The conventional model predictive control (MPC) suffers from high common-mode voltage (CMV) magnitude and large current ripples. In this paper, the reduced CMV MPC strategy with double vector (RCMV-MPCDV) is proposed for three-level inverters, which adopts double vector to reduce the CMV and current ripple simultaneously in per-sampling period. Firstly, based on geometric relationship of median, a novel cost function that evaluate median line distance is proposed to reduce the computational complexity, rather than evaluating the closest distance. Secondly, in order to reduce the computational burden, the double vector preselection (DVP) algorithm is presented with fewer evaluation times. Thirdly, the candidate vectors are reclassified and regrouped from partial basic vectors, which restricts CMV magnitude within one-sixth of DC-link voltage. In addition, to select more satisfactory double vectors for RCMV-MPCDV scheme, two improved methods are proposed to compensate the losing candidate vectors. The simulation and experimental results with steady-state performance and dynamic response validate the effectiveness of the proposed method.

Recently, a few fault-tolerant control strategies have been proposed for T-type three-level inverters to improve the reliability of the inverter system. However, the neutral-point (NP) voltage oscillations suppression is never considered in the previous fault-tolerant control strategies. In order to suppress the NP voltage oscillations when open-circuit fault occurs to NP switches, this paper proposes two novel fault-tolerant control strategies named as improved two-level switching tolerant control (I2LS-TC) and hybrid switching tolerant control (HS-TC). Both methods are implemented by adding an optimal time-offset to the three-phase turn-on times. The corresponding time-offset value of each method is simply calculated considering the phase current and dwell time of [O] state. Moreover, a NP voltage balancing algorithm to eliminate the dc deviation and a dc bias to further reduce the NP voltage oscillations are used in the HS-TC. The performance and comparison of the tolerant controls with respect to the NP voltage oscillations, current total harmonic distortion (THD) and dynamic response are illustrated and analyzed through the simulation and experimental results.

Parallel three-level inverters have attracted increasing attention due to the advantages to increase the power rating and the flexibility of power conversion systems. However, this type of configuration requires to suppress the zero-sequence circulating current (ZSCC) through either modulation or active control strategies, to avoid overloading problems of a single three-level inverter. For the operation of paralleled three-level T-type inverters ( $\text{3LT}^2 \text{Is}$ ), another requirement is to reduce the common mode voltage (CMV) associated with PWM method. To better solve these problems simultaneously, an optimized zero-sequence voltage injection method is proposed to both reduce CMV and eliminate ZSCC for multi-parallel-connected three-level inverters. A generalized CMV reduction strategy is developed, which can reduce the magnitude of CMV to one-second of the value using the conventional space vector modulation (SVM). The highlight of the method is that ZSCC is eliminated by adding an optimal compensation value to the reference duty signals with lower CMV. This compensation value is calculated simply by ZSCC, duty cycles and filter parameters. The better ZSCC suppression performance can be obtained in both steady and transient states with the proposed scheme, thus, improving the quality of currents.

Unsupervised domain adaptation (UDA) in person re-identification (re-ID) is a challenging task, aiming to learn a model with labeled source data and unlabeled target data to recognize the same person in the target domain across different cameras. Recently, a lot of popular and promising methods based on clustering are proposed for this task and achieve a sizable progress. However, in those methods, without target labels, the clustering algorithms will inevitably produce noisy pseudo-labels. Overfitting on these noisy labels is severely harmful to the performance and generalization of models. In order to address the above issues, we propose a novel framework, Adaptive Deep Clustering (AdaDC), to reduce the negative impact of noisy pseudo-labels. On one hand, the proposed approach employs different clustering methods adaptively and alternately to fully exploit their complementary information and avoid overfitting noisy pseudo-labels. On the other hand, there is a progressive sample selection strategy for reducing noisy label ratio in pseudo-labels, which is achieved by integrating different clustering results. Experiments present that the proposed approach can achieve state-of-the-art performance compared to the other recent UDA person re-ID methods on widely-used datasets. Moreover, there are some other analysis experiments conducted for verifying the effectiveness of the proposed approach.