Online Learning for Network Constrained Demand Response Pricing in Distribution Systems
For distribution system operators, it provides a data-driven method to price demand response without restrictive assumptions on uncertainty distributions, but results are incremental as they rely on numerical experiments without SOTA comparisons.
This paper develops an online learning approach for demand response pricing in distribution systems that continuously estimates price sensitivities and integrates with distributionally robust chance-constrained optimal power flow to ensure desired DR capacity, co-optimizing DR and conventional generation. Numerical experiments demonstrate effectiveness.
Flexible demand response (DR) resources can be leveraged to accommodate the stochasticity of some distributed energy resources. This paper develops an online learning approach that continuously estimates price sensitivities of residential DR participants and produces such price signals to the DR participants that ensure a desired level of DR capacity. The proposed learning approach incorporates the dispatch decisions on DR resources into the distributionally robust chance-constrained optimal power flow (OPF) framework. This integration is shown to adequately remunerate DR resources and co-optimize the dispatch of DR and conventional generation resources. The distributionally robust chance-constrained formulation only relies on empirical data acquired over time and makes no restrictive assumptions on the underlying distribution of the demand uncertainty. The distributional robustness also allows for robustifying the optimal solution against systematically misestimating empirically learned parameters. The effectiveness of the proposed learning approach is shown via numerical experiments. The paper is accompanied by the code and data supplement released for public use, see [27].