Academic Awards 2024 booklet

25 Real-time Implementation of Model Predictive Space Vector Modulation for a Three-Level Neutral-Point-Clamped Inverter The electrical grid is undergoing a transition to green energy by integrating renewable sources like solar photovoltaic panels and wind turbines. These sources require AC to DC power converters (inverters) to operate, while maintaining high efficiency under various load conditions. Modern control and optimization techniques for these converters require intensive computations on a microsecond scale. Additionally, the optimization algorithms involved are often NP-hard, with integer variables and non-linear objectives, making them difficult to implement in real power converters despite their potential benefits in increasing efficiency by computing optimal switching patterns for the semiconductor switches inside the inverters, every few hundred microseconds. In my bachelor thesis, I considered a new state-of-the art optimization algorithm developed by my supervisors, named “model predictive space vector modulation (MPSVM)”, which through simulations has proven to increase the efficiency of such power converters. Nevertheless, the optimization problem associated with MPSVM is NP-hard and was impractical to run in practice. In my work, I first developed heuristic methods to reduce algorithm's optimal solution search space. Then, I developed a dedicated hardware accelerator architecture to solve MPSVM in real-time and implemented it on an FPGA (field programmable gate array). The accelerator achieved results of solving MPSVM continuously in just under 20 microseconds, hence enabling MPSVM to run in microsecond speeds. 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8 9 10