EVA: Emergency Vehicle Allocation

• Autor(a) principal: Dias, Daniel Cacheira
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10400.22/23966
• topic: Emergency Medicine; Time Series Forecasting; Spatial Allocation; Engenharia de Inteligência Artificial

Emergency medicine plays a critical role in the development of a community, where the goal is to provide medical assistance in the shortest possible time. Consequently, the systems that support emergency operations need to be robust, efficient, and effective when managing the limited resources at their disposal. To achieve this, operators analyse historical data in search of patterns present in past occurrencesthat could help predict future call volume. This is a time consuming and very complex task that could be solved by the usage of machine learning solutions, which have been performed appropriately in the context of time series forecasting. Only after the future demands are known, the optimization of the distribution of available assets can be done, for the purpose of supporting high-density zones. The current works aim to propose an integrated system capable of supporting decision-making emergency operations in a real-time environment by allocating a set of available units within a service area based on hourly call volume predictions. The suggested system architecture employs a microservices approach along with event-based communications to enable real-time interactions between every component. This dissertation focuses on call volume forecasting and optimizing allocation components. A combination of traditional time series and deep learning models was used to model historical data from Virginal Beach emergency calls between the years 2010 and 2018, combined with several other features such as weather-related information. Deep learning solutions offered better error metrics, with WaveNet having an MAE value of 0.04. Regarding optimizing emergency vehicle location, the proposed solution is based on a Linear Programming problem to minimize the number of vehicles in each station, with a neighbour mechanism, entitled EVALP-NM, to add a buffer to stations near a high-density zone. This solution was also compared against a Genetic Algorithm that performed significantly worse in terms of execution time and outcomes. The performance of EVALP-NM was tested against simulations with different settings like the number of zones, stations, and ambulances.