IMT - AFA - DDRM
Responsable LIMOS : XIE XiaolanDébut du projet : 9 janvier 2022 - Fin du projet : 31 décembre 2024
Projet piloté par le LIMOS
Data-Driven Dynamic Resource Management for Random Time-Varying Demands in the Context of Covid-19 and future crises -- Capacity planning and resource management is a fundamental global challenge in the face of global crises such as the Covid-19 pandemic. One needs to understand the variabilities in time-varying demand for resources to hedge against the random demand fluctuations and take advantage of the risk-pooling effect. Covid-19 predictive and machine learning models used in capacity planning typically predict the mean values of the demands in both the temporal and spatial dimensions. However, they seldom provide reliable predictions or estimations of demand variabilities and, therefore, are insufficient for proper capacity planning. Some works focus on forecasting the evolution of the epidemic based on limited historical data and the resource needs over the days and weeks to come. However, the diversity of decisions, system dynamics over time, and demand uncertainties make decisions challenging even with good forecasts. During the crisis, regional or national resource pooling and allocation must be planned. Healthcare providers should identify their maximal capacity and decide how much capacity to share with others. National strategic reserve resources should also be planned, and emergency resources should be obtained from every organization. Once the capacity of the healthcare providers has been identified, and the sharing resources are allocated, the dynamic management of the resources must be put in place according to the epidemiological development. It is a question of managing the allocated capacity as well as possible according to the progression of the crisis and the limited available data and information. The number of confirmed cases, the number of hospitalizations (e.g., intensive care patients), the number of deaths, etc., are the primary demand data, which in fact may be underestimated during the crisis. There is also a time lag between the data collection and the actual crisis development, making demand prediction even more difficult. On the other hand, the number of places open for infected patients in intensive care and hospitalizations is the major supply data. These resources are counted with material, equipment, consumables, and personal support. The organization of staff working time must therefore be integrated into these decisions. Resources such as mobile resuscitation units from the military would be managed at this level. Using data analytics, machine learning predictive models, and data-driven optimization methods, it appears that playing on the management of resources according to the evolution of needs linked to the epidemic would make it possible to utilize better the limited resource and hedge against the temporal and spatial prediction uncertainties.
Organismes partenaires :
Financeur : None