China’s Smart Grid Breakthrough: AI-Powered Switch Cuts Fault Currents by 99%
In an era defined by the electrification of everything—from vehicles to entire cities—the stability and intelligence of power distribution networks have become as critical as the batteries that fuel the transition. Now, a team of researchers from Fuzhou University has unveiled a next-generation control system for flexible multi-state switches (FMS) that could redefine how urban and rural grids manage surges, faults, and renewable energy integration. At the heart of this innovation lies a hybrid architecture fusing recursive radial basis function neural networks with sliding-mode control—a solution that not only slashes grounding fault currents by up to 99% but also dramatically enhances grid resilience against the volatility of solar and wind generation.
For global investors and energy executives watching China’s rapid infrastructure evolution, this development signals more than a technical upgrade. It represents a strategic pivot toward self-healing, AI-augmented distribution systems capable of supporting megawatt-scale electric vehicle (EV) charging clusters, industrial microgrids, and distributed renewable assets—all without costly hardware overhauls or reliance on centralized backup sources.
Unlike conventional circuit breakers or static transfer switches that react slowly or require manual intervention, the FMS functions as a real-time “traffic controller” for electricity. Positioned at the junctions of multiple feeders in medium-voltage distribution networks (typically 10 kV), it dynamically balances power flows, stabilizes voltage profiles, and—most critically—neutralizes dangerous arc faults within milliseconds. What sets the Fuzhou team’s approach apart is its dual functionality: the same device that enables seamless power sharing between overloaded and underutilized feeders also doubles as an active arc-suppression system during single-phase-to-ground faults, a common yet hazardous occurrence in ungrounded or high-impedance grounded grids.
Traditional arc-suppression coils (Petersen coils) have long been used to limit fault currents, but they are passive, slow, and ineffective against high-resistance faults. Modern active solutions often demand dedicated DC power supplies and operate only during emergencies, leading to low utilization rates and high capital costs. The new FMS architecture eliminates these drawbacks by repurposing its existing DC bus—normally used for inter-feeder power transfer—as the energy source for fault compensation. This “multi-mission” design dramatically improves asset efficiency while reducing footprint and lifecycle expenses.
The true breakthrough, however, lies in the control algorithm. Conventional proportional-integral (PI) controllers struggle with the nonlinearities and parameter uncertainties inherent in real-world grids, especially when renewable penetration exceeds 30%. They exhibit sluggish response, steady-state errors, and poor robustness during sudden load shifts or component aging. Sliding-mode control (SMC) offers superior disturbance rejection but suffers from “chattering”—high-frequency oscillations that degrade power quality and stress semiconductor switches.
To overcome this trade-off, the team led by Liao Jianghua engineered an improved recursive radial basis function neural network (RRBFNN) that learns and adapts in real time. Unlike static neural networks, the RRBFNN incorporates feedback loops from previous control cycles, enabling it to anticipate system behavior and smooth out abrupt transitions. Crucially, the network’s input layer integrates not just current tracking errors but also sliding surface dynamics and historical control outputs—creating a richer context for decision-making.
In simulations conducted on a three-port 10 kV FMS model using MATLAB/Simulink, the RRBFNN-SMC controller outperformed optimized PI controllers across every metric. During steady-state power exchange between feeders, total harmonic distortion (THD) in the AC-side current dropped to just 0.78%, compared to 3.06% under PI control. More impressively, the fluctuation ratio—a measure of power instability—was reduced by over 70% for both active and reactive power channels.
But the most compelling results emerged during fault scenarios. When a single-phase ground fault was simulated on Feeder 1 through a 100-ohm transition resistance—a realistic condition for tree-contact or degraded insulation—the FMS injected a precisely calculated zero-sequence current within 60 milliseconds. The residual fault current plummeted to 0.54 amperes, well below the 5-ampere safety threshold for reliable arc extinction. Across a range of fault resistances—from 10 ohms (near-bolted fault) to 3,000 ohms (high-impedance)—the system maintained a fault current suppression rate (FCRR) above 98%, consistently outperforming PI-based counterparts.
This level of performance isn’t just academically impressive; it has direct implications for grid safety and reliability. Uncontrolled arc faults are a leading cause of wildfires in dry regions and can trigger cascading outages in dense urban networks. By suppressing fault currents to near-zero levels, the FMS prevents thermal damage to cables, transformers, and switchgear—extending asset life and reducing maintenance costs. Moreover, because the system operates continuously—not just during faults—it provides constant voltage regulation and power balancing, smoothing the integration of rooftop solar and EV charging stations that would otherwise cause local voltage sags or swells.
From a policy standpoint, this technology aligns perfectly with China’s “New Infrastructure” initiative and its dual carbon goals. The State Grid Corporation has already deployed pilot FMS units in cities like Tianjin and Hangzhou, primarily for load balancing and capacity enhancement. Integrating advanced arc-suppression capabilities into these existing platforms could accelerate nationwide rollout without requiring new hardware standards or regulatory approvals.
For international observers, the significance extends beyond China’s borders. As Europe and North America grapple with aging grids and rising distributed generation, the FMS model offers a scalable blueprint for modernization. Unlike full-scale grid overhauls—which can cost billions—the FMS retrofits into existing feeder interconnection points, making it ideal for incremental upgrades. Its ability to function without a dedicated DC source also lowers the barrier to entry for utilities in emerging markets.
Critically, the control strategy minimizes dependence on precise system parameters—a major advantage in real-world deployments where line impedances, load profiles, and even temperature can drift over time. The RRBFNN’s adaptive weights continuously recalibrate based on observed behavior, ensuring consistent performance even as components age or environmental conditions change. This self-tuning capability is essential for maintaining reliability in remote or unmanned substations, where manual recalibration is impractical.
Looking ahead, the researchers acknowledge one limitation: their current model does not address dynamic transitions between operating modes (e.g., switching from PQ control to V/f mode during islanding events). Future work will focus on seamless mode-switching protocols and hardware-in-the-loop validation under more complex fault scenarios, including multi-point grounding and cyber-physical disturbances.
Nevertheless, the foundation has been laid. By merging neural intelligence with robust control theory, the Fuzhou team has created a system that is not only smarter but also more economical and safer. In a world where every kilowatt-hour counts and every second of uptime matters, such innovations are not merely technical achievements—they are enablers of the clean energy transition itself.
As global utilities seek to future-proof their networks against climate volatility, cyber threats, and the unpredictable demands of electrified transport, solutions like this FMS controller will likely become standard components of the 21st-century grid. And while the technology originated in a university lab in southeastern China, its principles are universally applicable—offering a compelling case study in how AI, when thoughtfully integrated into physical infrastructure, can deliver both resilience and sustainability.
Author: Liao Jianghua, Gao Wei, Tang Junyi, Yang Gengjie
Affiliation: College of Electrical Engineering and Automation, Fuzhou University, Fuzhou 350108, China
Journal: Electrical Engineering, Vol. 25, No. 5, May 2024
DOI: 10.19595/j.cnki.1008-2366.2024.05.002