Unlocking the Potential of Model-Predictive Control in Non-domestic Building Energy Management: Automated Configuration and Optimisation of Control

Lead Research Organisation: University of Sheffield
Department Name: Architectural Studies

Abstract

Non-domestic buildings currently generate 18% of the UK's carbon emissions, >40% of which is due to space heating/cooling. Innovations in control are predicted to save > 25% of this figure making a sizeable contribution to the UK's carbon reduction target. Such innovations, however, must also address building refurbishment as well as new build since the rate of building replacement is rather low.

This proposal aims to effect a step change in the building energy management of non-domestic buildings by making model-predictive control (MPC) an economically-viable control technology for buildings. MPC is well-suited to controlling buildings, which typically have a large time delay between the application of an input and observable response. Importantly, MPC is based explicitly on optimisation as distinct from current building management systems, which use empirically-crafted rules. Currently, the generation and updating of the predictive dynamical model, which lies at the heart of MPC, has to be done manually by highly-skilled control engineers. Although MPC has been demonstrated to achieve energy savings >25% in non-domestic buildings in a research setting with hand-crafted predictive models, the cost of hand tuning these models is currently prohibitive for commercial deployment. We will use advanced machine learning techniques to automatically produce the predictive dynamical model from data acquired from the building-in-operation; hence we will capture key characteristics of the actual building, such as the profound influence of occupants on the building's energy performance. More than this, our approach will be able to periodically update the predictive model to accommodate the inevitable changes in building fabric, use and local environment, all of which will affect the building's thermal dynamics and hence its energy efficiency.

To cost-effectively demonstrate the potential of the research idea, we will make extensive use of building simulation by computer to explore a diverse range of different building forms, weather conditions, and modifications. We will also develop a stochastic model suitable for use in the simulation of non-domestic buildings allowing us to to better account for the impact of building occupants on energy performance. We will supplement these extensive simulation studies with a practical demonstration on different real buildings, one of which will be a passive design school; controlling peak temperatures in passive design buildings is a known challenge particularly in the summer months.

The overall outcome of this project will be to facilitate a step change in building control, producing an acceptable internal environment with the minimum use of primary, non-renewable energy. This derives from the fact that MPC is explicitly based on optimisation. As well as contributing to carbon reduction targets and greater workplace well-being, this will also produce significant cost savings for building operators.

We will disseminate knowledge about MPC and its potential contributions to carbon reduction to the UK building services community using a variety of channels, including an end-of-project workshop.

Publications

10 25 50
 
Title Multi-Agent Stochastic Simulation platform 
Description A prototype Multi-Agent Stochastic Simulation (MASS) platform has been developed. This is being deployed to generate data to help train the Model Predictive Control algorithms that are being developed within this project. This MASS platform will shortly be disseminated under open source licence agreement. 
Type Of Technology Software 
Year Produced 2017 
Impact This is the most sophisticated platform for simulated occupants' stochastic behaviours and their energetic impacts; this latter through co-simulation with a detailed building simulation programme called EnergyPlus.