Real-time Decision-making for Demand Response

Integrating renewable sources of generation to the current electric power system in high proportion requires significant transformations, amongst which increasing the flexibility of the system is key. Demand response programs aim to modulate the power consumption of end users to support the grid and can be leveraged to increase its flexibility.

In this presentation, we discuss past work on online optimization for fast timescale demand response of flexible loads under uncertainty. First, we formulate a setpoint tracking approach for uncertain flexible loads under different levels of feedback: full, bandit, partial bandit, and Bernoulli. These types of feedback can accommodate several communication and measurement settings, and thus the amount of available data the decision maker has access to. Building on our previous work and as part of an experimental implementation project, we propose a specialized online optimization algorithm for binary decision-making. This work is motivated by demand response of air conditioners and we present a demand response example where binary controls, lockout and temperature constraints are modelled. We then consider distributed decision-making for demand response. We formulate a dynamic and distributed online convex optimization algorithm with the tightest regret bound so far for this family of algorithms. We apply our approach to demand response of commercial buildings where the heating, ventilation, and air conditioning system's (HVAC) power consumption is adjusted on 4-second timescale to provide regulation services.