Tolerant Sequential Model Predictive Control Based on Lexicographic Optimization Method for T-Type Three-Phase Three-Level Inverters

Due to the existence of the neutral point (NP) voltage, controlling the three-level inverters has essentially become a multiobjective optimization problem (MOOP) that needs to provide stable output voltages for the load and maintain the NP voltage simultaneously. Traditionally, this MOOP is converted into a single-objective optimization problem by weighting factors. However, since the physical dimensions of the two control objectives are usually different, it is challenging to choose proper weighting factors to obtain a satisfactory performance according to a specific theory. To address this issue, a tolerant sequential model predictive control (TSMPC) utilizing a lexicographic optimization method is proposed in this article. This method establishes two distinct layers for the output voltage and NP voltage, arranging them in sequence according to the importance of control objectives to evaluate all voltage vectors. By using an explainable tolerance value rather than conventional weighting factors, the proposed TSMPC algorithm presents superior performance over traditional MPC approaches. Finally, the feasibility and effectiveness of the proposed TSMPC algorithm have been verified through relevant experiments and the stability of this algorithm has also been analyzed.

I. Introduction

Amidst the global shift towards low-carbon energy and the pursuit of carbon neutrality, transitioning the energy landscape of the transportation and power generation sectors is extremely crucial. Because these two sectors are significant consumers of conventional fossil fuels, research into transportation electrification and renewable energy generation technologies has become very popular [1], [2], [3], [4]. Inverters, as the critical link between dc and ac systems, have emerged as vital components in renewable energy generation systems and electric vehicles [5], [6]. Meanwhile, strategies for controlling inverters have been explored continuously. Proportional-integral control, a linear control algorithm, is widely used in industry due to its relatively simple control principles and low computational requirements [7], [8], [9], [10]. However, there exists a trade-off between the steady-state and dynamic performances of this algorithm. In addition, the inadequate performance of systems with strong nonlinearity, which proves challenging to approximate linearly, is also a significant issue that cannot be overlooked [11], [12]. As the pursuit of better performance on nonlinear systems intensifies, and thanks to the enhancement of the computing power of microprocessor chips, model predictive control (MPC) is beginning to be widely used in power electronics because of its good performance on multivariable and nonlinear systems [13], [14], [15].