论文部分内容阅读
利用人工神经网络研究低纬电离层参量的预测，首先我们研究从某一个月的电离层月中值预测下一个月的月中值。由于低纬电离层昼夜遵从不同的变化规律，我们将一天２４小时分成两到三个时间段进行分别预测，达到降低预测误差的目的。平均预测误差一般可以小于５－８％。其次我们将电离层看成一个系统，太阳辐射通量作为这个系统的输入，利用人工神经网络寻求太阳辐射通量与电离层Ｆ层参量之间的非线性关系，实现直接从太阳辐射通量预测电离层的月中值的目的。我们利用海南和广州两个台站１１年资料训练网络，采用训练后的网络预测电离层Ｆ层的临界频率的月中值，预测结果优于ＩＲＩ－９０的和更接近观测值。初步研究结果表明，人工神经网络能够充分利用大量的观测资料训练网络，训练后的网络不仅学习一些具体的例子，而且学会了从这些例子中所概括出的一般变化规律，寻求电离层复杂的非线性行为。
Using Artificial Neural Networks to study the prediction of ionospheric parameters at low latitudes, we first study the mid-month value of the next month from a monthly lunar ionospheric horizon. Because the low-latitude ionosphere complies with different laws of day and night, we divide it into two or three time segments 24 hours a day to make prediction separately, so as to reduce the prediction error. The average prediction error can generally be less than 5-8%. Secondly, we regard the ionosphere as a system. The solar radiation flux is used as the input of this system. The artificial neural network is used to seek the nonlinear relationship between the solar radiation flux and the ionospheric F-layer parameters and directly predict the solar radiation flux The purpose of the ionospheric mid-month. Using the 11-year data training network of two stations in Hainan and Guangzhou, we used the trained network to predict the mid-term of the critical frequency of the ionospheric F-layer. The prediction is better than the IRI-90 and closer to the observed data. The preliminary results show that the artificial neural network can make full use of a large number of observation data to train the network. The trained network not only learns some concrete examples, but also learns the general variation law summarized from these examples and seeks for the complicated ionosphere Linear behavior.