Reinforcement Learning And Social Based Approach to Post-disaster Reconstruction Planning

Natural disasters always have several effects on human lives such as in the form of causalities and destruction of the built environment. These kinds of situations always become challenging for the governments to tackle these incidents and to rebuild the economic, social, and physical infrastructures and facilities with the available resources, more specifically, in the defined budget and time. The governments always define plans and policies in accordance with the law and political strategies to reconstruct damaged infrastructure (buildings, roads, and bridges) that should maximize the social benefits of the affected community. Due to the severity of the damage, for instance to assess all the needs of the involved citizens, private companies, and public institutions, the plans and policies definition is a critical and difficult task. That’s why a huge amount of resources is always required to bring life back to normality, which makes reconstruction very challenging for all responsible stakeholders. To this end, in this thesis, we develop an approach (REPAIR) to decision-support system by using deep reinforcement learning technique (Double Deep Q-Network) for the planning of post-disaster city reconstruction by considering available resources, meeting the needs of the broad community stakeholders (like citizens’ social benefits and politicians’ priorities) and keeping in consideration city’s structural constraints (like dependencies among roads and buildings). In particular, from enriched GIS data, REPAIR elaborates a graph-based representation of the considered area and runs Double Deep Q-Network (DDQN) algorithm to generate a set of alternative reconstruction plans by satisfying the posed requirements. The generated plans are then provided to decision-making for the selection of which one to actuate. To check the applicability of the whole approach, we applied it on two real use-cases, i.e. the historical center of Sulmona city (Italy) and L’Aquila city (Italy), using detailed GIS data and information on the urban land structure and buildings vulnerability. The proposed approach for post disaster reconstruction planning (REPAIR) is generic, determines a set of alternative plans for local administrators who select the ideal one to implement, and it can be applied to areas of any extension as long as the decision-makers share the same goals. The described approach is comprised of a more general data science framework which is developed to produce an effective response to natural disasters in post-disaster reconstruction planning. Keywords: Decision-support System; Natural Disaster; Deep Reinforcement Learning; Social Benefits; City Reconstruction Planning; Data Science, Geographical Information Systems.