Speed Versus Accuracy in Embedded Bio-Inspired Control: PSO-Tuned Fuzzy and PI Controllers on a 48 V Synchronous Boost Converter

A swarm-tuned controller must run within a low-cost processor’s real-time budget. We validate in hardware a proportional–integral (PI) controller and a Mamdani fuzzy logic controller (FLC), both tuned by the same Particle Swarm Optimization (PSO) routine (PSO-PI and PSO-FLC) under an integral of time-weighted absolute error criterion and executed on one STM32F103C8T6-controlled 153 W, 225 kHz synchronous boost converter under line, load, and reference steps. Fuzzy inference occupies 40–50 μs of the 100 μs control period, against 2–3 μs for the PI. The PSO-PI settled faster in all nine scenarios, significantly in eight, by a median factor of 2.48; after a half-to-full load step it recovered in 225 ms with a 3.83 V (7.98%) dip, against 630 ms and 6.00 V (12.50%). The PSO-FLC reached higher efficiency in eight of nine scenarios, by 0.86 percentage points on average and 0.77 points at full load (87.94% against 87.17%; ±0.07 points for the difference with common instrument errors). Both held the settled output within 0.35% of the reference and remained stable under ±20% output-capacitance mismatch. Relative to simulation, PSO-FLC settling times on line and reference steps grew 2.2–7.9 times, against 0.65–1.57 for the PSO-PI, consistent with inference latency as a major contributor. Execution time should be priced in the tuning objective.