When Does Grid-State-Aware Parameter Adaptation Help Metaheuristic Dispatch of Photovoltaic Distributed Generation? A Reproducible Multi-Feeder Benchmark

Metaheuristic optimisation is widely used for distribution-network operation, yet claims of superiority are often based on single cases without equal budgets, deterministic references, or reproducible seeds. This study evaluates a fuzzy-adaptive particle swarm optimiser using voltage-deviation and renewable-ramp inputs (FA-PSO-UR) for the dispatch of inverter-interfaced photovoltaic (PV) distributed generation (DG). It is compared with a genetic algorithm (GA), classical particle swarm optimisation (PSO), differential evolution (DE), a grey wolf optimiser (GWO), and time-varying PSO (TV-PSO) on the IEEE 33- and 69-bus feeders. The frozen benchmark comprises 1,080 static runs and 720 sequential 24-hour runs under smooth and cloud-driven PV profiles, with paired seeds and identical function-evaluation budgets. DE achieves the lowest static median in all six cases, whereas FA-PSO-UR is engineering-equivalent to PSO in all six and to TV-PSO in five. All sequential schedules are feasible. FA-PSO-UR achieves the lowest median objective in three of four sequential scenarios, with median gaps of 0.0056–0.0212% to the best-known deterministic references. TV-PSO ranks first for the IEEE 69-bus smooth profile, where the two methods remain engineering-equivalent. FA-PSO-UR shows practical superiority only over selected weaker baselines, not over PSO or TV-PSO. Physical-state-aware adaptation therefore provides selective dynamic benefit rather than universal dominance.