关键词: 应急决策, 风险评估, 贝叶斯网络, 不精确概率, 证据理论

Abstract: After COVID-19 Pandemic broke out in China, China’s epidemic prevention and control policy has always adhered to scientific measures and precise prevention and control, which not only maximized the protection of people’s life safety and health, but also minimized the impact of the epidemic on economic and social development. The measures of regional classification control have achieved good results. However, the accuracy of the judgement is still open to question. In order to adjust and optimize the prevention and control measures according to the time and situation, it is of great significance to conduct research on the accuracy of regional risk level classification of COVID-19.
A Bayesian network decision-making model based on imprecise probability is proposed to evaluate the accuracy of regional risk classification of the new epidemic. First, the imprecise probability is used as the input parameter of the Bayesian network, that is, the key events in the epidemic are described with low probability and high probability. Then, by analyzing the regional risk level classification policy of COVID-19, we constructed a Bayesian network to represent the regional risk level classification process, and the mathematical description of the classification rule is carried out. Based on the analysis of the existing evidence collected in the epidemic prevention and control process, the imprecise probabilities of network nodes are extracted, and basic probability assignment functions of three states (true, false and uncertainty) are provided, so as to determine the nodes and range of nodes in a Bayesian network. Finally, an extended D-S (Dempster-Shafer) fusion technique is used to transfer the uncertainty of network nodes and to obtain joint probabilities on nodes from imprecise probabilities. The accuracy estimation of risk classification is obtained.
According to Prevention and Control of COVID-19 (Fifth Edition), we applied relevant data published by the media and clinical data of Central South Hospital of Wuhan University to our Bayesian network decision-making model. The results show that: The accuracy of regional risk classification of medium risk is higher, while those of low and high risk are low. The following suggestions are drawn based on above results: (1)The uncertainty of the assessment results in low-risk areas is significant. The vast majority of regions in the country belong to this risk level, with a large base, and the analysis should be particularly precise, otherwise small errors will bring great fluctuations. (2)The uncertainty of the assessment results in high-risk areas is the greatest. Although there are few high-risk areas designated, strict prevention and control of high-risk areas have a significant negative impact on the production and life of the people in the area and surrounding areas. Over prevention and control can also harm the phased achievements of the epidemic. (3)More evidence is needed to improve the accuracy of regional risk classification. On the one hand, the accuracy of the accuracy assessment system’s research results can be improved by improving the accuracy of the evaluation criteria; More evidence can be added to make the system’s rules more complete, especially for judgments of low and high risks. For example, improving the accuracy of confirmed cases through multiple nucleic acid tests. When epidemic prevention and control enter normalization, precision prevention and control should also focus on the economic and social activity order of low and high-risk areas, so as to meet the needs of scientific, professional, intelligent and refined emergency management.
Compared with fuzzy Bayesian network method and risk assessment method based on reliability theory, the proposal uses extended D-S evidence fusion technology to reduce the amount of calculation. And the information has been fully applied to solve the problem of knowledge uncertainty caused by data scarcity, data discontinuity, data incompleteness and prior ignorance. At the same time, the proposal can also solve the problem of information fusion when there are significant evidence conflicts, thereby improving the accuracy of decision-making. Our research focuses on the emergency management of public health events from another perspective, providing theoretical support for scientific decision-making and precise implementation. However, the regional risk classification of COVID-19 is a dynamic process. With the change of anti-epidemic situation, the adjustment and optimization of prevention and control measures can study the dynamic Bayesian network or the closed-loop dynamic Bayesian network topology.

Key words: emergency decision-making, risk assessment, Bayesian network, imprecise probability, evidence theory