Abstract
Silicon Carbide (SiC) power devices are increasingly applied in high-efficiency systems such
as electric vehicles, renewable energy converters, industrial drives, and aerospace power
systems because of their superior thermal capability and switching performance. In DC-AC
inverters, however, achieving optimum performance depends on the careful selection of
switching frequency and dead-time. Conventional fixed proportional-integral control
strategies use static switching parameters and fail to adapt under varying operating
conditions, which can lead to reduced efficiency, increased thermal stress, and shorter device
lifespan. This creates a need for intelligent and adaptive optimisation methods that can
improve inverter operation in real time.
In this study, a 13 kW, 850 V Wolfspeed reference inverter was modelled in
MATLAB/Simulink to generate performance data over a switching-frequency range of 20 to
50 kHz and a dead-time range of 200 to 400 ns. A supervised machine learning model based
on a neural network was developed to predict inverter efficiency from these operating
parameters. The trained model was then integrated with a Genetic Algorithm to dynamically
determine the optimum switching frequency and dead-time parameters. The model was
further validated against datasheet values obtained from Wolfspeed, Infineon, and Rohm.
The baseline simulation produced an efficiency of approximately 98.26% under fixed control
settings of 20 kHz and 300 ns. The ML–GA optimisation framework identified operating
points that improved efficiency to as high as 99.63%, while also reducing switching losses
and maintaining stable power output. These findings show that intelligent optimisation can
significantly improve inverter performance beyond that of conventional fixed-parameter
control. Improved inverter efficiency also has wider practical benefits, including lower
energy losses, reduced cooling requirements, lower carbon emissions, and improved
sustainability in high-demand applications such as electric mobility and renewable energy
systems. These results confirm that adaptive, data-driven control strategies are well-suited to
unlocking the full performance potential of wide bandgap semiconductor devices across a
broad range of high-efficiency power electronic applications.
Index Terms: Silicon Carbide (SiC), DC-AC inverter, machine learning, genetic algorithm,
switching frequency, dead time, power electronics, neural network, MATLAB/Simulink, ...