Abstract
The implementation of optimal power flow (OPF) methods to perform voltage and power flow regulation in electric networks is generally believed to require extensive communication. We consider distribution systems with multiple controllable Distributed Energy Resources (DERs) and present a data-driven approach to learn control policies for each DER to reconstruct and mimic the solution to a centralized OPF problem from solely locally available information. Collectively, all local controllers closely match the centralized OPF solution, providing near optimal performance and satisfaction of system constraints. A rate distortion framework enables the analysis of how well the resulting fully decentralized control policies are able to reconstruct the OPF solution. The methodology provides a natural extension to decide what nodes a DER should communicate with to improve the reconstruction of its individual policy. The method is applied on both single- and three-phase test feeder networks using data from real loads and distributed generators, focusing on DERs that do not exhibit inter-temporal dependencies. It provides a framework for Distribution System Operators to efficiently plan and operate the contributions of DERs to achieve Distributed Energy Services in distribution networks.
Introduction
The rapid integration of Distributed Energy Resources (DERs), such as residential solar photovoltaics and electric vehicles, presents significant challenges for grid management, particularly regarding voltage stability and power flow regulation. The traditional approach to managing these systems is through Centralized Optimal Power Flow (OPF), which calculates optimal setpoints for all controllable devices to minimize system-wide losses while maintaining operational constraints.
However, centralized OPF imposes a significant "communication tax," requiring high-bandwidth, low-latency networks to exchange data between a central controller and every node in real-time. While decentralized heuristics exist, they often lack optimality and can lead to instability. This work addresses the fundamental challenge: achieving the performance of a global optimizer using only local, decentralized information.
Methods
The proposed framework treats decentralization as a data compression problem, where the goal is to reconstruct the optimal global solution using limited local information. The methodology follows a four-step pipeline:
- Data Generation: Thousands of centralized OPF simulations are performed using historical grid data to establish a "ground truth" dataset of optimal device settings.
- Information Analysis: Rate Distortion Theory is employed to quantify the information content of local measurements relative to the global grid state. This analysis identifies whether a DER has sufficient local data to make near-optimal decisions or if supplemental communication with neighboring nodes is required.
- Supervised Learning: A multiple stepwise linear regression model is trained for each DER. To ensure stability, the models utilize exogenous variables (e.g., local load and solar generation) as inputs, avoiding endogenous variables like local voltage that are directly influenced by the controller's actions.
- Deployment: The resulting policies are deployed as local open-loop controllers that predict optimal setpoints based on real-time local conditions.
The approach was validated using high-resolution power consumption and solar generation data from Pecan Street Inc., applied to a realistic 129-node Arizona feeder model and the standard IEEE 13-node test feeder for three-phase unbalanced networks.
Key Results
The data-driven framework demonstrates that decentralized ML policies can effectively mimic centralized OPF behavior with minimal performance loss.
- Near-Optimal Performance: The decentralized policies matched centralized OPF solutions with an average deviation of only 0.15%. As shown in the performance comparison, the regression-based approach significantly reduces system losses compared to "no control" or simple "constant power factor" heuristics.
- Constraint Satisfaction: Despite the lack of real-time global coordination, the controllers respected local capacity limits, with constraint violations occurring in only 0.00025% of validation tests.
- Emergent Collaboration: The models automatically learned location-specific "personalities." For instance, inverters at the end of long lines prioritized local voltage support, while those near substations focused on minimizing system losses.
Conclusion
This research provides a principled framework for decentralizing grid control by framing it as an information reconstruction problem. By leveraging machine learning to learn from centralized optimization, DERs can provide sophisticated grid services using only local data. Future work will address limitations such as static network topologies and the incorporation of energy storage systems, which introduce inter-temporal dependencies. Hybridizing these open-loop policies with feedback mechanisms could further enhance robustness against unmeasured grid disturbances.
Summary updated