CAJ | 학술논문

为了提高对流层湿延迟的计算精度和全球定位系统水汽反演的准确性，结合BP神经网络算法的自主学习、记忆、计算和智能处理功能，利用气象探空数据建立了计算对流层湿延迟的BP神经网路模型，其模型结构为4×15×1．分别利用霍普菲尔德模型、多元线性回归模型及BP神经网络模型计算对流层湿延迟．对比分析了3种模型的对流层湿延迟计算结果，得到结论：霍普菲尔德模型存在系统误差，精度较低；多元线性回归模型和BP神经网络模型的精度都优于霍普菲尔德模型；BP神经网络模型精度较霍普菲尔德模型改进约50％，较多元线性回归模型学习中误差改进约71．7％，检验中误差改进约2％．
위료제고대류층습연지적계산정도화전구정위계통수기반연적준학성，결합BP신경망락산법적자주학습、기억、계산화지능처리공능，이용기상탐공수거건립료계산대류층습연지적BP신경망로모형，기모형결구위4×15×1．분별이용곽보비이덕모형、다원선성회귀모형급BP신경망락모형계산대류층습연지．대비분석료3충모형적대류층습연지계산결과，득도결론：곽보비이덕모형존재계통오차，정도교저；다원선성회귀모형화BP신경망락모형적정도도우우곽보비이덕모형；BP신경망락모형정도교곽보비이덕모형개진약50％，교다원선성회귀모형학습중오차개진약71．7％，검험중오차개진약2％．
In order to improve the calculation accuracy of the tropospheric wet delay and the preci-sion of the global positioning system water vapor inversion, a back propagation ( BP) neural network model with the structure of 4 ×15 ×1 is introduced to calculate the tropospheric wet delay.The mod-el establishment involves the combination of the functions such as self-learning, memory, calculation and intelligent processing of the BP neural network algorithm as well as the usage of radiosonde da-ta.Three models are taken to calculate the tropospheric wet delay, which are the Hopfield model, the multiple linear regression model and the BP neural network model, respectively.Analysis results show that the Hopfield model is relatively low in precision because of its systematic error, whereas the other two are superior in this respect.The accuracy of the BP neural network model is improved about 50%compared to the Hopfield model, while both the learning error and inspection error are increased by 71.7%and 2%, respectively, compared to the multiple linear regression model.