Model-Based Online Adaptive Space Vector Modulation

Abstract

This paper proposes a model-based online adaptive space vector modulation (oaSVM) strategy for a three-level neutral-point-clamped (3L-NPC) inverter based on the convex-combination principle of conventional space vector modulation (SVM). The method formulates triangle selection and switching-pattern scheduling as a receding-horizon mixed-integer nonlinear program using predictive models of the electrical system and semiconductor losses, while enforcing neutral-point potential regulation. To improve performance under distorted load conditions, the approach incorporates a harmonic oscillator model and stationary Kalman filters (KFs) to estimate the harmonic components of relevant measured signals. These estimates are used to build a reference trajectory from the fundamental component and define a tracking-error term that drives the predicted output toward the desired fundamental component imposed by the control loop and disturbance. In this way, the proposed oaSVM shapes the trade-off among current total harmonic distortion (THD), semiconductor losses, and neutral-point potential balance through the cost-function weights. Simulation results show that, compared with conventional SVM, the proposed method extends the achievable operating region for a fixed modulation period, yielding lower THD or lower losses depending on the selected objective, while preserving neutral-point potential balance.