Detection of Neutral Disconnection in LV Distribution Network by Neural Network

Neutral conductor disconnection in low-voltage distribution networks is a critical operational challenge that can lead to overvoltage, undervoltage, equipment damage, and even safety hazards for consumers. This fault is particularly common in three-phase four-wire networks supplying unbalanced loads, often caused by conductor breakage, jumper burnout, or improper connections, and it carries significant technical and financial consequences. In this study, a novel neural network–based method is proposed for both detection and localization of neutral disconnection faults. To this end, a standard low-voltage distribution network was modeled in MATLAB/Simulink. Various scenarios were simulated by considering voltage unbalance, load variation, and neutral disconnection at different points of the network, and the three-phase and neutral currents at the line origin were extracted as input features. These data were then fed into a binary neural fault detection (BNFD) model to determine whether a neutral disconnection had occurred. The results demonstrate that the proposed approach can identify neutral disconnection faults with an accuracy exceeding ۹۸%. Importantly, this performance was achieved without the need for direct measurement of neutral currents across the entire network or reliance on costly equipment, highlighting the method’s strong potential for industrial application and smart distribution network operation. Finally, recommendations are provided for further development and enhancing the model’s accuracy under real-world operating conditions.

Keywords:

Neutral Disconnection , Three-Phase Four-Wire Distribution Network , Neural Network , Simulation , Binary Classification