Zenith Wet Delay (ZWD) represents an important parameter in the Global Navigation Satellite Systems (GNSS) positioning, meteorology, and weather forecasting as it assembled to the Integrated Water Vapour (IWV) that is important for weather prediction. It is difficult to be predicted due to its temporal and spatial resolution. With the increasing development in machine learning, Artificial Neural Network (ANN) started to be used for predicting different nonlinear problems. In this paper, ANN is used for predicting ZWD from previous epoch of temperature, pressure, and Water Vapour Pressure (WVP) using twelve years (2008–2019) of data from 505 globally distributed stations. Four scenarios were followed in this study, using previous epochs from 6, 12, 18 and 24 h lags, as input for ANN. The results showed that ZWD can be predicted using ANN for all the stations using the 6-h lag scenario at the published required level of accuracy, 3.0 cm Root Mean Square Error (RMSE) and 60.2% of them have RMSE lower than 1.5 cm with 97.6% of them have the correlation coefficient (R) values ≥ 0.9. Although the accuracy of the remaining three scenarios was found to be, as expected, lower than the first scenario, their predicted ZWD showed promising R-values ≥ 0.9 to be 78.6%, 64.1% and 58.2% of the stations, for the three remaining scenarios respectively. While 68.5%, 68.9% and 65.7% of the stations were below 3.0 cm RMSE. The conclusion is that, the ANN with the suggested strategy can produce a reliable ZWD. The finding of this paper could help to improve the predicted ZWD for GNSS positioning and meteorology as well as weather forecasting.

真力时湿延迟（ZWD）是全球导航卫星系统（GNSS）定位，气象和天气预报的重要参数，因为它被组装到对天气预报至关重要的综合水汽（IWV）中。由于其时间和空间分辨率，很难预测。随着机器学习的不断发展，人工神经网络（ANN）开始用于预测不同的非线性问题。在本文中，ANN用于使用来自505个全球分布式站点的十二年（2008-2019）数据，根据以前的温度，压力和水蒸气压力（WVP）来预测ZWD。在这项研究中，我们采用了四个场景，使用之前的6、12、18和24 h时差作为ANN的输入。结果表明，在公布的要求精度水平，3.0 cm均方根误差（RMSE）和60.2％的均方根误差低于1.5 cm的情况下，可以使用ANN预测所有站点使用6小时滞后情景的ZWD。其中97.6％的相关系数（R）值≥0.9。尽管发现其余三种情况的准确性均低于第一种情况，但他们预测的ZWD显示出有希望的R值≥0.9，对于其余三种情况，其R值分别为电台的78.6％，64.1％和58.2％场景。而68.5％，68.9％和65.7％的台站均方根均方根误差（RMSE）低于3.0厘米。结论是，采用所建议策略的人工神经网络可以产生可靠的零轴WD。本文的发现可以帮助改进对GNSS定位和气象学以及天气预报的ZWD预测。