Engineering Sustainable and Adaptive systems in Dynamic and Unpredictable Environments - School of Computing Seminar Series

Rui P. Cardoso, Imperial College London

Abstract
Electronic institutions are socially-inspired multi-agent systems, typically operating under mutually agreed rules, policies, and/or contracts. Previous work in the field, inspired by studies from social, political, and economic science, has exposed some crucial open questions regarding the ability of systems with hard-wired adaptation mechanisms to cope with unpredictable environmental variation. To address this, we investigate an innovative integration of existing socially-inspired approaches with biologically-inspired computing paradigms that provide insight into how populations of solutions can be evolved and adapted over multiple timescales in response to a dynamically changing problem space. The goal is to engineer electronic institutions which can remain sustainable over time in the face of unpredictable changes in the environment. Focusing on community energy systems in which several communities produce and consume energy, we develop a model of the system that can operator under different operating policies. We evaluate an offline method for evolving new operating policies via Genetic Programming, and an online method that uses Reinforcement Learning to adapt and improve existing policies. The results show that the evolved policies significantly outperform a benchmark policy, and provide novel insights into how evolutionary mechanisms can provide both innovation and adaptation within an electronic institution.

Bio
Rui Cardosa is currently an MSc student at Imperial College London. The talk is based on work undertaken for his MSc Dissertation in conjunction with Prof. Jeremy Pitt (visiting Prof. at Edinburgh Napier University) and Prof. Emma Hart and has been submitted to ISOLA 2018 in Cyprus, the 8th International Symposium On Leveraging Applications of Formal Methods, Verification and Validation