Physics-informed prediction with hybrid gbestABC–NSGA-II metaheuristics in stochastic multi-objective optimal power flow for renewable-EV grids

Abstract The increasing penetration of wind generation, solar photovoltaic (PV) systems, and plug-in electric vehicles (PEVs) introduces uncertainty and conflicting operational requirements into modern power systems. This study proposes a stochastic Wind–PV–PEV multi-objective optimal power flow framework based on a hybrid global-best-guided Artificial Bee Colony and NSGA-II ( gbestABC–NSGA-II ) algorithm. Wind speed, solar irradiance, and PEV charging/discharging power are represented using Weibull, lognormal, and normal probability distributions, respectively, and evaluated using Monte Carlo scenarios. The proposed algorithm combines guided ABC exploration with NSGA-II elitist sorting and diversity preservation. Enhanced epsilon-constraint handling supports feasibility restoration, while an epsilon-dominance and crowding-distance archive maintains a bounded and distributed Pareto set. AHP–TOPSIS is employed to select a preference-aware Best Compromise Solution. The framework is evaluated through 12 bi-, tri-, and quad-objective cases on modified IEEE 30-, 57-, and 118-bus systems. Under the common computational budget used in this study, the reported comparisons show that the proposed method yields competitive Pareto solutions relative to MOABC, NSGA-III, MOEA/D, NSGA-II, and MOPSO. In the reported cases, APL is up to approximately $$27\%$$ 27 % lower and VMD is approximately $$18\%$$ 18 % – $$52\%$$ 52 % lower than the corresponding values of selected comparators. Pareto-front quality assessments further show GD values more than $$25\%$$ 25 % lower and SP values up to $$28.5\%$$ 28.5 % lower, while DM is up to $$37.3\%$$ 37.3 % higher than the stated comparator values. Wilcoxon signed-rank analyses ( $$p<0.05$$ p < 0.05 for the reported pairwise comparisons) provide statistical support for differences in the investigated Pareto-quality observations. A physics-informed neural network is additionally incorporated for rapid power-flow screening, and every candidate dispatch retained for implementation is verified using the complete Newton–Raphson AC power-flow model before acceptance.