中国区域BP-Adaboost强预测器对流层天顶延迟建模研究

选取中国大陆构造环境监测网（陆态网）提供的155个测站2014~2018年对流层延迟产品，基于BP-Adaboost算法将多个弱神经网络预测器集成为强预测器，建立新的无气象参数对流层延迟计算模型。利用陆态网2019年参与建模的141个建模测站、未参与建模的62个测站的对流层延迟产品和中国区域86个无线电探空站解算出的对流层延迟精确值对BP-Adaboost模型进行精度评定，结果表明，新模型的平均偏差分别为0.62 mm、-1.16 mm和12.32 mm，均方根误差分别为25.30 mm、26.72 mm和46.29 mm，优于常见的无气象参数模型；BP-Adaboost模型在内陆地区或海拔2 km以上地区具有更高的精度，能够满足中国大陆区域卫星导航用户实时对流层延迟改正的需求。

Study on Modeling of Tropospheric Zenith Delay in China with BP-Adaboost Strong Predictor

We retrieve tropospheric delay data from 155 stations from 2014-2018 Crustal Movement Observation Network of China (CMONOC). We use the BP-Adaboost algorithm to integrate multiple weak neural network predictors into a strong one in order to establish a new tropospheric delay model without meteorological parameters. The accuracy of the BP-Adaboost model is evaluated using the tropospheric delay products of 141 CMONOC stations in 2019, 62 stations excluded in modeling and 86 radiosonde stations in China. The results show that the biases of the new model are 0.62 mm, -1.16 mm and 12.32 mm, and the root mean square errors are 25.30 mm, 26.72 mm and 46.29 mm, respectively, which are better than the common models without meteorological parameters. In addition, the BP-Adaboost model could achieve higher accuracy in inland areas or areas above 2 km above sea level, meeting the real-time tropospheric delay correction needs of Chinese satellite navigation users.