Jos Van Den Haspel Distributed Control of Refrigerators for the Smart Grid

Weather-dependent energy sources such as wind and solar power lack the flexibility to adjust their power supply in response to the demand. To facilitate the integration of these renewable sources the electricity demand needs to respond to the available electricity supply. Thermostatically controlled loads such as refrigerators and electric water heaters are useful for this due to their autonomous operation, thermal bu↵er property, and high aggregate power demand. The work in this thesis is aimed at controlling a large-scale population of these thermostatically controlled loads to ensure that the aggregate power demand of these appliances matches the available renewable power supply. Instead of a large centralized controller a distributed approach is chosen to provide robustness, modularity, and scalability. By using distributed model predictive control (DMPC), local controllers are able to anticipate on future events and to directly integrate local constraints such as user comfort levels. However, the majority of thermostatically controlled loads is constrained to binary on/o↵ inputs, complicating the DMPC optimization due to discrete ‘jumps’ in the negotiations among controllers. Existing DMPC approaches make use of additional heuristics to terminate the optimization when these jumps result in non-converging oscillations. The contributions of the work in this thesis are twofold. First, the potential of demand response using thermostatically controlled loads is investigated and translated into initial guidelines for a feasible match between the renewable supply and aggregate demand. Secondly, a serial implementation of feasiblecooperation DMPC is proposed to control the appliances. Closed-loop simulations show that unlike parallel DMPC methods from the literature such an approach converges towards the centralized MPC solution. Moreover, communication and computation requirements of the serial approach scale linearly with the amount of subsystems, without requiring a centralized coordinating agent.