Enhancing Voltage Stability under GCC Constraints with the AI-Driven Optimization of Distributed Generators

This paper presents a novel, data-driven, multi-objective optimization framework that uses Grasshopper Optimization Algorithm (GOA), Adaptive GOA (AGOA), and Flower Pollination Algorithm (FPA) to manage the active and reactive power outputs of Distributed Generators (DGs) within the limits of the AI-modeled Generator Capability Curve (GCC). This improves the voltage profiles and reduces the voltage unbalance in the distribution systems. The proposed method entails two steps. First, the GCC is reconstructed using a Deep Learning (DL) model with 20 neurons and 16 hidden layers. This model is trained utilizing numerical data for 2.5 MW and 3.0 MW DGs and achieves a minimum Mean Squared Error (MSE) of 1×10⁻⁸. Second, the reconstructed GCC is integrated as a dynamic constraint in the optimization model to guide the DG dispatch. Three metaheuristic algorithms were applied to optimize the DG operation under unbalanced loading conditions at buses 10 and 15. AGOA had the best performance, reducing the voltage unbalance from 2.2129% to 1.4086% at bus 10 and from 2.0820% to 1.4295% at bus 15. AGOA also restored the voltage at bus 15 from 0.8698 p.u. to over 0.926 p.u. and achieved the lowest convergence fitness (<1.43). These results confirm AGOA's effectiveness in enhancing the voltage stability and phase balance, emphasizing the advantages of integrating DL-based GCC modeling with adaptive metaheuristic optimization for reliable and efficient DG operation.