Active Disturbance Rejection Versus Model Predictive Controllers for CC–Taper Bidirectional EV Charger Control Based on Genetic Algorithm and Particle Swarm Parameter Optimization

Bidirectional electric vehicle (EV) chargers must regulate current and voltage accurately in both grid-to-vehicle (G2V) and vehicle-to-grid (V2G) modes while respecting the tight real-time computational budget of embedded hardware. This paper compares three controllers, active disturbance rejection control (ADRC), a linear model predictive controller (MPC), and a reduced-order quadratic-programming predictive variant (rMPC), for a bidirectional charger supplying a 96-cell nickel–manganese–cobalt battery modeled as a second-order equivalent circuit. The gains of all three controllers are tuned systematically with particle swarm optimization (PSO) and a genetic algorithm (GA) against a weighted objective covering current tracking, terminal-voltage error, disturbance recovery, and control effort. Each controller is evaluated over a constant-current/taper charging profile, a V2G discharge, and injected load disturbances, with robustness quantified by a 500-run Monte Carlo sweep of inductance, resistance, and temperature. After tuning, every controller meets the 5% normalized-error target across all phases, with constant-current tracking error below 2%. ADRC’s mean per-step solve time (∼14.5 μs) is 19–30× smaller than that of the predictive controllers. A 20-seed statistical comparison at equal budget shows PSO and GA to be statistically equivalent tuners, and the decisive practical distinction among three comparably accurate and robust controllers is computational: ADRC offers bounded, negligible worst-case timing, while rMPC and MPC offer marginally faster disturbance recovery at a far higher and less predictable solve cost that exceeds the real-time budget at the 99th percentile.