基于区间预测模型的流感趋势预测  

Influenza Trends Forecast Based on Interval Prediction Model

作　　者：韩帅[1] 李树刚[1] 

机构地区：[1]上海交通大学机械动力工程学院,上海200240

出　　处：《计算机仿真》2014年第9期237-242,261,共7页Computer Simulation

摘　　要：在流感暴发趋势预测模型的研究中,传统点预测是估计预测平均值的随机变量,不包含置信水平和预测区间范围等辅助决策的有用信息,导致决策者不能很好把握流感发展趋势。为了解决上述问题,提出利用神经网络上下限估计方法(LUBE)建立预测区间(PI)发展了流感趋势区间预测模型,提出了评价预测区间的宽度范围组合指标(CWC),运用蚁群算法对神经网络区间预测模型进行训练,并运用上述模型对传染病等应急医疗数据进行了仿真。为了衡量预测区间性能,改进模型与Delta、Bayesian、Holt指数平滑和支持向量机等常用预测模型建立的预测区间进行了对比。结果表明蚁群算法神经网络区间预测模型能够对流感趋势进行更为有效的分析和预测。The most existing literature researches on Influenza Trends are based on point predict models. The point prediction, as the estimated prediction mean of the random variable, does not contain the confidence level and the prediction interval width which are useful for the decision-makers to confirm the Influenza Trends and make more effective decisions. To solve this problem, a neural network lower upper bound estimation method （LUBE） which builds prediction intervals（PI） was used to develop Influenza Trends interval prediction model in this paper, a combinational coverage width-based criterion （CWC） was proposed to evaluate the prediction intervals, and an Ant Colony Algorithm was used to train the model. The model was simulated by the real emergency data. To assess the perform- ance of prediction interval, the model was compared with fore well-known prediction models including Delta model and Bayesian model, the exponential smoothing model and SVM model. The result show that the Ant Colony Algorithm neural network interval prediction model can effectively predict Influenza Trends.

关 键 词：区间预测 评估指标 蚁群算法 神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]