As the capacity of high-voltage transformerless energy storage systems expands to the scale of hundreds of megawatts, unexpected power deficit transients caused by unplanned outages can severely impact power grid stability. Focusing on high-voltage transformerless energy storage systems based on modular multilevel topology, this article systematically analyzes and categorizes the factors affecting the reliable operation of large-capacity energy storage systems. For faults and warnings emerging in battery systems and battery management systems, a fault isolation strategy that preserves voltage output capability is proposed. Following fault isolation, a system-level fault-tolerant operation control method is developed for energy storage systems under DC-side fault isolation conditions, enabling hybrid coordinated operation between faulty submodules and healthy submodules. Additionally, a circulating current injection strategy is proposed to compensate for power factor deviations in bridge arms caused by hybrid operation. The proposed method ensures that when any faulty battery cluster is taken offline, the remaining energy storage submodules can continue normal charging and discharging operations while maintaining the system power factor stable, thereby significantly enhancing the reliability and security of large-capacity energy storage systems.
1. Introduction
Energy storage systems can balance the fluctuations of wind and solar energy and enhance the stability of the power grid, making them a core element in the flexible regulation and active support of new power systems. The vigorous development of energy storage systems is a strategic necessity for achieving the “dual carbon” goals and ensuring energy security. Among various storage technologies, electrochemical energy storage has become the fastest-growing type due to its high energy density, high efficiency, flexible deployment, and environmental friendliness.
Traditional low-voltage energy storage systems have limited single-unit capacity. Constructing large-scale storage power stations requires hundreds of units operating in parallel, monitored by complex information systems. The multi-layer information architecture slows down response speed, and power must be stepped up through two stages of transformers before grid connection, reducing operational efficiency and introducing multi-machine stability risks. High-voltage transformerless battery energy storage systems achieve cluster-level segmentation and management through modular power conversion interfaces. These systems break through the constraints imposed by parameter differences among batteries on the single-unit large-capacity scaling of energy storage systems. Without requiring power frequency transformers, they can be directly connected to 6–35 kV or higher voltage grids, substantially improving the single-unit capacity and conversion efficiency of energy storage systems.
At present, energy storage systems are evolving toward larger single-unit capacities. Under the same modulation ratio, the delta topology increases the number of battery cells and power by a factor of 1.73 compared to the single-star topology; the modular multilevel full-bridge topology doubles the cell count and power compared to the single-star topology; and the modular multilevel half-bridge topology quadruples the cell count and power relative to the single-star topology. However, high-voltage transformerless energy storage equipment not only performs power conversion but also undertakes energy storage and management tasks. The potential risk points and reliability issues are more prominent compared to purely power electronic equipment. In particular, large-capacity grid-forming high-voltage transformerless energy storage systems play an important role in supporting new power systems, necessitating a re-examination of fault online processing methods to improve equipment operational reliability.
For modular multilevel power electronic converters, existing submodule fault-tolerant operation methods can be classified into hardware redundancy and software control. Hardware redundancy adds redundant submodules that can be switched in when faults occur, but this increases cost and reduces efficiency while offering limited fault tolerance capability. Software control strategies modify relevant controller parameters after faulty submodules are removed, primarily focusing on objectives such as neutral point shift control, inter-phase state-of-charge balancing, and modulation wave correction. However, these methods do not classify faults according to the types and characteristics specific to energy storage systems, making it difficult to meet the reliability improvement requirements of large-capacity high-voltage transformerless energy storage systems.
Unlike traditional high-voltage transformerless reactive power compensators and modular multilevel converter valves—which are purely power electronic devices—high-voltage transformerless battery energy storage systems contain additional potential failure points. Beyond switching elements and passive components, the battery system serving as the energy carrier and the battery management system (BMS) monitoring system operation introduce a wider variety of fault information and alarm types. The probability of failure and the frequency of alarm events requiring handling are considerably higher. The operational reliability of energy storage systems depends heavily on three dimensions of reliability: power electronic devices, battery systems, and BMS. I have therefore undertaken this study to systematically address these reliability challenges.
When a battery cluster or BMS fails or issues warnings, my proposed strategy uses the isolation switches configured in the BMS high-voltage box to disconnect the battery cluster and BMS from the power unit, achieving fault isolation on the DC side. Since both the positive and negative busbars of the battery cluster are equipped with circuit breakers from the BMS system, no additional hardware cost is incurred. After the battery cluster is disconnected, the power unit can continue operating in a reactive power mode, fully utilizing its voltage support capability to the AC side.
Compared to previous fault tolerance methods that directly bypass the faulty submodule at the AC side, my proposed strategy preserves the voltage output capability of the faulty submodule’s power unit and coordinates the operation between faulty and healthy submodules. The active power of the system is appropriately reduced after the fault, and a circulating current injection strategy is employed to compensate for power factor deviations between the upper and lower bridge arms, without changing the system’s power factor. The proposed method enables the modular multilevel high-voltage transformerless energy storage system to achieve isolation and removal of any number of faulty battery clusters while the remaining non-faulty submodules continue charging and discharging operations.
This article is organized as follows. Section 2 presents the statistical analysis and classification of factors affecting energy storage system reliability. Section 3 describes the system topology and the proposed battery cluster fault isolation strategy. Section 4 details the fault-tolerant operation control method for energy storage systems after fault isolation. Section 5 presents the simulation verification under various operating conditions. Section 6 provides the concluding remarks.
