I. Introduction
Since the module multilevel converter (MMC) was proposed by Lesnicar and Marquardt [1], it is widely researched by scholars and industry because of the advantages, such as modular structure, low switching frequency, and high output voltage. So far, it has been widely used in high voltage direct current (HVdc) transmission systems, motor drive, and static reactive power compensation [2], [3], [4]. Nonetheless, dozens or even hundreds of power switches are installed in MMC for medium and high voltage applications, which are considered to be the most vulnerable devices because 31% of system failures are caused by power semiconductor devices [5]. Consequently, the switch open-circuit (OC) fault of the submodule (SM) is unavoidable in the MMC operation, which makes the current path in the faulty SM alter under certain conditions. In this case, the capacitor voltage of the faulty SM rises rapidly and the SM output voltage is abnormal. The rapid rise of the capacitor voltage not only increases the voltage stress of the device, which is easy to cause secondary faults, but also distorts the output voltage, and generates odd harmonic components in the circulating current. If the switch OC fault is not dealt with in time, the MMC even stops running or is damaged by the fault [5], which is not allowed in the HVdc transmission system. While, the OC faults are not recognized promptly in the early stages of their occurrence. Therefore, high sensitivity is inevitably necessitated in switch open-circuit fault diagnosis and localization (FDL) methods. In addition, the FDL methods should have the advantages of high computational efficiency, noise isolation, simple detection circuit, and low cost, to meet the requirements of swift and accurate positioning, decent robustness and strong anti-interference ability. Hence, numerous FDL methods have been presented to handle the faults [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28], [29], [30], [31], [32], [33], [34], [35]. These methods can be broadly categorized as hardware-based [6], [7], [8], [9], [10], artificial intelligent-based [11], [12], [13], [14], [15], [16], [17], and model-based [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28], [29], [30], [31], [32], [33], [34], [35].