Synapse-Inspired Energy Networks: A Neuromorphic Approach to Microgrid Protection without Communication Links

Abstract

Traditional protection systems for microgrids, which rely on high fault currents and continuous communication, struggle to keep up with the changing dynamics and cybersecurity concerns of decentralized networks. In this study, we introduce a novel biologically inspired protection system based on neuromorphic principles, where each distributed energy resource (DER) functions as a simple neuron. These neurons process local changes in voltage, current signals, and converting them into spike patterns that represent the severity of disturbances. Just as neurons communicate via synapses in biological systems, we exploit transmission cables to coordinate between DERs, enabling them to share information and respond to faults collectively. Fault detection and circuit breaker activation are driven by a First-To-Spike (FTTS) mechanism, similar to the concept of traveling wave protection, but without needing GPS synchronization or communication links. A key innovation is the ability to use the timing of spikes to locally determine the nature of a fault, offering an intelligent, adaptive response to disturbances. Performance shows tripping latency of 10-58 ms, surpassing conventional relays and even traveling-wave methods (60 ms), while maintaining detection accuracy above 98% and spatial selectivity over 97%, enabling real-time, communication-free, scalable protection for plug-and-play microgrids.

Figures (25)

• Figure 1: Neuromorphic foundation of spike-based protection for AC microgrids: (a) A two-bus AC microgrid setup with inverter-based DERs connected via a tie-line, protected by circuit breakers at both ends. Each DER functions as an independent spiking unit, allowing fully decentralized and communication-free protection. (b) Each DER is modeled as a Leaky Integrate-and-Fire (LIF) neuron, where inverter-side capacitance and line resistance together replicate the behavior of a biological neuron's membrane using an equivalent RC circuit. (c) The neuron accumulates input spikes over time, and once the membrane potential crosses a voltage threshold, it emits an output spike—mimicking how biological neurons fire. (d) Increasing disturbance severity shortens the spike interval, resulting in faster spike emission. This inverse-latency characteristic is mathematically equivalent to the IEC 60255 Standard Inverse time curve, establishing a direct analytical bridge between classical IDMT protection and the proposed neuromorphic encoding.
• Figure 2: Schema of the proposed spike-based neuromorphic protection framework, which begins with (a) event-driven sensing, where a load change around 0.5 s and a fault near 1.5 s create observable changes in voltage, current, and power—resulting in a rise in the disturbance index $D(t)$. These signals are then passed through (b) spike encoding, where local deviations are converted into input spike trains based on their severity. In membrane integration and adaptive thresholding, the neuron integrates the input spikes, and the membrane potential $V_{\mathrm{m}}(t)$ rises until it crosses the adaptive threshold $V_{\mathrm{th}}(t)$, triggering an output spike. Finally, (c) the first-spiking neuron sends a trip command to its local circuit breaker, isolating the faulted line. The entire response is fast, local, and communication-free decision-making through asynchronous spiking dynamics.
• Figure 3: Influence of electrical and neuromorphic parameters on spike-driven fault response. (a) DERs connected through lower impedance lines (Z=$2\,\Omega$) detect faults faster than those on higher impedance lines (Z=$6\,\Omega$), as stronger disturbances lead to earlier spikes. The observed inverse-time behavior matches well with classical relay logic. (b) A neuron with a smaller membrane time constant ($\tau = 86\,\mu s$) spikes earlier than one with larger $\tau = 161\,\mu s$, highlighting how spike timing can be tuned biologically to reflect fault location. (c) Tripping time varies with fault severity---quickest for three-phase faults (LLL), slower for line-to-line (LL), and slowest for single-line-to-ground (LG)---mimicking classical selectivity. (d) The 3D surface shows how tripping time decreases with increased disturbance and lower fault resistance, replicating the behavior of inverse-definite minimum time (IDMT) curves using a fully local, neuromorphic logic.
• Figure 4: Accuracy and selectivity trends in spike-based neuromorphic protection: (a) Grouped bar plot illustrating the tripping accuracy of the spike-based neuromorphic protection scheme across different operating scenarios, including low and high membrane time constants and load disturbance cases. The framework maintains high accuracy ($>99\%$) under low $\tau$, around $81.3\%$ under high $\tau$, and $100\%$ under load disturbances, highlighting robustness and selectivity.(b) Box plot showing tripping accuracy across varying fault resistances and fault types. While the system performs reliably up to $0.5~\Omega$, occasional misclassifications occur in LG faults due to weaker disturbance signatures. The results affirm the role of spike timing in capturing fault severity and supporting accurate, selective protection decisions.
• Figure 5: Energy-efficient and fault-sensitive spike response. The neuromorphic protection system shows selective activity based on disturbance severity. In (a), the disturbance index $D_i(t)$ rises gradually during load changes but shows a sharp increase during fault events. This is mirrored in (c), where the input spike activity remains sparse under load shifts but becomes dense and urgent under fault conditions. In (b), the membrane potential $V_\text{m}(t)$ increases over time and surpasses the adaptive threshold $V_\text{th}(t)$ only during critical disturbances, triggering output spikes. Subplot (d) highlights the timing of these output spikes, showing that more severe faults trigger faster responses—58 ms for an AG fault and just 10 ms for an ABCG fault. Finally, (b) shows that the total number of spikes scales with fault severity: 477 spikes for AG, 688 for ABG, and 831 for ABCG, demonstrating how the system maintains energy-efficient behavior during normal events while intensifying its response during critical faults.