Multi-Objective Particle Swarm Tuning of Dual-Active-Bridge Control for Bidirectional Wireless Vehicle-to-Grid Interfaces: A Critical Review of Concurrent Harmonic, Switching-Loss and Dynamic-Response Optimisation

Abstract

Bidirectional wireless vehicle-to-grid interfaces combine a loosely coupled inductive link with high-frequency bridges on both the infrastructure and the vehicle side, producing a converter that is structurally equivalent to a dual-active-bridge operating through a resonant network with a variable, load-dependent coupling coefficient. Control of such an interface must satisfy several requirements at once: grid and coupler current waveforms must remain within harmonic limits, conduction and switching losses must stay low enough to justify the thermal and economic penalty of wireless transfer, and the power-flow reversal that defines vehicle-to-grid service must be executed without excessive transient excursions. These requirements are not independent, and the modulation freedoms that improve one frequently degrade another. Multi-objective particle swarm optimisation has been proposed as a tuning framework capable of exposing the trade-off surface rather than concealing it inside a weighted scalar cost, yet its application to this specific converter class remains fragmentary. This review examines how total harmonic distortion, switching loss and dynamic response have been formulated, optimised and validated in the dual-active-bridge and bidirectional inductive power transfer literature, and assesses the extent to which population-based multi-objective tuning has actually delivered on its stated promise. Literature retrieved through open scholarly indexes to 21 July 2026 was appraised for methodological adequacy, evidential strength and the fidelity of the models used inside the optimisation loop. The evidence supports three conclusions: objective conflict is real and is repeatedly demonstrated in single-objective studies that report collateral degradation; harmonic distortion is defined inconsistently and is frequently omitted from the optimisation vector altogether; and dynamic response is the least well represented objective, usually assessed after the fact rather than optimised concurrently. Confidence in reported improvements is limited by low-fidelity loss models, simulation-dominant validation and the near-absence of coupling-variation robustness testing. Research priorities are identified accordingly.