UCLA

Bidirectional power-transistor rectifier-inverter circuits have broad applications, ranging from uninterruptible power supplies, and electric vehicle motor drives, to renewable energy wind turbines. Sizable and expensive DC-link capacitors are commonly placed between the rectifier and inverter to decouple the circuit's nonlinear dynamics and facilitate control. Aiming to minimize the buffering capacitor and gain high performance for the nonlinear dynamics via the pulse width modulated (PWM) control of the transistor switches, this dissertation contributes to this field in three aspects. First, we developed a high-fidelity nonlinear discrete-time model of the continuous-time nonlinear system to facilitate model-based control design. This model is also more accurate than existing models published in the literature. Second, we proposed a novel framework of multi-input multi-output (MIMO) nonlinear model predictive control (NMPC), where both the non-linear model and approximate linearized model are used to strike the balance of the computational complexity and control performance. Third, we implemented the NMPC in real-time by field programmable gate array (FPGA), where the controller is modeled by a neural network trained offline by machine learning. We performed a simulation and experiment of the power converter control on a three-phase permanent magnet synchronous motor (PMSM) drive and demonstrated superior performance in comparison to the feedback linearization control (FBLC), power-balancing schemes, and conventional cascaded PI control.