基于ARIMA-BP组合模型的全国卫生总费用预测  

Prediction of Total Health Expenditure in China Based on ARIMA-BP Combination Model

作　　者：王秋彤 王旭 范馨月 

机构地区：[1]贵州大学数学与统计学院,贵州 贵阳 [2]华北理工大学以升创新教育基地,河北 唐山

出　　处：《运筹与模糊学》2023年第3期1515-1524,共10页Operations Research and Fuzziology

摘　　要：目的:对卫生总成本的特征分析及变化趋势预测,能够为我国有关行政部门医疗卫生政策的法规制定、合理的卫生费用规划提供依据。方法:搜集我国1991~2020年卫生总费用统计数据通过方差倒数法构建ARIMA-BP组合模型。模型检验通过后,对2021年的卫生成本进行预测和评价。结果:将1991~2017年全国卫生总费用作为训练集,采用ARIMA(0,2,0)模型和BP模型拟合效果较好,通过滑动窗口方法生成新的数据样本后构建三层BP神经网络模型,结合方差倒数法,构建ARIMA-BP组合模型在2018~2020预测效果优于ARIMA(0,2,0)和BP,ARIMA模型、BP神经网络模型和ARIMA-BP组合模型的平均相对误差分别为:1.127%、1.052%、0.05%。结论:ARIMA-BP组合模型的预测效果为三种模型中最佳的,可通过该模型预测未来的卫生总费用,得到较为可靠的卫生总成本预算。Objective: The characteristic analysis and change trend prediction of the total health cost can provide a basis for the formulation of medical and health policies and reasonable health cost planning of relevant administrative departments in China. Methods: The statistical data of China’s total health expenditure from 1991 to 2020 were collected, and the ARIMA-BP combination model was constructed by the reciprocal of variance method. After the model test is passed, the health cost in 2021 is predicted and evaluated. Results: Taking the total national health expenditure from 1991 to 2017 as the training set, ARIMA(0,2,0) model and BP neural network model have good fitting effect. After generating new data samples through the sliding window method, a three-layer BP neural network model is constructed. Combined with the reciprocal of variance method, the ARIMA-BP combined model is constructed. The prediction effect of 2018~2020 is better than ARIMA(0,2,0) and BP. The average relative errors of ARIMA model, BP neural network model and ARIMA-BP combined model are 1.127%, 1.052%, 0.05% respectively. Conclusion: ARIMA-BP combined model has the best prediction effect among the three models. This model can predict the total health cost in the future and obtain a more reliable budget of the total health cost.

关 键 词：ARIMA模型 BP神经网络 ARIMA-BP组合模型 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]