基于Holt-Winters指数平滑法的入院人次预测

Forecast the Number of Inpatients with Holt-Winters Exponential Smoothing Methods

目的应用时间序列Holt-Winters指数平滑法进行入院人次预测,为合理配置住院部资源提供科学的依据。方法运用SPSS22.0以2012年1月-2016年12月入院人次数据为基础建立Holt-Winters季节模型,此模型评价指标是预测误差(MAPE)和平均绝对误差(MAE),且预测误差的随机性采用Ljung-Box和画直方图来检查。采用2017年的实际数据进行建立模型的验证后,选择精确度较高模型预测2018年入院人次。结果在月度预测中选取α=0.001、β=0.803、γ=0.001时,得出MAPE=3.011%,实现较准确的预测。对模型进行白噪声残差分析,R^2=0.795,RMSE=213.998,标准化的BIC(L)=10.910,Ljung-Box Q(18)=20.293。结果表明估计具体模型是适合的,Holt-Winters算法较为准确地拟合了峰值,实现了吻合的预测效果。结论 Holt-Winters指数平滑法能很好地模拟入院人次在时间序列上的变动趋势,可对未来的入院人次进行预测,为医院诊疗工作的开展,资源的合理配置提供依据。

Objective To predict the number of hospitalized patients by using Holt-Winters exponential smoothing methods and provide scientific basis for rational allocation of inpatient resources. Methods Based on the data from 2012 to 2016,the Holt-Winters seasonal model was established by SPSS22.0. The evaluation index was the prediction error（MAPE） and the mean absolute error（MAE）,and the randomness of the prediction error was examined by Ljung-Box and drawing histogram.After validation of the model with the actual data in 2017,the model with high accuracy was selected to predict the number of hospitalized people in 2018.Results When selecting alpha= 0.001,beta =0.803 and gamma =0.001 in monthly prediction,we got MAPE=3.011% and achieved more accurate prediction.The white noise residuals of the model were analyzed,R^2=0.795,RMSE=213.998,standardized BIC（L）=10.910,Ljung-BoxQ（18）=20.293.It showed that the estimation model was suitable.The Holt-Winters algorithm accurately matched the peak value and achieved a consistent prediction effect.Conclusions The Holt-Winters exponential smoothing method could well simulate the changes in the time series of the hospitalized people,which could be used to predict the hospitalization time in the future,then provides the basis for hospital diagnosis and treatment work and the rational allocation of resources.