Abstract
Demand response (DR) programs aim to engage distributed small-scale flexible loads, such as thermostatically controllable loads (TCLs), to provide various grid support services. Linearly Solvable Markov Decision Process (LS-MDP), a variant of the traditional MDP, is used to model aggregated TCLs. Then, a model-free reinforcement learning technique called Z-learning is applied to learn the value function and derive the optimal policy for the DR aggregator to control TCLs. The learning process is robust against uncertainty that arises from estimating the passive dynamics of the aggregated TCLs. The efficiency of this data-driven learning is demonstrated through simulations on Heating, Cooling & Ventilation (HVAC) units in a testbed neighborhood of residential houses.
Original language | English (US) |
---|---|
Article number | 106780 |
Journal | Electric Power Systems Research |
Volume | 189 |
DOIs | |
State | Published - Dec 2020 |
Keywords
- Linearly Solvable MDP
- Markov Decision Process
- TCL ensemble
- Thermostatically Controlled Loads
- Z-learning
ASJC Scopus subject areas
- Energy Engineering and Power Technology
- Electrical and Electronic Engineering