基于逆传播神经网络的麻疹短期发病预测研究

Short- term Prediction of the Measles Based on BP Neural Network

目的建立用于麻疹短期预测的逆传播神经网络(BPNN)模型,并对麻疹的短期发病数进行预测,为制定麻疹的预防措施提供理论依据。方法确定预测模型的基本结构,以2011年1月—2012年11月全国麻疹的月发病数为训练样本,以2012年12月的发病数为检验样本,采用BPNN算法训练预测模型。利用该模型对2013年1—6月的麻疹发病数据进行预测。结果所建立的BPNN模型在仿真预测样本点的平均预测相对误差为0.774%,检验样本的相对误差为1.296%。利用该BPNN模型采用分步预测法得到2013年1—6月麻疹的发病数预测值,将预测得到的数值乘以2 000,得到全国麻疹2013年1—6月的发病数,分别为787、786、603、523、573、629。结论 BPNN模型具有良好的预测精度,适合用来进行麻疹的短期发病预测。

Objective To establish a BP neural network （BPNN） model for measles short - term prediction to predict the occurrence of measles, and to provide theoretical evidence for preventive measures of measles. Methods The basic structure of the prediction model was established. The monthly amount of measles from January 2011 to November 2012 was taken as train- ing sample, and the amount of measles in December 2012 was taken as testing samples. The BPNN calculating method was used to train the prediction model, which was used to predict the amount of measles from January to June in 2013. Results The aver- age relative error of simulated predicted samples was 0. 774%, and the predicted error of tested samples was 1. 296%. By using the BPNN model and predicting step by step, we got the predicted incidence of measles from January to June 2013, they were 787, 786, 603, 523, 573, 629 after multiplied by 2000. Conclusion The BPNN model has high prediction accuracy, and it is suitable to predict short - term incidence of measles.