A comparison of indoor temperature models for building demand response optimisation using MILP

Electricity consumption for heating, ventilation and air-conditioning (HVAC) is projected to keep growing, as the means improve and demand for indoor comfort increases. Rising HVAC demand poses challenges to grid management that demand response can mitigate, namely by inducing end-users to use building inertia to their advantage. Making the most of building inertia for demand response requires indoor temperature models that are simultaneously accurate and compatible with optimisation methodologies. Various models have been adopted for building demand response optimisation using mixed-integer linear programming (MILP) but it remains unclear which should be used for best results. We investigated the effect of building model complexity on accuracy and MILP solver performance in demand response problems. We relied on a MILP formulation compatible with state-space system representations to instantiate four indoor temperature models, all of which parameterised to represent the same building. Their accuracy was established via simulations and case studies were used to assess solver performances. We found accuracy and computational cost to be correlated with model order. Model selection processes should thus balance accuracy and computational performance for timely and informed decision-making within a given computational budget. Future efforts should be directed towards reducing the computational burden of adopting high order models.