【摘要】 目的 比较有无协变量的ARIMA模型在大连市青年学生HIV感染数预测中的效果。方法 首先对网络、性病序列和HIV序列进行单因素相关分析，以了解网络、性病序列和HIV感染数序列是否存在相关关系及滞后关系，其次，以分析得出的关系最强且具有预测意义的变量和HIV感染数作为基础数据，建立包含协变量的ARIMA模型和不包含协变量的一般ARIMA模型，并进行前瞻性预测，以平均绝对误差作为评价指标比较两种模型的预测效果。结果 2013—2021年期间大连市共报告青年学生HIV感染者841例。单因素相关分析结果显示，2013—2018年百度指数中关键词艾滋病的搜索指数与当月HIV感染数呈显著正相关关系(r=0.302,P=0.006),淋病与滞后2个月的HIV感染数呈显著负相关关系(r=-0.250,P=0.024)。考虑到预测价值，最终选择以淋病发病数据和HIV感染数据作为基础数据，建立了纳入协变量的ARIMA模型和无协变量的一般ARIMA模型，并运用上述模型分别对2019年至2021青年学生HIV感染数进行预测，平均绝对误差分别为17.62%、66.17%。结论 对于大连市青年学生的HIV感染数预测，与无协变量的一般ARIMA模型相比，更适合联合运用性病发病数以及HIV感染数所建立的ARIMA模型，但模型的平均绝对误差较大，在今后的研究中仍需进一步改进。更多还原

【Abstract】 Objective To compare the effect of ARIMA models with and without covariates in predicting the number of HIV infections among young students in Dalian. Methods First, univariate correlation analysis was performed on the network, STD sequence and HIV sequence to understand whether there was a correlation and lag relationship between them.Secondly, variables with the strongest correlation and predictive value and HIV infection numbers were used as the baseline data to establish an ARIMA model with covariates and a general ARIMA model without covariates, and to predict the HIV number from 2019 to 2021.The average absolute errors were used as evaluation indexes to compare the prediction effects of the two models. Results A total of 841 cases of HIV infection among young students were reported in Dalian from 2013 to 2021.The results of univariate correlation analysis showed that the search index of the keyword AIDS in the Baidu Index in a given month from 2013 to 2018 was significantly positively correlated with the number of HIV infections in that month(r=0.302,P=0.006),and gonorrhea was negatively correlated with the number of HIV infections with a lag of 2 months(r=-0.250,P=0.024).Using gonorrhea incidence number and HIV infection number as the basic data, an ARIMA model with covariates and a general ARIMA model without covariates were established to predict the number of HIV infection among young students from 2019 to 2021,and the average absolute errors were 17.62% and 66.17%,respectively. Conclusion Compared with the general ARIMA model without covariates, the ARIMA model based on the combined use of STD incidence and HIV infection is more suitable for predicting the number of HIV infections among young students in Dalian, but the average absolute error of the model is still large, which needs further improvement in the future research.更多还原