RBF神经网络在甲肝流行趋势预测中的应用研究  

Study on the application of RBF Neural Networks in forecasting prevalence of Hepatitis A

作　　者：刘艳 郑彬 邬文亮 胡建利[2] 朱叶飞[2] 刘文东[2] 

机构地区：[1]江苏省南京市六合区疾病预防控制中心,南京211500 [2]江苏省疾病预防控制中心,南京210009

出　　处：《江苏预防医学》2014年第1期7-10,共4页Jiangsu Journal of Preventive Medicine

摘　　要：目的探讨径向基函数(RBF)神经网络在甲肝趋势预测中的应用价值,同时为其他传染病的预测研究提供参考和借鉴。方法以1996-2012年江苏省甲肝月发病数据序列为基础,采用不同的样本构建方式建立不同的RBF神经网络,并根据预测效果的评价选择最优网络用于预测分析。结果江苏省甲肝流行具有逐渐下降的长期趋势,月发病数呈双峰型分布,最高峰出现在每年的3-4月(春峰),次高峰出现在7-8月(夏峰)。总共拟合了9个不同样本结构的RBF神经网络,其中以既往3年历史同期发病数以及最近3期发病数为网络输入,以当前期发病数为相应网络输出的RBF网络预测效果最佳,其拟合优度及外推预测精度分别为90.22%、89.10%。据此模型推算,2013年江苏省甲肝发病数为544例,将延续逐渐下降的长期趋势。结论通过对神经网络样本构建方式的优化研究,使建立的RBF神经网络同时拟合了历史同期的变化趋势和近期的波动规律,对甲肝的流行趋势预测取得了良好的效果,具有一定的实用价值和推广意义。Objective To study the application value of Radical Basis Funetion（RBF） Neural Networks m torecasting preva-lence of Hepatitis A; to provide reference for the prediction research on other infectious diseases. Methods Based on monthly case array of Hepatitis A in Jiangsu Province from 1996 to 2012 ,different RBF nets were fitted by different methods of sample building. The optimal net was selected according to the evaluation of prediction accuracy for forecasting application. Results The epidemics of Hepatitis A showed long-term gradual declining trend in Jiangsu. Two monthly incidence peaks were ob-served every year. The major peak occurred between March and April （spring peak）, the minor peak occurred between July and August （summer peak）. Totally 9 RBF nets were fitted with different sample structures. The best RBF net was fitted u-sing case numbers of the last three months and the same months in the last three years as the network inputs, while the case number of the current month as the network outputs. The goodness of fit and extrapolated prediction accuracy was 90.22% and 89.10%, respectively. According to this RBF net, there would be 544 Hepatitis A cases in 2013 and long-term gradual de-clining trend would continue. Conclusion In this study, the method of network sample building was optimized, so that the RBF net can fit both the prevalence trends of the same period in the history and the recent fluctuation of Hepatitis A, which improves the prediction accuracy significantly and is worthy to be applied in other fields.

关 键 词：径向基神经网络 甲型肝炎 预测 

分 类 号：R512.61[医药卫生—内科学]