Due to a dramatic increase in grid-connected renewable energy resources, energy storage systems (ESSs) are believed to be a must in future power systems, among which the modular multilevel converter (MMC)-based battery energy storage systems (BESSs) are one of the most modular, efficient and flexible topologies. Uneven active power distribution among sub-modules (SMs) in the arms of an MMC-based BESS is necessary in certain applications. The main contribution of this paper is to present a general analysis of the inter-SM active power disparity problem which incorporates the inherent operational constraints of the MMC converter. An analytical method to derive inter-SM active power disparity limits is introduced. The proposed analysis can help facilitate the design of MMC-based BESS for applications such as recycled batteries and hybrid battery chemistries, which can both require significant inter-SM active power disparity. The analysis formulates a criteria vector and criterion value that describes whether an MMC-based BESS is capable of supplying demanded output powers while subject to inter-SM active power disparity. Simulation and experimental results are obtained on a single-phase system with varying numbers of SMs per arm, which verifies the feasibility and generality of the proposed analytic method.

In the operation of battery energy storage systems (BESSs) based on the cascaded H-bridge (CHB) converters, it is desirable to balance the state of charge (SoC) among the submodules (SMs) within each phase arm. However, there are constraints on the active power distribution among the SMs that intend to balance the SoC. For instance, each SM in a practical system is designed with a specific power rating and can be damaged if it is exceeded. Therefore, a novel rule-based method is proposed in this article to achieve a fast SoC balancing, while respecting the power constraints of the SMs. The proposed method is general and readily applicable to hybrid BESSs, where the SMs are integrated with heterogeneous energy storage. Experimental results are obtained to verify the effectiveness of the proposed algorithm and to compare its performance with respect to existing methods.

In a battery energy storage system (BESS) based on the cascaded H-bridge (CHB) converter or modular multilevel converter (MMC), the active power distribution among the submodules (SMs) can be highly unbalanced, especially when batteries with different power ratings are used. Meanwhile, there are certain limits for the unbalanced power distribution among the SMs. It is a challenge to determine whether the active power distribution is viable for cascaded multilevel converters with unequal capacitor voltages because the complexity grows exponentially as the number of SMs increases. This article proposes a generalized analytic solution to this problem, which reduces the complexity significantly and is readily applicable to converters with large number of SMs. The validity of this method is proved theoretically and verified with simulation and experimental results on both the MMC and CHB converter.

The dc solid-state transformer (dcSST) plays a vital role in interconnecting diverse dc sources and loads in dc microgrids. However, output voltage regulation and submodule power balance control have always been two essential control challenges of the dcSST. To address these challenges, this article proposes a multiagent soft actor-critic-based active disturbance rejection control (MASAC-ADRC) method. The ADRC method is used for uncertainty estimation and compensation in modular dcSST. Specifically, the controller gains of the ADRC method are optimized adaptively by the MASAC method, thus enhancing the adaptability to changing conditions from the environment. Compared with the existing controller gain optimization methods, the proposed method does not rely on predetermined datasets. Instead, it provides a tailored strategy, employing a neural network to map optimal ADRC parameters from the measured states. Through the incorporation of diverse environmental scenarios, encompassing variant input voltage and output power, the MASAC-ADRC method achieves superior dynamic performance. Experimental validation underscores the efficacy of the proposed algorithm, showcasing its superiority over alternative approaches. The proposed method yields enhancements exceeding 50% in dynamic performance metrics such as overshoot, settling time, and mean square error when compared with existing methods.

Solar photovoltaic (PV) power generation inherently fluctuates due to erratic weather conditions. Although an energy storage system (ESS) can effectively mitigate these fluctuations, conventional methods require a large ESS capacity to control both increasing and decreasing rates of PV power change. This article explores an opportunity to reduce the required ESS capacity in PV power smoothing applications by delegating the power smoothing during positive irradiance transients to a flexible power point tracking algorithm. Therefore, the ESS only needs to regulate the power ramp rate during negative irradiance transients and it no longer requires the reserved capacity for positive transients. After discharging the ESS, the proposed control fully restores it without violating the allowed ramp rate. The efficacy of the proposed power ramp rate control under rapid irradiance transients is demonstrated experimentally using a laboratory-scale setup. In addition, based on simulated case studies using a specific real-field one-day irradiance profile, the proposed control allows around $64\%$ reduction in the required ESS capacity compared to the conventional ESS-based control.

This article proposes a distributed uniform control approach for a dc solid state transformer (DCSST) that feeds constant power loads. The proposed approach utilizes a multiagent deep reinforcement learning (MADRL) technique to coordinate multiple control objectives. During the offline training stage, each DRL agent supervises a submodule (SM) of the DCSST, and outputs real-time actions based on the received states. Optimal phase-shift ratio combinations are learned using triple phase-shift modulation, and soft actor-critic (SAC) agents optimize the neural network parameters to enhance controller performance. The well-trained agents act as fast surrogate models that provide online control decisions for the DCSST, adapting to variant environmental conditions using only local SM information. The proposed distributed configuration improves redundancy and modularity, facilitating hot-swap experiments. Experimental results demonstrate the excellent performance of the proposed multiagent SAC algorithm.