Real-Time Power Balancing Via Decentralized Coordinated Home Energy Scheduling

Tsung-Hui Chang, Mahnoosh Alizadeh, Anna Scaglione

Abstract

As smart grids integrate more flexible loads, independent scheduling by Home Energy Management (HEM) systems often leads to "rebound peaks," where simultaneous load shifting to low-price periods strains grid reliability. This paper introduces CoHEM (Coordinated Home Energy Management), a decentralized framework designed to balance local demand and supply cooperatively. By formulating the problem as a multi-stage stochastic optimization and utilizing dual decomposition, CoHEM allows neighborhoods to reach a consensus on optimal scheduling without compromising user privacy. Simulations demonstrate a 53% reduction in peak power deviation and a 70-fold increase in communication efficiency compared to existing methods.

Introduction

Demand Side Management (DSM) is a cornerstone of modern smart grid strategy, aiming to align consumer electricity usage with grid capacity. However, traditional approaches face a critical "rebound effect": when utilities broadcast low-price signals, uncoordinated HEM systems may simultaneously schedule high-power tasks (e.g., electric vehicle charging) for the same window, creating new artificial peaks. Existing remedies, such as Direct Load Control (DLC), are often perceived as invasive by consumers, while dynamic pricing without coordination remains volatile and unpredictable.

The authors propose a decentralized coordination mechanism that respects user-defined deadlines and privacy. The goal is to ensure that the aggregate neighborhood load matches the available power supply through a "pseudo-price" signal that nudges behavior without requiring central oversight of individual household habits.

Methodology

The CoHEM framework treats the scheduling problem as a multi-stage stochastic optimization to account for the inherent uncertainty in appliance request timings.

Appliance Modeling

Deferrable loads, such as dishwashers, dryers, and Plug-in Hybrid Electric Vehicles (PHEVs), are modeled as Markov Decision Processes (MDP). Each appliance's state is defined by its current operation mode, remaining job length, and "slack" time—the window available before a hard deadline must be met.

State transition diagram for a single-profile appliance modeled as an MDP.

For more complex appliances, the model incorporates multiple operation profiles, allowing the system to choose between different power consumption patterns to better fit the grid's needs.

State transition diagram for an appliance with multiple selectable operation profiles.

Decentralized Algorithm and Privacy

To avoid a centralized "master controller" that would require access to private household data, the authors employ Lagrange dual decomposition. Homes exchange "pseudo-price" signals (dual variables) with immediate neighbors. This signal acts as a local coordination mechanism: if aggregate demand is too high, the pseudo-price rises, incentivizing HEMs to defer loads further.

Privacy is maintained because homes only share their expected aggregate load and the dual variables. No raw data regarding specific appliance usage or daily routines is transmitted to the utility or neighbors. The system uses a Consensus Subgradient Method to iteratively reach a neighborhood-wide optimal policy.

Experimental Setup

The framework was validated using synthetic datasets based on established energy models:

  • Appliance Profiles: Usage probability distributions for 18 residential appliance types were sourced from established literature (Paatero & Lund, 2006).
  • Market Data: Real-world Locational Marginal Prices (LMP) from Ameren (June 2012) were used to simulate realistic market conditions.
  • PHEV Specifications: Modeled as 3kW loads with random charging durations (1–6 hours) and variable arrival windows.

Key Results and Insights

The CoHEM algorithm demonstrates significant improvements over "selfish" (uncoordinated) scheduling:

  • Peak Reduction: In a 100-home simulation, uncoordinated systems increased power deviation to 4350.4 kW. CoHEM reduced this to 2002.7 kW, a 53% improvement.
  • Communication Efficiency: By using a stochastic approach rather than traditional Model Predictive Control (MPC), the system required 70x fewer message exchanges (3,000 vs. 213,750), making it highly scalable.
  • Economic Impact: When integrated with day-ahead procurement, the framework reduced total retailer costs by 28.6%.
  • Robustness: The system remained effective even with only 50% participation from the neighborhood.

Conclusion and Future Work

CoHEM provides a scalable, privacy-preserving solution to the rebound peak problem. By aligning individual incentives with grid stability through decentralized coordination, it offers a "win-win" for both utilities and consumers. Future research will focus on integrating thermostatically controlled loads (like HVAC), managing high price volatility, and incorporating home-level renewable generation and battery storage.

Summary updated

BibTeX
@article{chang2013real,
  author = {Tsung-Hui Chang and Mahnoosh Alizadeh and Anna Scaglione},
  title = {{Real-Time Power Balancing Via Decentralized Coordinated Home Energy Scheduling}},
  journal = {IEEE Trans. Smart Grid},
  year = {2013},
  doi = {10.1109/tsg.2013.2250532},
}
Figure