Measurement-aware zero-phase iterative learning control for robust regulation of nonideal boost converters in electric vehicle fast charging

Integrating battery energy storage with DC-DC boost converters for electric vehicle fast charging exposes the regulator to ageing-induced parameter drift, periodic load pulses, and, critically, the nonidealities in the output-voltage sensing chain. This paper proposes a measurement-aware, zero-phase iterative learning control scheme for robust output-voltage regulation of a nonideal boost converter whose parameters are matched to those of a commercial Texas Instruments TPS6102x battery regulator. The controller combines an inner proportional-integral stabilizing loop with an outer zero-phase learning law that updates a feedforward correction based on the sensor-captured output trajectory; a forward-backward robustness filter suppresses the amplification of measurement noise and quantization across successive trials without introducing phase lag. The averaged small-signal model is validated against a 960 kHz switched simulation, with agreement within 0.2% at the operating point and a transient-waveform correlation of 0.91, while the bandwidth limitations imposed by the lightly damped resonance and the nonminimum-phase dynamics on feedback are quantified from the loop frequency response. Because feedback alone is bandwidth-limited, the learning feedforward recovers the intrinsic converter speed for the repetitive task and reduces the periodic tracking error by more than an order of magnitude relative to a fairly tuned baseline. A bounded-error convergence theorem shows that, under real sensing, the tracking error converges to a neighbourhood whose size is proportional to the measurement-disturbance level; this is confirmed numerically, with the asymptotic error scaling approximately linearly with the noise standard deviation once the noise dominates the filter bias. Systematic sweeps of converter resolution, noise, and sensor delay show that the zero-phase measurement-aware law remains bounded where a naive learning law diverges, and a 500-sample Monte Carlo study confirms robustness to component tolerances, load variation, and parasitic aging. The primary physical task is the rejection of a repeating high-current charging pulse: over successive charging cycles the scheme reduces the periodic output-voltage sag from 35% to below 0.01%, whereas the best feedback comparator leaves a 17% residual, and the sinusoidal reference used elsewhere serves as a spectral diagnostic that isolates the feedback-bandwidth ceiling at a single known frequency. The learning law is further verified in closed loop on the cycle-by-cycle switched large-signal converter, where it converges and stays bounded. A per-unit scale-invariance analysis then establishes the conditions under which the design obtained on the low-power benchmark transfers to fast-charging power levels, as confirmed on a 400 V, 6.1 kW stage that reproduces the benchmark convergence exactly.