A Smart Disaster Management System for Future Cities Using Deep Learning, GPUs, and In-Memory Computing

Natural and manmade disasters have increased significantly over the past few decades. These include, among many others, the recent floods in Japan (June/July 2018) and Barcelona attack of August 2017. The Japan floods had left around 200 people dead, 70 were reported missing, and over eight million people were ordered to evacuate their homes, with 2bn USD the estimated cost of flood rebuilding. Disaster management plays a key role in reducing the human and economic losses. Our earlier work has focused on developing a disaster management system leveraging technologies including vehicular ad hoc networks (VANETs) and cloud computing to devise city evacuation strategies. The work was later extended to incorporate traffic management plans for smart cities using deep learning techniques. In-memory computations and graphics processing units (GPUs) were used to address intensive and timely computational demands of deep learning over big data in disaster situations. This paper extends our earlier work and provides extended analysis and results of the proposed system. A system architecture based on in-memory big data management and GPU-based deep learning computations is proposed. We have used road traffic data made publicly available by the UK Department for Transport. The results show the effectiveness of the deep learning approach in predicting traffic behavior in disaster and evacuation situations. This is the first system which brings together deep learning, in-memory data-driven computations, and GPU technologies for disaster management.