Abstract
Modern power systems integrate renewable distributed energy resources (DERs) as an environment-friendly enhancement to meet the ever-increasing demands. However, the inherent unreliability of renewable energy renders developing DER management algorithms imperative. We study the energy-sharing problem in a system consisting of several DERs. Each agent harvests and distributes renewable energy in its neighborhood to optimize the network's performance while minimizing energy waste. We model this problem as a bandit convex optimization problem with constraints that correspond to each node's limitations for energy production. We propose distributed decision-making policies to solve the formulated problem, where we utilize the notion of dynamic regret as the performance metric. We also include an adjustment strategy in our developed algorithm to reduce the constraint violations. Besides, we design a policy that deals with the non-stationary environment. Theoretical analysis shows the effectiveness of our proposed algorithm. Numerical experiments using a real-world dataset show superior performance of our proposal compared to state-of-the-art methods.
Introduction
As the global energy landscape shifts toward sustainable alternatives, the integration of Distributed Energy Resources (DERs)—such as localized solar arrays and wind turbines—has become a cornerstone of modern smart grids. While these resources offer a path to greener power, they introduce significant operational challenges due to their inherent volatility. Weather-dependent generation is notoriously unpredictable; solar output can plummet with passing clouds, and wind speeds can shift abruptly.
Managing these "fickle" resources requires a framework that can handle three primary hurdles:
- Uncertainty: Traditional power management models often rely on precise forecasts of generation and demand. In practice, these are moving targets that change in real-time.
- Information Gaps: In many decentralized systems, the exact mathematical relationship (the gradient) between a change in energy distribution and the overall system loss is unknown. Operators only receive "bandit" feedback—the final observed performance.
- Privacy Concerns: Prosumers (nodes that both produce and consume energy) are often reluctant to share detailed internal consumption or generation data with a central authority or neighboring nodes.
This paper addresses these challenges by proposing a distributed framework for energy sharing that optimizes network performance while respecting local constraints and user privacy.
Methodology: Bandit Convex Optimization and Adaptive Learning
The authors model the energy-sharing problem as a Bandit Convex Optimization (BCO) problem. In this setting, nodes must make decisions about energy distribution without knowing the underlying cost functions, relying instead on noisy observations of their performance.
Distributed Resource Sharing (DRS)
The core algorithm, Distributed Resource Sharing (DRS), is a gradient-free approach. Instead of calculating exact gradients, each node "probes" the environment by adding small random perturbations to its actions. To handle the physical reality that a node cannot distribute more energy than it possesses, the authors employ a regularized Lagrangian function. A critical innovation is the adjustment step, a projection mechanism that ensures all suggested actions are physically feasible without compromising the algorithm's theoretical convergence.
Meta-Algorithm for Non-Stationary Environments (MA-NSDRS)
Energy generation patterns are rarely static; they change with seasons, weather cycles, and time of day. To address this non-stationarity, the authors developed the Meta-Algorithm for Non-Stationary DRS (MA-NSDRS). This algorithm maintains a "pool of experts," where each expert is a DRS instance running with a different learning rate. By using an exponentially weighted average, the system can dynamically track and favor the expert that is currently performing best, allowing the grid to adapt to sudden shifts in energy availability.
Experimental Validation and Results
The proposed algorithms were validated using real-world data from the New York State Energy Research and Development Authority (NYSERDA). The dataset includes energy production records and geographical positions for solar photovoltaic (PV) systems in Staten Island. The researchers simulated networks of 6 and 20 nodes to evaluate scalability and efficiency.
Key Findings
- Theoretical and Practical Efficiency: The algorithms achieve a dynamic regret bound of $\tilde{O}(T^{3/4})$, ensuring they can keep up with the best changing strategy over time. In simulations, MA-NSDRS significantly outperformed state-of-the-art benchmarks like BanSaP, achieving lower cumulative loss and faster convergence (see Figure 4a).
- Constraint Satisfaction: The adjustment strategy proved essential. While standard optimization might "over-promise" energy, the proposed method effectively minimized constraint violations, ensuring grid stability (see Figure 4b).
- Optimized Energy Flow: The system successfully reallocated surplus energy from high-production nodes to those facing shortages, even across indirect network connections, thereby minimizing overall energy waste.
Conclusion and Future Work
This research provides a robust, privacy-preserving framework for managing the inherent volatility of renewable energy in distributed networks. By leveraging bandit feedback and adaptive meta-learning, the proposed system achieves state-of-the-art performance in dynamic environments.
Future research directions include improving the scalability of the meta-algorithm, which currently has a computational complexity of $O(T^2)$. Additionally, incorporating more complex power flow models—accounting for voltage drops and line impedance—and modeling correlations between neighboring nodes (e.g., shared weather patterns) could further enhance the practical utility of this framework in real-world smart grids.
Summary updated