New Optimization Method Enhances Efficiency of Multi-Zone Hybrid AC/DC Distribution Networks
In a significant leap forward for smart grid technology, researchers have unveiled a novel bi-level collaborative optimization scheduling method designed specifically for medium- and low-voltage alternating current/direct current (AC/DC) hybrid distribution networks with interconnected multi-zone architectures. This innovation promises to address critical challenges posed by the rapid integration of renewable energy sources and new types of electrical loads, such as electric vehicles (EVs) and 5G infrastructure, into modern power systems.
As the global push toward carbon neutrality intensifies, distribution networks are undergoing a fundamental transformation. Traditional radial, single-directional grids are giving way to dynamic, meshed, and multi-voltage-level systems capable of bidirectional power flow and real-time coordination. However, this evolution introduces new complexities—voltage fluctuations, power losses, and difficulties in balancing supply and demand across heterogeneous zones. The newly proposed method directly tackles these issues by enabling seamless cooperation between medium-voltage (MV) and low-voltage (LV) layers while leveraging the flexibility of energy storage systems (ESS) and vehicle-to-grid (V2G) technologies.
At the heart of this breakthrough is a dual-layer optimization framework that treats the MV and LV networks as interdependent yet distinct decision-making entities. The lower layer focuses on individual LV zones—each potentially hosting a unique mix of photovoltaic (PV) arrays, EV charging stations, 5G base stations, and stationary storage. Within each zone, the model determines the optimal charging and discharging schedules for ESS and V2G assets to minimize operational costs, which include electricity procurement, curtailed solar generation, network losses, and compensation for EV user participation.
Simultaneously, the upper layer operates at the MV level, where the primary objectives shift toward minimizing system-wide voltage deviations and total operational expenditures. This layer coordinates the power exchange among interconnected LV zones through voltage-source converters (VSCs), effectively acting as a traffic controller that redirects surplus energy from over-generating zones (e.g., those with abundant midday solar output) to deficit zones (e.g., those experiencing peak EV charging demand or high 5G load).
What sets this approach apart is its iterative, feedback-driven convergence mechanism. After an initial optimization pass in the LV layer, the resulting power profiles and voltage conditions are fed upward to inform the MV layer’s decisions. The MV layer then computes optimal inter-zonal power flows and sends updated boundary conditions—such as nodal voltages and injected power—back down to the LV zones. This cycle repeats until both layers reach a stable equilibrium, ensuring that local autonomy and global coordination are harmonized.
The research team validated their methodology using a modified IEEE 33-node test system, augmented with realistic LV subnetworks representing three distinct zones: one DC-based zone dominated by rooftop solar, one AC zone centered around EV charging infrastructure, and another AC zone hosting multiple 5G base stations. Each zone was equipped with localized storage and interconnected via 100 kVA VSCs. The simulation incorporated real-world data on solar irradiance, EV usage patterns (including stochastic arrival times and daily mileage distributions), and 5G load profiles tied to communication traffic intensity.
Three comparative scenarios were evaluated. In the first, each zone operated in complete isolation—no interconnection, no coordination. This baseline case resulted in significant solar curtailment (76 kW of wasted PV generation) and severe voltage violations, with node voltages in the solar-rich zone spiking to 1.076 per unit—well above the standard 1.05 pu limit. Network losses were high, and overall system efficiency suffered.
The second scenario introduced physical interconnections between LV zones via VSCs but maintained separate, non-coordinated optimization for MV and LV layers. While this allowed surplus solar power to be redirected to neighboring zones—reducing curtailment and lowering LV operational costs by nearly 30%—the lack of MV-level feedback limited the system’s ability to fully exploit its flexibility. Voltage stability improved but remained suboptimal, with noticeable fluctuations persisting during peak generation and demand periods.
The third and final scenario implemented the full bi-level collaborative framework. Here, the synergy between layers unlocked the system’s full potential. Solar curtailment was virtually eliminated, and the maximum allowable PV penetration in the solar zone increased by over 112% compared to the isolated case—rising from 40 kW to 85 kW without violating voltage constraints. Total LV operational costs dropped by 31.4%, while MV network losses decreased by 26.3%. Most impressively, the peak-to-valley load difference shrank by more than 25%, and the highest observed voltage was reduced by 5.85%, bringing the entire network well within safe operating margins.
Beyond economic and technical gains, the method enhances grid resilience and renewable energy hosting capacity. By enabling real-time redistribution of power across zones, it mitigates the risk of local overloads or under-voltage events. Storage and V2G assets are deployed not just for cost savings but as strategic buffers that smooth out intermittency and provide ancillary services. The 5G base stations—often viewed solely as inflexible loads—are integrated into the optimization framework through their empirically derived load models, ensuring their critical communication functions are never compromised while allowing their energy consumption to be shaped in response to grid conditions.
From a computational standpoint, the solution strategy is equally sophisticated. The LV layer problems, characterized by linear and convex constraints, are solved efficiently using second-order cone programming—a technique well-suited for handling AC/DC power flow equations under safety limits. The MV layer, however, presents a multi-objective challenge: minimizing both cost and voltage deviation. These competing goals are reconciled using a Pareto-optimal approach, where a genetic algorithm explores the solution space to generate a set of non-dominated trade-offs. A normalization and aggregation technique then identifies the most balanced compromise solution, ensuring fairness between economic and technical priorities.
This dual-solver architecture strikes an ideal balance between accuracy and scalability. While genetic algorithms can be computationally intensive, their application is confined to the relatively smaller MV layer, whereas the more numerous LV subproblems benefit from the speed and reliability of convex optimization. The iterative coupling ensures that the final solution is globally consistent without requiring a monolithic, intractable model.
The implications of this work extend far beyond the laboratory. As utilities worldwide grapple with the dual pressures of decarbonization and digitalization, scalable coordination mechanisms for heterogeneous distribution assets are no longer optional—they are essential. The proposed method offers a practical blueprint for grid operators seeking to unlock the full value of distributed energy resources without overhauling existing infrastructure. By retrofitting legacy feeders with smart VSCs and deploying advanced control algorithms, utilities can transform passive distribution networks into active, self-optimizing platforms.
Moreover, the framework is inherently extensible. Future iterations could incorporate probabilistic forecasting to handle the uncertainty of renewable generation and EV mobility, integrate demand response programs, or even coordinate with upstream transmission systems. The modular design also facilitates plug-and-play integration of new technologies—whether hydrogen electrolyzers, behind-the-meter batteries, or AI-driven load aggregators.
Critically, the research adheres to the highest standards of scientific rigor and transparency. All assumptions are clearly stated, models are grounded in established power engineering principles, and validation is performed against widely accepted benchmark systems. The inclusion of detailed cost structures—covering not just energy but also equipment degradation, user incentives, and converter losses—ensures that the results reflect real-world economics rather than theoretical ideals.
In an era where every kilowatt-hour counts and grid stability is paramount, this bi-level optimization method represents more than just an algorithmic improvement—it is a paradigm shift in how we conceive and operate the distribution grid. By bridging the gap between local autonomy and system-wide coordination, it paves the way for a more efficient, resilient, and sustainable power future.
Authors: Keyan Liu, Wanxing Sheng, Huiyu Zhan (Power Distribution Technology Center, China Electric Power Research Institute, Beijing 100192, China); Bo Tong, Lu Zhang, Wei Tang (College of Information and Electrical Engineering, China Agricultural University, Beijing 100091, China)
Published in: Electric Power Construction, Vol. 45, No. 1, January 2024
DOI: 10.12204/j.issn.1000-7229.2024.01.004