# Harnessing Flexible and Reliable Demand Response Under Customer Uncertainties

## TL;DR

This paper develops a stochastic optimization framework for demand response that accounts for uncertainties in renewable energy, customer demands, and costs, proposing online control policies and distributed algorithms to improve reliability and efficiency.

## Contribution

It introduces a joint capacity planning and demand response design model with uncertainty considerations, along with online control policies and distributed algorithms for practical implementation.

## Key findings

- Achieves near-optimal social costs with proposed algorithms.
- Significant social cost savings over baseline methods.
- Enhanced policies allow flexible commitment levels.

## Abstract

Demand response (DR) is a cost-effective and environmentally friendly approach for mitigating the uncertainties in renewable energy integration by taking advantage of the flexibility of customers' demands. However, existing DR programs suffer from either low participation due to strict commitment requirements or not being reliable in voluntary programs. In addition, the capacity planning for energy storage/reserves is traditionally done separately from the demand response program design, which incurs inefficiencies. Moreover, customers often face high uncertainties in their costs in providing demand response, which is not well studied in literature. This paper first models the problem of joint capacity planning and demand response program design by a stochastic optimization problem, which incorporates the uncertainties from renewable energy generation, customer power demands, as well as the customers' costs in providing DR. We propose online DR control policies based on the optimal structures of the offline solution. A distributed algorithm is then developed for implementing the control policies without efficiency loss. We further offer enhanced policy design by allowing flexibilities into the commitment level. We perform real world trace based numerical simulations. Results demonstrate that the proposed algorithms can achieve near optimal social costs, and significant social cost savings compared to baseline methods.