Multi-Stack PEM Fuel Cell Optimization with PSO-Tuned PI Control
Project Overview
This team project, developed as part of the "Electronic Energy Management Systems" course at the Technical University of Crete, investigates the efficiency of a multi-stack PEM fuel cell energy management system compared to conventional single-stack configurations. The work is based on the research paper "A new control algorithm for increasing efficiency of PEM fuel cells – Based boost converter using PI controller with PSO method" by Y. Benteşen Yakut (International Journal of Hydrogen Energy, 2024).
The core idea is to replace a single high-power fuel cell with multiple smaller fuel cells connected in parallel, sharing a single DC-DC boost converter. A control algorithm dynamically activates or deactivates individual fuel cell stacks depending on load demand, significantly reducing hydrogen consumption — approximately 5x reduction compared to a single-stack system under the same load conditions.
Configuration
5 Parallel PEM Stacks
Five 1.26 kW stacks (13 cells each) replacing a single 6 kW stack (65 cells), sharing one DC-DC converter.
H₂ Reduction
~5x Lower Consumption
Control algorithm activates only the required number of stacks per load, dramatically reducing hydrogen fuel usage.
PI Optimization
PSO Algorithm
Particle Swarm Optimization tunes Kp and Ki parameters of the PI controller, minimizing voltage oscillation at the converter output.
Load Testing
1–6 kW Variable Load
System tested across 7 load levels (1, 2, 3, 4, 4.5, 5, 6 kW) with dynamic fuel cell activation.
System Architecture
The system consists of several key components, each modeled and simulated in MATLAB/Simulink: the PEM fuel cell stacks (mathematically modeled using the Mann model with Nernst voltage, activation, ohmic, and concentration polarization losses), a DC-DC boost converter that steps up the low fuel cell voltage and regulates current distribution, a PI controller that monitors converter output and adjusts the duty cycle to maintain stable voltage, and a switch manager that reads load power and activates the appropriate number of fuel cell stacks.
Key Innovation:
Unlike previous studies that use a separate DC-DC converter for each fuel cell stack, this design uses a single converter for the entire parallel system. The PSO-optimized PI controller adapts dynamically as stacks are added or removed, maintaining output voltage stability within 2% across all load conditions.
Hydrogen Flow Rate Comparison
| Load (kW) | Active Cells | Parallel H₂ (lpm) | Single H₂ (lpm) |
|---|---|---|---|
| 1 | 2 | 1.0 | 6.5 |
| 2 | 3 | 3.0 | 13.75 |
| 3 | 4 | 4.2 | 21.56 |
| 4.5 | 5 | 6.82 | 34.64 |
| 6 | 5 | 9.84 | 50.1 |
Implementation Steps
The project was completed across four milestones, building the full system in MATLAB/Simulink:
Mathematical modeling of the PEM fuel cell stack using the Mann empirical model (Nernst voltage, activation, ohmic, and concentration losses), reproducing the paper's voltage-current and power curves — starting voltage 64.29 V, dropping to 37.42 V at maximum current.
Design of the DC-DC boost converter (IGBT-based) with a PI controller and low-pass output filter, targeting a stable 100 V DC output.
Implementation of the switch manager / controller that activates the required number of fuel cell stacks (1–5) based on measured load power.
Simulation under variable loads (1–6 kW) with PSO-tuned PI parameters, eliminating the output voltage oscillations present before optimization and achieving a stable steady state near 100 V.
Results & Conclusions
The completed system successfully delivers a stable DC output voltage using only a single boost converter, while dynamically responding to changes in both load and power supply. Before PSO tuning, the output exhibited a persistent ~10 Hz oscillation of ±10% around the target; after applying the PSO-optimized PI parameters, these oscillations were eliminated across all tested loads.
Key Takeaways:
Fuel cells can power loads with a stable DC voltage without needing multiple DC-DC converters, dramatically reducing installation cost. The parallel, independently-switchable architecture makes the system flexible and scalable, and the PI controller's tuning proved not just desirable but essential for stable operation.
The report also identifies future improvements: a multi-phase converter for higher voltage step-up and reduced oscillation, more advanced control (PID, Fuzzy Logic, or Neural Networks) given the PI controller's sensitivity, and a more sophisticated stack-activation logic that accounts for branch efficiency, wear, and fuel availability.