论文部分内容阅读
目的分析气象因素与海南省万宁市疟疾发病率的相关性,比较BP神经网络模型和逐步回归模型对疟疾发病率的预测效果。方法收集1995年1月—2007年12月万宁市每月气象数据和疟疾发病率数据,应用Spearman等级相关分析方法分析气象因素与疟疾发病率之间的相关性,分别用BP人工神经网络方法和逐步回归方法建立疟疾发病率的气象因子拟合模型,预测2008年各月的疟疾发病率。结果万宁市疟疾月发病率与前1个月的平均气温、最高气温、最低气温、降雨量、日照时间均呈正相关(均P<0.05),与前1个月的平均相对湿度、平均气压均呈负相关(均P<0.01);将7种气象因素作为输入变量,疟疾发病率作为输出变量,构建内含1个隐含层的BP神经网络模型,在隐单元数为16时拟合效果最优,经过300次训练达到设定的最小训练误差为0.001,模型的均方误差和决定系数R~2分别为0.002 7和0.99;将7种气象因素作为自变量,疟疾发病率作为因变量构建逐步回归模型,进入模型的变量为平均气温和平均相对湿度,模型的决定系数R~2为0.40;应用2种模型对2008年各月疟疾发病率进行预测,平均绝对误差分别为1.24/10 000和0.44/10 000。结论万宁市疟疾发病率与气象因素明显相关,利用气象因素构建的BP神经网络模型较逐步回归模型具有更好的发病率拟合效果,但逐步回归模型的预测效果更好,BP神经网络模型的泛化能力需要进一步提高。
Objective To analyze the correlation between meteorological factors and the incidence of malaria in Wanning City, Hainan Province, and to compare the prediction effect of malaria incidence with BP neural network model and stepwise regression model. Methods Monthly meteorological data and malaria incidence data of Wanning City from January 1995 to December 2007 were collected. Spearman rank correlation analysis was used to analyze the correlation between meteorological factors and incidence of malaria. BP artificial neural network And a stepwise regression method to establish a meteorological factor fitting model for the incidence of malaria and predict the incidence of malaria in each month of 2008. Results The monthly incidence of malaria in Wanning was positively correlated with the mean temperature, the highest temperature, the lowest temperature, the rainfall and the sunshine duration in the previous month (all P <0.05), and the average relative humidity, average pressure (All P <0.01). Seven meteorological factors were used as input variables and malaria incidence as output variables. A BP neural network model with one hidden layer was constructed. When the number of hidden units was 16, The best training effect was achieved. After 300 trainings, the minimum training error was 0.001, and the mean square error and the coefficient of determination R ~ 2 were 0.002 7 and 0.99 respectively. Taking seven meteorological factors as independent variables and malaria incidence as a factor The variables were used to construct the stepwise regression model. The variables entering into the model were average temperature and average relative humidity, and the coefficient of determination R ~ 2 was 0.40. Two models were applied to predict the incidence of malaria in each month of 2008 with the average absolute error of 1.24 / 10 000 and 0.44 / 10 000. Conclusion The incidence of malaria in Wanning is significantly related to the meteorological factors. The BP neural network model constructed by using meteorological factors has a better incidence rate fitting effect than the stepwise regression model, but the stepwise regression model has a better prediction effect. The BP neural network model The generalization ability needs to be further improved.