Experimental Exploration of Model Predictive Control for 3-Phase Split-Source Inverters (EXPLORE-MP3S)

Objectives: 
The primary aim of the EXPLORE-MP3S project was to implement and evaluate finite set 
model predictive control (MPC) strategies for a 3-phase split-source inverter (SSI) using a real
time hardware-in-the-loop (HIL) platform. This involved: 
• Applying finite set MPC algorithms to the SSI. 
• Comparing MPC performance with the modified space-vector pulse-width modulation 
(MSVPWM), a recognized PWM strategy for SSIs. 
Scope and Motivation: 
The 3-phase SSI is a novel inverter topology with several benefits, including reduced passive 
component count and continuous input current. Finite set MPC has gained popularity in power 
electronics due to its intuitive design, ability to handle constraints, and excellent dynamic per
formance. The project sought to explore these advantages by applying MPC algorithms in a 
real-time HIL setup and comparing it to MSVPWM.
Methodology: 
The project utilized a Typhoon HIL emulator and a DSP-based HIL controller (Vindobona) for 
simulation and real-time testing. Various MPC algorithms were tested, including: 
• MPC with two cost functions, with forbidden simultaneous switching in all three phase 
legs, and with a single zero switching state allowed (MPC-2CF) 
• MPC-2CF with both zero switching states allowed (MPC-2CF-2Z) 
• MPC-2CF with allowed simultaneous switching in all three phase legs (MPC-2CF-3L) 
• MPC-2CF with switching penalization by means of a weighting factor (MPC-2CF-SW) 
• MPC with a single cost function, with forbidden simultaneous switching in all three 
phase legs, and with a single zero switching state allowed (MPC-1CF) 
Performance metrics included average switching frequency, steady-state errors, total har
monic distortion (THD), and dynamic performance. Initial simulations and virtual HIL testing 
were conducted before real-time implementation. 
The project was structured in the following key phases: 
• Week 1: Laboratory setup and equipment familiarization were completed as planned. 
• Week 2: MSVPWM-based control of the SSI was successfully implemented using the 
HIL platform. 
• Week 3: Challenges arose with the MPC algorithm HIL implementation, requiring mod
ifications to the controller firmware and SSI model, leading to incomplete results. 
• Week 4: HIL implementation showed numerically stable responses but revealed issues 
with high ripple, total harmonic distortion (THD), and lower-than-expected switching 
frequency. Experimental validation was not pursued further. 
• Week 5: Efforts to address low switching frequency and waveform quality were met 
with limited success. Experimental validation was abandoned, and the project focused 
on documenting HIL-based results.
Challenges:
The real-time HIL testing encountered numerical inconsistencies and adaptation issues, lead
ing to the decision to forgo experimental validation within the project's timeframe. The imple
mentation of MPC faced difficulties due to synchronization problems between the Typhoon HIL 
model and the Vindobona board, requiring adjustments to execution times and firmware. Nu
merical instability was a significant issue, aggravated by high snubber values in the SSI model. 
The HIL controller's adaptation for MPC and the debugging process proved more complex than 
anticipated. Despite optimization efforts, the MPC algorithms did not meet performance expec
tations. Issues included high ripple, THD, and reduced switching frequency. These problems 
were linked to the snubber value and model inaccuracies. While a full comparison of MSVPWM 
versus MPC was not feasible, the project still provided valuable insights into the performance 
of different MPC algorithms and HIL-based validation emerged as a valuable contribution. 
Key Findings: 
Performance Comparison: 
• MSVPWM demonstrated superior performance compared to all MPC variants, espe
cially in terms of switching frequency, harmonic content, and overall efficiency. 
• MPC-2CF emerged as the most effective MPC variant, providing a good balance of 
performance metrics. 
Transient Responses: 
• In open-loop control, MPC-2CF exhibited minimal overshoot and quick response to load 
changes, outperforming MSVPWM, which experienced significant overshoot and oscil
lations. MPC-1CF struggled with stability during load reductions. 
• For closed-loop control, both MSVPWM and MPC-2CF effectively controlled load volt
age, but MPC-2CF experienced longer settling times due to the PI controller's influence 
on reference signals. 
Impact of Parameters: 
• Increasing the sample time in MPC-2CF-3L led to increased current ripple and THD, 
highlighting a trade-off between switching frequency and performance. 
• Switching penalization in MPC-2CF-SW reduced switching frequency but adversely af
fected ripple and THD. 
• Discrepancies in inductance values between MPC code and HIL models significantly 
impacted performance, particularly in current ripple and steady-state error and 
achieved output voltage. 
• MPC-2CF-2Z and MPC-2CF-3L exhibited similar inductor current ripples, but MPC
2CF-3L required higher maximum sampling frequency to achieve comparable perfor
mance to MPC-2CF. 
Delay Compensation: 
• Implementing delay compensation improved inductor current ripple and switching fre
quency, enhancing performance indicators such as load current THD and input current 
error. 
Overall, the study highlights that while MSVPWM remains superior in many respects, MPC 
algorithms, particularly MPC-2CF, offer competitive performance with various trade-offs. The 
findings suggest that optimal control strategy selection depends on specific performance pri
orities and operational conditions.