Abstract
Distributed energy storage and flexible loads are essential tools for ensuring stable and robust operation of the power grid in spite of the challenges arising from the integration of volatile renewable energy generation and increasing peak loads due to widespread electrification. This paper proposes a demand-side management policy to coordinate self-interested energy prosumers based on receding horizon games, i.e., a closed-loop receding-horizon implementation of game-theoretic day-ahead planning. Practical stability and recursive constraint satisfaction of the proposed feedback control policy is proven under symmetric pricing assumptions using tools from game theory and economic model predictive control. Our numerical studies show that the proposed approach is superior to standard open-loop day-head implementations in terms of peak-shaving, disturbance rejection, and control performance.
Introduction
As the global energy landscape shifts toward volatile renewable sources like solar and wind, and electrification increases through electric vehicles and heat pumps, the power grid faces significant "peak load" challenges. Upgrading physical infrastructure is often prohibitively expensive. Consequently, researchers are focusing on Demand-Side Management (DSM)—incentivizing users to shift their energy consumption to off-peak hours.
Modern energy users are increasingly "prosumers"—consumers who also produce and store energy via batteries and solar panels. These agents are self-interested, seeking to minimize their own costs rather than optimize for the grid. Traditional game-theoretic DSM models typically rely on "day-ahead" planning, an open-loop approach where agents commit to a 24-hour plan at midnight. This approach suffers from two major drawbacks: "end-of-day effects," where prosumers dump stored energy at the end of the horizon to save money, creating new artificial peaks; and inflexibility, as static plans cannot adapt to mid-day disturbances like sudden cloud cover or equipment failure.
Methodology: Receding Horizon Games
The authors propose a Receding Horizon Game (RHG) framework to address these issues. Instead of a single daily optimization, the system solves a Generalized Nash Equilibrium (GNE) problem at every time step over a moving future window, effectively "closing the loop."
Each prosumer is modeled as a linear time-invariant system with states representing battery State-of-Charge (SoC) and "energy debt" (flexible load requirements). The grid's physical limits are represented as coupling constraints, meaning one player's feasible actions depend on the collective choices of others. The study targets "variational" GNEs (v-GNEs), which ensure economic fairness by applying the same marginal cost for grid constraints to all prosumers.
The framework was validated using high-resolution consumption and solar data from the Pecan Street Dataport (New York, 2019). The simulation involved 10 active prosumers and 5 passive consumers, utilizing lithium-ion battery models and time-of-use pricing.
Key Insights: Stability and Fairness
A critical discovery in this work is the connection between these games and Potential Games. Under symmetric pricing—where all prosumers in a neighborhood pay the same rate—the complex game behaves like a single optimization problem. This allows the authors to leverage stability proofs from Economic Model Predictive Control (MPC) to guarantee that the system is "recursively feasible" and will converge to a stable equilibrium.
Furthermore, by shifting to a receding horizon, the "end-of-day" battery depletion effect is eliminated. Since prosumers are always looking 24 hours ahead, there is never a "final hour" that incentivizes emptying the battery, leading to more consistent grid stabilization.
Results and Performance
The RHG framework demonstrates significant improvements over traditional methods in peak shaving and resilience.
Numerical studies revealed that the RHG scheme reduced peak aggregate load by 49%, outperforming standard day-ahead optimization by 6%. Beyond peak shaving, the system excels at disturbance rejection. In a simulated contingency scenario where grid capacity dropped by 60% for five hours, the RHG successfully shifted loads in real-time to maintain safety limits. In contrast, open-loop day-ahead schemes failed to adapt, leading to constraint violations.
Limitations and Future Outlook
While the paper provides the first formal proof of closed-loop stability for constrained DSM games, it assumes symmetric pricing. Future research will investigate non-symmetric pricing models where different prosumers may have different contracts. Additionally, moving toward fully distributed computation for "smart meter" environments and incorporating stochastic forecasts to handle weather uncertainty are key next steps for making RHG practical for real-world grid management.
Summary updated