耦合PSO与扩展RBF神经网络估计NWM模型ZTD计算精度

Estimating the ZTD accuracy of NWM model with PSO and extended RBF neural network

针对基于数值气象模型获取的对流层天顶延迟精度估计依赖外部基准的问题,本文构建耦合了粒子群算法与扩展径向基函数神经网络的ZTD精度估计模型,模型样本特征集利用NWM自身气象数据和地形特征数据构建,目标集以GNSS ZTD产品为参考值构建,模型规模结构通过层次聚类和模糊C均值聚类确定,模型参数通过粒子群算法优化。以欧洲中期天气预报中心提供的ERA5气压分层产品为NWM特例进行了模型训练和结果验证。结果表明,模型估计精度和泛化能力较好,平均估计精度优于4 mm,可在任意位置实现不依赖于外部参考基准的ZTD精度估计。

To solve the problem that the accurate estimation of the zenith tropospheric delay(ZTD)obtained from the numerical weather model(NWM)depends on external benchmarks,a ZTD accuracy estimation model coupled with particle swarm algorithm and extended RBF neural network is constructed.The model sample is built using NWM s meteorological and terrain feature data.The target set is constructed using GNSS ZTD products as reference values.The model scale structure is determined by hierarchical clustering and fuzzy C-mean clustering,and the particle swarm algorithm optimizes the model parameters.The ERA5 pressure stratification product provided by the European Centre for Medium-Range Weather Forecasts(ECMWF)is used to train the model and verify the results for the NWM particular case.The results show that the model has good estimation accuracy and generalization capability.The average estimation accuracy is better than 4 mm and can achieve ZTD accuracy estimation at any location without relying on an external reference frame.