A Multi-Agent Reinforcement Learning (MARL) Decision-Support tool for Energy-Efficient Incremental Residential Development

Incremental housing, a system in which a house is unfinished but in conditions of habitability and is upgraded at a pace based on the financial capacities of its dwellers, is the most common form of residential development in the global south. Its nature entails risks for both dwellers and communities, so several aid programmes have emerged to minimise them while further providing low-cost housing. Although improvement is usually achieved, energy-related environmental variables tend to be neglected. A more holistic approach is needed so that the social and environmental benefits of these programmes are maximised. In this context, this thesis proposes an agent-based modelling (ABM) workflow for social simulation whose objective is to help aid agencies pursue net-zero incremental neighbourhoods while considering the complex and highly dynamic socio-economic interactions present in this building development strategy. By following the exploratory modelling paradigm and a post-rationalist approach to planning theory, the model explored plausible outcomes of policy implementation and agent interaction scenarios in incremental housing neighbourhood development using Peru as a case study. The focus of this initial approach was on the operational energy use of buildings, leaving for further development other aspects that would allow reaching net-zero neighbourhoods.