Machine learning-based strategy for demand response in distribution systems using thermostatically controlled loads

Thermostatically controlled loads (TCLs) show a great potential for demand side management (DSM) in smart grids. They can be aggregated and controlled autonomously, i.e. without the need for communication. A Space Heating (SH) load is a typical TCL with a heat storing capability that is a great demand response (DR) resource - a fast response option at low cost. In this paper, an autonomous DR control scheme is proposed to alleviate voltage violations of the local distribution system by managing SH loads. The proposed scheme attempts to set the SH operating point intelligently based on local measurements only. The proposed scheme bases its decisions on the voltage at the premise and the premise's voltage-to-power sensitivity (measured locally). By deploying machine learning, the proposed control scheme ensures that multiple SH loads, located at different nodes on the distribution system, contribute in a fair manner to mitigate voltage violations in the system without the need for communication. This controller also ensures an acceptable comfort level for the end-users. Simulation results verify the efficacy of the proposed control scheme in revamping voltage profile while maintaining fairness among and comfort of end-users.