Acta Geodaetica et Cartographica Sinica ›› 2022, Vol. 51 ›› Issue (9): 1911-1919.doi: 10.11947/j.AGCS.2022.20210117

• Geodesy and Navigation • Previous Articles     Next Articles

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

ZHANG Shuang1,2, CHEN Xihong1, LIU Qiang1, LIU Zan1,3, WANG Qingli1   

  1. 1. Air and Missile Defense College, Air Force Engineering University, Xi'an 710051, China;
    2. Troops 93305, Shenyang 110000, China;
    3. Troops 93567, Baoding 074100, China
  • Received:2021-03-15 Revised:2022-03-30 Published:2022-09-29
  • Supported by:
    The National Natural Science Foundation of China (No. 61701525)

Abstract: 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.

Key words: zenith tropospheric delay, accuracy estimation, particle swarm algorithm, radial-based neural network, numerical weather mode

CLC Number: