CAJ | 학술논문

为了提高电离层短期区域预报效果，提出了基于支持向量机方法考虑太阳活动、地磁活动、中高层大气、地理位置等因素对电离层的影响．对中国地区电离层F2层临界频率（foF2）提前1 h的区域预报模型，将支持向量机的预报模型与输入同样参数的反向传播神经网络和国际参考电离层模型从多方面进行对比分析，结果显示，支持向量机模型的年平均预报相对误差相对神经网络和国际参考电离层模型在太阳活动高年分别降低了2.5％和9.6％，在太阳活动低年分别降低了1.9％和7.5％．在低纬度地区，支持向量机模型的预报优势更加显著，在高年和低年相对反向传播神经网络分别降低了3.2％和2.7％．对暴时，支持向量机模型也表现出一定的预报能力．这表明支持向量机模型应用在中国区域电离层foF2短期预报上，相对反向传播神经网络和国际参考电离层模型更有优势．
위료제고전리층단기구역예보효과，제출료기우지지향량궤방법고필태양활동、지자활동、중고층대기、지리위치등인소대전리층적영향．대중국지구전리층F2층림계빈솔（foF2）제전1 h적구역예보모형，장지지향량궤적예보모형여수입동양삼수적반향전파신경망락화국제삼고전리층모형종다방면진행대비분석，결과현시，지지향량궤모형적년평균예보상대오차상대신경망락화국제삼고전리층모형재태양활동고년분별강저료2.5％화9.6％，재태양활동저년분별강저료1.9％화7.5％．재저위도지구，지지향량궤모형적예보우세경가현저，재고년화저년상대반향전파신경망락분별강저료3.2％화2.7％．대폭시，지지향량궤모형야표현출일정적예보능력．저표명지지향량궤모형응용재중국구역전리층foF2단기예보상，상대반향전파신경망락화국제삼고전리층모형경유우세．
Ionospheric short‐term forecasting is very important to radio communication , navigation and radar systems . In this paper , in order to improve the regional prediction accuracy of ionosphere , a model of regional prediction of the ionospheric F2 layer critical frequency in China area 1 hour in advance is set up based on the support vector machine (Support Vector Machine , referred to as SVM for short) method . In this model , the influence of solar activity , geomagnetic activity , the upper atmosphere , geographical location and other factors on the ionosphere is taken into consideration . Results of this model is compared to Back‐Propagation referred to as BP for short the neural network of the same input parameters and the IRI model ( International Reference Ionosphere , referred to as IRI for short) . The results show that the average relative error of annual prediction of SVM in high solar activity years decreases by 2.5% and 9.6% , respectively , compared with the neural network and the IRI models and in low solar activity decreases by 1.8% and 7.5% , respectively . In the low latitude area , the prediction of SVM has more significant advantages over the BP neural network . In the high and low solar activity years it decreases by 3.2% and 2.7% , respectively . During the storm time SVM also shows a relatively good prediction ability . This proves that the developed model based on SVM in the paper has more advantages over the BP neural network and IRI model .