Adaptive neuro-fuzzy based droop control for voltage stabilization in multi-node bipolar direct current microgrids

In comparison with conventional unipolar configurations, multi-node bipolar direct current microgrids (BDCMGs) offer multiple voltage-level interfaces that enhance system reliability, operational flexibility, and energy utilization. This study proposes an advanced control strategy for direct current (DC) bus voltage stabilization in BDCMGs employing a bidirectional single-ended primary-inductor converter–Ćuk (SEPIC–Ćuk) fused converter topology. The bidirectional capability of the converter ensures consistent voltage regulation under varying load conditions and supports seamless bidirectional power flow. However, voltage deviations arising from load imbalance and line resistance require intelligent and adaptive control mechanisms. To address these challenges, an adaptive fuzzy logic based droop controller integrated with an adaptive neuro-fuzzy inference system (ANFIS) is developed. ANFIS combines the reasoning capability of fuzzy logic with self-learning characteristics of neural networks, enabling automatic tuning of membership functions and rule parameters through a hybrid algorithm. The proposed method represents an artificial intelligence-based control approach for voltage stabilization in BDCMGs. In the proposed control scheme, converter output voltage and the battery state of charge (SOC) are utilized as controller inputs to generate an adaptive droop coefficient that regulates both the positive and negative DC bus voltages, thereby maintaining overall system stability. This approach allows real-time adjustment of droop coefficients based on voltage deviation, voltage unbalance index, and SOC, leading to improved dynamic performance and enhanced system robustness. Simulation results in Matrix Laboratory (MATLAB) and the Simulink simulation environment by MathWorks validate the proposed ANFIS-based control strategy, demonstrating voltage imbalance mitigation, stable DC bus regulation, and improved power sharing within multi-node BDCMGs.