In modern micro-grid systems, the control of power electronic converters faces numerous challenges, including the uncertainty of parameters of the controlled objects, variations in the operating environment, and the increased complexity of models for controlled objects. Meanwhile, traditional control strategies for power electronic converters are characterized by high model dependency, slow dynamic response, and static control parameters, which make them inappropriate for the development of modern micro-grid systems. In response to this problem, researchers have proposed reinforcement learning (RL) control strategies to achieve fast adaptive control through real-time interaction with the environment. However, there is currently a lack of comprehensive literature reviews specifically focusing on RL control for power electronic converters. To fill this gap, this article provides a summary of existing published research papers. Specifically, the mainstream RL algorithms are listed. The existing RL control strategies for various power electronic converter topologies are introduced. The sim-to-real methods for practical implementation are illustrated. The underlying design considerations are discussed. Moreover, the future research prospects are explored.

In the vessel integrated power system (IPS), several common events, such like replacement of the aging batteries, isolation of fault modules, and launching of pulse loads, can make the state-of-charge (SOC) of energy storage elements (ESE) inconsistent, which consequently requires different charging or discharging power command for the ESEs to balance their SOC. Modular multilevel converter õbidirectional DC-DC converter (MMC-BDC) can be utilized in the vessel IPS to connect ESEs with the medium voltage (MV) bus for its outstanding characteristics. However, the degree of imbalance in the module power of MMC-BDC is limited inherently, which can destabilize the system but is usually neglected by the researchers. Therefore, this paper investigates into the unbalanced operating principle of a cascaded MMC-BDC. The imbalance boundary is studied. A control strategy is designed for the unbalanced operation to optimize the voltage stress of the components. A method of allocating unbalanced module power to the sub-modules (SM) is raised on the basis of imbalance boundary. And a novel fast balancing charging strategy is proposed consequently. Finally, the proposed theory and strategies are verified through simulation and a 7.2kW prototype.

In order to equip more high-energy pulse loads and improve power supply reliability, the vessel integrated power system (IPS) shows an increasing demand for high-voltage and large-capacity energy storage systems. Based on this background, this paper focuses on a super capacitor energy storage system based on a cascaded DC-DC converter composed of modular multilevel converter (MMC) and dual active bridges (DAB). The cascaded converter is called MMC-DAB for short. This paper analyzes the topology and modulation strategy of the system in detail. Taking into account the shortcomings of the traditional bidirectional power control strategy, this paper proposes a control strategy where the DAB module of each branch independently controls the voltage of the sub-module capacitor. The mathematical models of the traditional and the proposed control strategy are established. The stability of the two control strategies is analyzed and compared. Finally, a MW-level engineering prototype of the MMC-DAB energy storage system is designed and manufactured, and the effectiveness of the proposed control strategy was verified through experiments.

The modular triple-port DC-DC converter (MTDC) is recommended as the energy router for DC power distribution of the vessel integrated power system, and the stability assessment of MTDC is an issue of concern. However, the interactions among different branches are so significant that the conventional criterion may not have the ability to judge stability readily. Thus, research work related to stability analysis of MTDC is further explored, and a new criterion for MTDC is presented in this article. Firstly, the basic principle of MTDC is introduced, such as the topology, operating modes and control strategies. After that, the port impedances of all sub-converters are given, including the transfer functions of the port current to the non-corresponding port voltage in other branches. The small-signal model of the cascaded system is obtained by considering the system as an interconnected network. Then, the equivalent loop gains are precisely derived which indicates the cascaded stability criterion of MTDC. Finally, experiments were conducted in a 10 kW/500 V MTDC engineering prototype to verify the effectiveness of proposed stability assessment method.

Due to the widespread adoption of electric vehicles, integrated charging stations have been increasingly applied. However, because of the extensive number of components, when a fault occurs, how to identify the faulty components is a great challenge. In this paper, considering the power conversion system of integrated charging stations, the impact of internal open-circuit faults in IGBTs and capacitors, along with fault identification methods, are discussed. By combining signal processing and deep learning for fault localization, the neural network model based on the bidirectional gated recurrent unit (Bi-GRU) and attention mechanism is proposed, which can flexibly learn crucial information, enhance the perception of important segments in sequences, and achieve higher accuracy in open-circuit fault identification. Through comparative analysis for the validation set among different models, the effectiveness and advantages of the proposed model and analysis are proved. The proposed method has good versatility and can be used for fault identification and operation reliability improvement for other power electronics-based systems.

Modular multilevel converter-bidirectional dc–dc converter (MMC-BDC) can be used in the second generation of vessel integrated power system (IPS) to interconnect the medium-voltage dc (MVdc) bus and distributed energy storage elements (ESEs) for its outstanding characteristics. When MMC-BDC operates with submodule (SM) power well balanced, traditional phase-shifted modulation (TPSM) reduces the current ripple at the MVdc side of MMC-BDC for the frequency multiplication effect. However, when it comes to the unbalanced operation conditions, TPSM can result in large current ripple. In this article, the relationship of the MVdc-side current ripple and the switching signal patterns is revealed. An optimal phase-shifted modulation (OPSM) method based on offline genetic algorithm (GA) is proposed to suppress the MVdc-side current ripple by adjusting the phase-shifted ratios of the driving signals of each SM. Combining the proposed modulation method with a proposed online stepwise energy management strategy, all unbalanced conditions can be covered with limited sets of optimal phase-shifted ratios. Simulation and experiment results show that the MVdc-side current ripple of MMC-BDC is remarkably reduced with the proposed OPSM.