Show simple item record

AdvisorSalem, Saeed Mousa Ahmed
AdvisorBadawy, Ahmed
AuthorTLULI, REEM BASSAM
Available date2024-07-08T07:23:50Z
Publication Date2024-06
URIhttp://hdl.handle.net/10576/56498
AbstractEfficient allocation of ambulances is crucial for Emergency Medical Services (EMS) to respond promptly and deliver life-saving care on time. The challenge lies in the ambulance redeployment problem, which aims to devise optimal deployment strategies to minimize response times and maximize coverage in a given area. Traditional approaches to ambulance allocation problem rely on heuristics and predefined rules, often struggling to adapt to the dynamic nature of emergencies. In response, this thesis proposes a dynamic ambulance redeployment system to reduce ambulance response time, thus increasing the chances of saving lives. When an ambulance becomes available, the system recognizes it and intelligently reallocates it to the appropriate ambulance station, including ambulances that have come via patient transfers. By doing this, ambulance stations become more equipped to handle crises in the future. It is necessary to take into account several dynamic factors at each station concurrently due to the complexity of this operation. It is almost hard to control these elements with manual rules. We propose combining and prioritizing the dynamic factors of each station into a single score by means of a DNN, which we term the deep score network, in order to overcome this complexity. Through the utilization of DNN, we propose a Deep Reinforcement Learning (DRL) framework that efficiently trains the deep scoring network. Our dynamic ambulance redeployment algorithm is presented here for real-world applications based on this learning. Similarly, we apply the proposed framework on dynamic Charlie vehicle redeployment. Experimental results on real-world data from Qatar EMS show that our method clearly outperforms the stateof- the-art baseline methods. For example, for dynamic ambulance redeployment, the average response time of patients can be reduced by ⇠ 100 seconds (20%) with our proposed method, and the percentage of patients picked up within 10 minutes can be improved from 64.8% to 79.8%. As for the dynamic redeployment of Charlie vehicles, the average response time of critical patients can be reduced by ⇠ 125 seconds (13.33%) with our proposed method, and the percentage of critical patients treated within 10 minutes can be improved by approximately 11.08%.This improvement leads into more effective rescue operations for people in danger.
Languageen
Subjectambulance redeployment
deep reinforcement learning
deep neural network
emergency medical services
response time
TitleA DEEP REINFORCEMENT LEARNING-ENABLED DYNAMIC REDEPLOYMENT SYSTEM FOR MOBILE AMBULANCES IN QATAR
TypeMaster Thesis
DepartmentComputer Science and Engineering
dc.accessType Open Access


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record