Behavioural data-driven coalitional control for buildings
Lead Research Organisation:
Imperial College London
Department Name: Electrical and Electronic Engineering
Abstract
Buildings are responsible for about 40% of carbon emissions and consume about 40% of all produced energy in the UK. Transforming how buildings use and produce energy is a fundamental steppingstone to achieving net-zero carbon emissions and sustainable economic growth. The abundance of data, flexible technologies and advanced control approaches open exciting opportunities to achieve cost-effective system decarbonisation and create places where people love to live for the increased comfort standards. A radical transformation of the building sector is possible using real-time monitoring, learning capabilities, advanced control strategies, distributed optimisation and coordination.
Our research demonstrates that the energy consumption of buildings has a vast potential to be flexible and support an efficient grid operation. However, it is unclear how to design distributed control architectures and schemes managing millions of buildings in real-time to simultaneously achieve societal and individual consumer benefits. The proposed project seeks answers to critical open questions: How can we efficiently harness the adaptability of millions of diverse buildings to support the entire energy system while optimizing individual objectives concurrently? How can we harness data reliably to develop scalable, transferable control methods, bringing them closer to practical application?
The aim of this research is to develop distributed solutions to reliably manage energy use across groups of buildings. We will consider for the first time the advantage of dynamically forming coalitions according to the environment's variability and individual real-time energy needs. To realise this, we have set the following objectives:
1. Extend the latest data-driven behavioural control and uncertainty modelling approaches, state-of-the-art distributed optimisation methods and reinforcement learning techniques. These methods should be scalable to bridge the gap between lab-scale demonstrations and real-world implementation.
2. Apply these innovative methods to models of building clusters. This will offer insights for shaping policies and driving innovation, bolstering their role in supporting the entire energy system.
The close collaboration with UK Power Networks and SSE Energy Solutions will support the data-driven modelling and development of novel adaptive distributed control architectures to maximise the research output impact.
A pressing question we will address is how to achieve both individual and societal benefits. Existing distributed solutions are focused on directly achieving a centralised objective. Such solutions do not fit the objectives of simultaneously achieving societal and individual objectives. Substantial performance limitations arise when pursuing exclusively conflicting objectives, since the buildings connected to the grid are strongly coupled.
Our research demonstrates that the energy consumption of buildings has a vast potential to be flexible and support an efficient grid operation. However, it is unclear how to design distributed control architectures and schemes managing millions of buildings in real-time to simultaneously achieve societal and individual consumer benefits. The proposed project seeks answers to critical open questions: How can we efficiently harness the adaptability of millions of diverse buildings to support the entire energy system while optimizing individual objectives concurrently? How can we harness data reliably to develop scalable, transferable control methods, bringing them closer to practical application?
The aim of this research is to develop distributed solutions to reliably manage energy use across groups of buildings. We will consider for the first time the advantage of dynamically forming coalitions according to the environment's variability and individual real-time energy needs. To realise this, we have set the following objectives:
1. Extend the latest data-driven behavioural control and uncertainty modelling approaches, state-of-the-art distributed optimisation methods and reinforcement learning techniques. These methods should be scalable to bridge the gap between lab-scale demonstrations and real-world implementation.
2. Apply these innovative methods to models of building clusters. This will offer insights for shaping policies and driving innovation, bolstering their role in supporting the entire energy system.
The close collaboration with UK Power Networks and SSE Energy Solutions will support the data-driven modelling and development of novel adaptive distributed control architectures to maximise the research output impact.
A pressing question we will address is how to achieve both individual and societal benefits. Existing distributed solutions are focused on directly achieving a centralised objective. Such solutions do not fit the objectives of simultaneously achieving societal and individual objectives. Substantial performance limitations arise when pursuing exclusively conflicting objectives, since the buildings connected to the grid are strongly coupled.
