Abstract:

In recent years， various epidemics （such as SARS in 2003， H1N1 in 2009， MERS in 2012， Ebola in 2014， etc.） have frequently occurred around the world， causing serious casualties， economic losses and social panic. After the outbreak， how to quickly isolate the infected persons and cut off the source of infection is an effective way to reduce the spread of the epidemic. In reality， the Chinese government has adopted a "multi-level quarantine" strategy in its response to COVID-19 to ensure that infected people are isolated as quickly as possible. On the one hand， the government converts gymnasiums into makeshift hospitals to accept mild patients， and on the other hand， the government also opens designated hospitals to take seriously ill patients. In this context， how to dynamically locate the two types of medical facilities according to the spread trend of the epidemic， and optimize the related resource and patient allocation problems integrally is the key to effectively control the epidemic.Since the 21st century， emergency facility location problems has been widely concerned and deeply studied by scholars. However， existing studies mainly focus on natural disasters such as earthquakes. Although fruitful results have been achieved， relevant models and algorithms are basically limited to the pre-disaster and post-disaster stages， and the commonly used two-stage stochastic planning and two-stage robust optimization are not applicable to the dynamic decision-making requirements of epidemic prevention and control. Compared with natural disasters， the evolution of the epidemic has dynamic and time-varying characteristics， which brings new challenges to the medical facility location problem.In this context， the medical facility location problem under epidemics is addressed as a multi-period and multi-type facility location problem with uncertain demand， as well as the integrated optimization problem of resource allocation and patient transportation. To solve this problem， the dynamic and time-varying characteristics of epidemic evolution are firstly considered， and the rolling horizon approach is used to divide the whole epidemic period into multiple discrete decision-making stages， so as to ensure that location strategies can be optimized and adjusted according to epidemic evolution timely. Secondly， in each decision stage， a robust location model is proposed and embedded into the rolling horizon framework after considering the random deviation between real demand data and the predicted ones. Thirdly， an algorithm is proposed and applied in the case study of COVID-19， which finally verified the validity of the model and the algorithm.

Key words: epidemic emergency, facility location, uncertain demand, rolling-horizon approach, robust optimization