Near-zero NDZ islanding detection for multi-source DGs via hybrid ANFIS and adaptive fuzzy classification

Abstract

Increased penetration of distributed generation (DG) increases the likelihood of inadvertent islanding, in which traditional active/passive approaches are plagued with large non-detection zones (NDZ) and power-quality trade-offs. This article introduces a hybrid islanding detector that combines an adaptive neuro-fuzzy inference system (ANFIS) and an adaptive fuzzy classifier, concurrently benefiting from active frequency drift and rate-of-change-of-frequency (RoCoF) while consuming multi-signal features-RMS/THD of voltage and current, frequency, and active/reactive power sensed at the PCC. This architecture eliminates fixed-threshold brittleness and reduces the NDZ to zero without compromising power quality. Innovative aspects are i) a stacked, real-time sampling approach (Ts = 5 ms) that supplies per-signal ANFIS modules and a main decision ANFIS, ii) subtractive clustering for generating fuzzy rules data-driven, and iii) low iq perturbation to maintain unity power factor when querying doubtful NDZ examples. MATLAB/Simulink experimentation on a seven-case, seven-stage multi-source PV-interfaced microgrid (including power-matched NDZ) demonstrates fast, robust trips at disconnection with retention of IEEE-1547 voltage/frequency envelopes; the structure achieves minimum/ideal detection times of 0.04 s and reliably indicates islanding at ~0.4 s in matched and mismatched conditions, achieving normal breaker trip expectations (<0.1 s). Comparative analysis with ROCOV/ROCOAP suggests faster detection and stronger robustness near NDZ, assisted by THD-based disambiguation in cases of non-informative active power. Through the provision of quicker, power-quality-confining islanding decisions with near-zero NDZ, the technique raises operator safety and system stability, supporting indirectly higher renewable hosting capacity and more robust, sustainable distribution networks.