基于神经网络的气象条件对泸州市肺结核发病率预测

Application of Neural Network in Forecasting Tuberculosis Caused by Meteorological Factors

在选取2006-2010年泸州市防疫站呼吸道疾病每个月发病人数和相对应的泸州市气象数据分析基础上,对肺结核患病人数的数据进行标准化,采用预报因子的模糊优选方法选择肺结核易发率预报的气象因子。应用3层BP神经网络,将3个月前月平均气压、3个月前月平均温度、2个月前月平均最低温度、3个月前月平均日温差和1个月前温度变化作为前馈神经网络输入层节点,以发病人数的标准化值作为网络输出,建立肺结核易发率5个输入、5个隐含节点和1个输出神经网络模型结构。模型设定训练精度为0.005,迭代后的精度为0.00495,研究结果有助于提高医疗气象预报服务水平。

Based on monthly incidence data of the respiratory diseases of the CDC of Luzhou city,Si Chuan province,from which the monthly number of pulmonary tuberculosis patients then is standardized,and the corresponding meteorological data, the meteorological predictors for tuberculosis prone rate are chosen by using the Fuzzy optimization method.The monthly average of air pressure of 3 months ago,temperature of 3 months ago,minimum temperature of 2 months ago,monthly average of daily range temperature of 3 months ago and temperature variation of one month ago are applied to 3-layer BP neural network as the standard feed-forward neural network input layer nodes.Meanwhile the standardized incidence of the following month is as the network output data.The neural network model of tuberculosis prone rate,whose structure is 5-5-1,i.e.5 input nodes,5 hidden nodes and 1 output node,is then established with training accuracy of 0.005 and post-iteration accuracy of 0.00495.The results are useful for improving the forecasting service level of medical meteorology.