A processing architecture for real-time predictive smart city digital twins

Smart Cities will create many benefits for society. One of the most significant will be real-time, active management of the systems making up a Smart City. Active management will be enabled by processing large amounts of real-time data, by machine learning to deliver timely insights and by large-scale simulation to create effective interventions. To realise active management of Smart Cities requires processing systems that can take in data at large volume and variety, and perform complex processing on these data, all in real-time. Digital twins are used to create a continuous, consistent view of the current state of the city, and simulation of the evolution of digitals twins is used to detect problems arising in the near future and to optimise the impact of interventions. In this paper we propose and validate a Smart City processing architecture that can deliver these benefits by building a digital twin of a city, combining real-time streamed data ingestion, simulation and analytics. Streamed data systems can deliver a real time response with large amount of incoming data but are not suited to time advancing simulations where the interaction between modelled objects is important. We propose a processing architecture that mitigates this limitation by separating object interactions from streamed data processing. Finally, we introduce DEDUCE-PT, a demonstration implementation of this Smart City processing architecture that is aimed at proactive management of public transport systems.