Swarm Behavior to Mitigate Rebound in Air Conditioning Demand Response Events

Thermostatically Controlled Loads (TCLs) have shown great potential for Demand Response (DR) events. However, it has been commonly seen that DR events using TCLs may cause demand rebound, especially in homogeneous populations. To further explore the potential for DR events, as well as the negative effects, a stability and resilience analysis were performed on multiple populations and verified with agent based modeling simulations.At the core of this study is an added thermostat criterion created from the combination of a proportional gain and the average compressor operating state of neighboring TCLs. Where DR events in TCLs are commonly controlled by set point manipulation, the modified thermostat behavior proposed in this study alters the effective dead band of each individual TCL. Previous work has shown the effectiveness of the proposed behavior to mitigating the demand rebound.By adding the average operating state of neighboring TCLs and a proportional gain, the systems feedback is changed, opening the possibilities to creating an unstable response. Stability limit are found from linearized systems, differing in delay schemes and connection architecture. The stability analysis was verified through agent-based modeling simulations on MATLAB. The linearization assumption was tested by simulating the systems while altering the parameters of population size and thermostat dead band.Resilience of several systems, differing in connection architecture, is computed and compared to results of a simulated denial of service attack on the system. Resilience for each architecture was calculated using the algebraic connectivity of the graph. The simulated attack is completed by removing the TCLs ability to communicate with in the agent based model.The stability analysis showed the effect of the gain value on the performance of the system and that the stability limit was directly affected by the effective deadband. As the deadband size was increased, the predicted results found from the analysis aligned with simulations of the system. Contrarily the resilience analysis was validated by simulations with smaller deadband sizes. Simulations of cyber-attacks also showed optimal attacks based on operating state of thermostats, as well as locations within the population.