Engineering energy flexibility into buildings: development and evaluation of a model-predictive control strategy for demand response with space-heating

Energy flexibility of buildings has the potential to support deeper and more efficient decarbonisation of electricity systems, especially if aspirations to electrify heating become true. A promising technology to realise the flexibility potential in buildings is model-predictive control (MPC), where a model of a building and forecasts are used to determine optimal control decisions. This research aimed to evaluate the ability of a model-predictive space heating control strategy to use building thermal mass in delivery of energy flexibility for demand response programmes.

The Aim was achieved by developing versions of a MPC strategy for delivery of flexibility and implementing it in a simulation and physical environment. The simulation was a community-scale co-simulation of thirty dwellings, which was used to explore the innate factors that affected the delivery of a fixed demand reduction with the MPC. The physical environment was a matched pair test house facility. Experiments in the test houses were used to address the elements that the simulation lacked: the real-life uncertainties and their effects on the delivery of flexibility.

The results showed that the delivery of flexibility was dependent on the weather, physical features of the buildings, the control architecture and market environment. In both simulations and experiments, moderate weather conditions improved the flexibility potential. Weather also had an impact on the baseline error and thus its uncertainty. Weather variables not included in the MPC predictions, like solar irradiation or wind speeds, tended to cause errors in the MPC predictions. Elements impacting energy efficiency were found to affect the ability to deliver demand reductions. The houses in the simulations and experiments were typical English dwellings with limited levels of insulation and air-tightness, which led to relatively rapid changes in indoor air temperatures, on average 2.0-2.3\textdegree C in 40-50 minutes, when heating was turned off in the test house experiments.

The control architecture and implementation was found to affect the overall optimality of control. In the simulations, this led to a 44.9\% increase in peak power, when the centralised strategy with knowledge of all buildings was changed to a decentralised MPC. This was because only some of the houses in the community were used to deliver the demand reduction. The market environment affected the response of the MPC in two ways. First, the choice of the baseline was shown to be important, especially when considering uncertainty. The MPC projections were found to have the lowest uncertainty out of the baseline scenarios included in the analysis. For a day in January, the estimated uncertainty in 10-minute mean power was 23.8~\% of the total heater capacity. Second, varying the pricing in the MPC objective function had effects on the response pattern. Either the MPC performed a load-shifting pattern, or it shed load, which were two very distinct response strategies.