径向基函数神经网络在传染病发病率预测中的应用

Introduction on a forecasting model for infectious disease incidence rate based on radial basis function network

传染病发病率的有效预测在传染病防治工作中意义重大,其预测理论和方法的研究一直是一个热点。现实中影响传染病发病的因素众多、相互关系复杂,各因素的作用机制通常不能或无法用精确的数学语言来准确描述。本文采用基于时间序列的径向基函数(RBF)神经网络模型对传染病发病率进行预测,以实现传染病发病序列的非线性逼近。在实例分析中,以某市1991-2002年乙型肝炎(乙肝)月发病率数据建模,经过网络的不断学习和训练,得到合适的预测模型后对2003年1-6月的月发病率进行预测。通过与2003年1-6月的实际发病率进行比较分析以验证建模的可靠性,并与传统的时间序列模型预测结果进行比较,结果表明应用RBF神经网络模型对乙肝发病率的短期预测精度更高、效果更好。

It is important to forecast incidence rates of infectious disease for the development of a better program on its prevention and control. Since the incidence rate of infectious disease is influenced by multiple factors, and the action mechanisms of these factors are usually unable to be described with accurate mathematical linguistic forms, the radial basis function （RBF） neural network is introduced to solve the nonlinear approximation issues and to predict incidence rates of infectious disease. The forecasting model is constructed under data from hepatitis B monthly incidence rate reports from 1991 - 2002. After learning and training on the basic concepts of the network, simulation experiments are completed, and then the incidence rates from Jan. 2003 - Jun. 2003 forecasted by the established model. Through comparing with the actual incidence rate, the reliability of the model is evaluated. When comparing with ARIMA model, RBF network model seems to be more effective and feasible for predicting the incidence rates of infectious disease,observed in the short term.