AI-driven ANN–PSO smart-inverter capacity allocation at preselected buses under meteorological uncertainty

Abstract

Introduction Radial distribution grids require renewable planning methods that jointly account for meteorological generation, WT/PV capacity allocation, and smart-inverter voltage support. Methods An ANN-PSO-VoltVAR framework was developed using 2012-2021 NASA POWER irradiance, wind-speed, and temperature data for In Salah, Algeria. Renewable uncertainty was represented by 12 stratified Weibull-Beta scenarios and an expected-cost/CVaR objective. The mixed decision vector optimized WT units, PV strings, and voltage-dependent inverter support at four preselected IEEE 85-bus locations. An exact backward/forward-sweep module on the standard 33-bus feeder independently verified the load-flow implementation. Results The tuned ANN achieved an RMSE of 1.6323 kW, an MAE of 1.2314 kW, and an R-squared value of 0.99999. The verified five-start mean-profile solution reduced mean active loss to 15.9900 kW and voltage deviation to 0.051082 p.u., while increasing minimum voltage to 0.98213 p.u. and minimum VSI to 0.96307. Relative to IFLO, the respective improvements were 54.57%, 22.49%, 1.21%, and 8.41%. Discussion The ablation analysis showed that capacity sizing accounted for most of the loss reduction, whereas Volt-VAR support produced the clearer additional voltage-security gains. Because the IEEE 85-bus electrical response is benchmark-calibrated rather than fully reconstructed, the findings should be interpreted as a reproducible proof-of-concept. Direct full-feeder validation on the IEEE 85-bus system and other networks remains necessary.