应用ARIMA-GRNN模型预测深圳市宝安区流感样病例发病率

摘要

Objective To establish and verify an ARIMA-GRNN model to predict influenza-like illness incidence. Methods In-fluenza-like illness monthly incidence from Jan 2010 to Sep 2015 in Bao'an district was integrated to fit ARIMA model using SPSS 19. 0 software;MATLAB 7. 0 and GRNN model were used for error correction. Prediction accuracy was evaluated using monthly incidence from Oct 2014 to Sep 2015. Monthly incidence from Oct to Dec in 2015 was predicted by established model. Results The best fitting model was ARIMA(2,1,0)(0,1,1)12 model, upon combination with GRNN model, relative error between prediction and actual value was in the range of 2. 00% to 3. 12%. Conclusion The established ARIMA(2,1,0)(0,1,1)12-GRNN model was well integrated, the predicted short-term prevalence trend was consistent with actual incidence of influenza-like illness in Bao'an district.%目的：建立预测流感样病例发病率的ARIMA-GRNN模型,并验证其可行性。方法应用SPSS 19．0软件,对2010年1月-2015年9月宝安区流感样病例月发病率进行模型拟合,并应用MATLAB 7．0 联合GRNN模型进行误差修正。以2014年10月-2015年9月月流感样病例发病率作为考核样本评价模型预测效果,并预测2015年10-12月月流感样病例发病率。结果 ARIMA(2,1,0)(0,1,1)12拟合2010年1月-2015年9月宝安区每月新增感染率的变动趋势较为理想,联合GRNN模型后,预测结果与实际值相对误差最大值为3．12%,最小为2．00%；结论 ARIMA(2,1,0)(0,1,1)12-GRNN模型在宝安区流感样病例发病率短期趋势的预测与实际发病率吻合,拟合效果较好。