值等权相加得到最终的组合模型预测值。通过多个GNSS数据集的分析验证,相较于单一的STL和XGBoost模型预测结果,STL-XGBoost模型预测结果的MAE均值降低了32%,RMSE均值降低了30%,预测结果具有更高的准确性,且与原始时间序列具有较强的相关性。本文提出的组合模型在预测精度和稳定性方面有显著提升,为GNSS高程时间序列分析提供了一种有效的预测方法。
In order to more accurately predict the GNSS elevation time series and solve the problems of imperfect feature selection and poor stability of the traditional GNSS elevation time series prediction method,this paper proposes a prediction method based on the STL-XGBoost combined model.This method uses STL decomposition to separate the GNSS data to obtain the trend term,and removes the non-trend term from the original GNSS data as the input data of the XGBoost model with time as the feature,and uses its powerful nonlinear modeling ability to predict the GNSS elevation time series.At the same time,the trend term is predicted by STL decomposition,and the nonlinear prediction value and the trend term prediction value are equally weighted to obtain the final combined model prediction value.Through experimental verification of multiple GNSS data sets,compared with the prediction results of the single STL and XGBoost models,the MAE mean of the prediction results of the STL-XGBoost model is reduced by 32%,and the RMSE mean is reduced by 30%.The prediction results have higher accuracy and are strongly correlated with the original time series.The combined model proposed in this paper has significantly improved the prediction accuracy and stability,providing an effective prediction method for GNSS elevation time series analysis.
作者
何会齐
谢伟
孔冷进
HE Huiqi;XIE Wei;KONG Lengjin(Shenzhen Comprehensive Geotechnical Engineering Investigation and Design Co.,Ltd.,Shenzhen 518172,China)
出处
《测绘通报》
CSCD
北大核心
2024年第S02期282-287,共6页
Bulletin of Surveying and Mapping