2. Factors Affecting Reliable Operation of Energy Storage Systems
The high-voltage single-unit large-capacity energy storage system concentrates stored energy to a greater degree, making safe and stable operation critically important. Unlike traditional high-voltage transformerless reactive power compensators and modular multilevel converter valves—which are purely power electronic devices—high-voltage transformerless battery energy storage systems have more potential failure points. In addition to switching elements and passive components in the converter, the battery system serving as the energy carrier and the BMS may exhibit more diverse fault information and alarm types, with higher failure probabilities and more frequent alarm events requiring handling. Traditional fault handling and redundancy strategies directly perform AC-side bypass removal of faulty submodules when redundant submodules are available. In the absence of submodule redundancy capability, the system resorts to power reduction or shutdown.
Based on operational data from an actual high-voltage transformerless energy storage power station, I have summarized and analyzed the fault types and occurrence probabilities of each component in Table 1, excluding certain human factors. The statistics reveal that with the advancement of semiconductor power device technology and device fault protection techniques, the probability of power switch failures is significantly lower than the probability of faults and warnings in other components of the energy storage system.
| Fault Source | Fault Category | Fault Information | Probability |
|---|---|---|---|
| Battery System | Cell-level alarms | Cell overvoltage, cell undervoltage, charging/discharging overtemperature, charging/discharging undertemperature, large temperature differences among cells, large temperature variation within cells, large voltage differences among cells, SOC too high, SOC too low, SOC inconsistency among cells | 90% |
| Cluster-level alarms | Cluster total voltage overvoltage, cluster total voltage undervoltage, cluster charging/discharging overcurrent, cluster SOC too low, cluster SOC too high, cluster insulation too low | ||
| System-level alarms | Excessive inter-cluster voltage difference, excessive inter-cluster current difference, excessive inter-cluster capacity difference | ||
| BMS System | Slave control unit (BMU) | BMU communication disconnection, sensor faults | 7% |
| Master control unit (BCM) and total control unit | BCM wiring harness connection abnormality, communication faults, bus matching fault, communication instability | ||
| Power Unit | Switching devices | Open-circuit and short-circuit faults caused by electrical overstress, thermal failure of devices | 2.5% |
| Film capacitors | Failure or open circuit, breakdown short circuit, leakage current |
The battery system, serving as the energy carrier, and the BMS, monitoring the system operating status, produce more fault information and alarm types. The probability of failure and the frequency of alarm events requiring handling are considerably higher. Battery system faults may include cell short circuits, overcharging, overdischarging, overheating, capacitor short circuits, wire disconnection, DC positive-negative short circuits, switch short circuits, and switch open circuits. These faults can, if not handled promptly, lead to system shutdown or even escalation of the incident. If AC-side fault tolerance strategies were still adopted in such scenarios, significantly more energy storage submodules would need to be configured, and over-provisioning the battery capacity of the energy storage system would increase costs.
3. System Structure and Fault Isolation Strategy for Battery Clusters
3.1 Operating Principle of Modular Multilevel Energy Storage Systems
The modular multilevel energy storage converter system comprises six bridge arms. Each bridge arm consists of N identical power modules and bridge arm inductors connected in series. Each power module contains a full-bridge structure and a battery cluster. The typical modular multilevel energy storage structure is illustrated in Figure 1. Here, \(L_0\) represents the bridge arm inductance, and \(L_s\) represents the grid-side inductance.
For the convenience of analysis, let \(u_{sk}\) denote the k-phase output voltage (where \(k = a, b, c\)), \(u_{pk\_j}\) and \(u_{nk\_j}\) denote the output voltage of the j-th module in the upper and lower bridge arms, respectively (for \(j = 1, 2, \dots, N\)). \(u_{pk}\) and \(u_{nk}\) represent the sum of module output voltages in the upper and lower bridge arms of the corresponding phase, and the voltage at the connection point of the upper and lower bridge arms is the phase output voltage. \(i_{pk}\) and \(i_{nk}\) represent the upper and lower bridge arm currents of the corresponding phase, respectively.
For the modular multilevel high-voltage transformerless large-capacity battery energy storage system, applying Kirchhoff’s voltage law yields the voltage relationship of the upper and lower bridge arms:
$$u_{pk} = \frac{u_{sk}}{2} – u_{sk} – L_0 \frac{di_{pk}}{dt} + L_s \frac{di_{sk}}{dt}$$
$$u_{nk} = \frac{u_{sk}}{2} + u_{sk} + L_0 \frac{di_{nk}}{dt} + L_s \frac{di_{sk}}{dt}$$
Define the common-mode voltage \(u_{comk}\) of the upper and lower bridge arms and the equivalent output voltage \(u_k\) of the modular multilevel energy storage converter as:
$$u_{comk} = \frac{u_{nk} + u_{pk}}{2}$$
$$u_k = \frac{u_{nk} – u_{pk}}{2}$$
Define \(i_{sk}\) as the grid current of phase k, and \(i_{diff\_k}\) as the circulating current between the upper and lower bridge arms of phase k:
$$i_{sk} = i_{pk} – i_{nk}$$
$$i_{diff\_k} = \frac{i_{pk} + i_{nk}}{2}$$
From these definitions, the relationship between the circulating current and common-mode voltage of the energy storage converter system can be expressed as:
$$u_{comk} = -\frac{L_0}{2} \frac{di_{diff\_k}}{dt}$$
$$u_k = -\frac{L_s}{2} \frac{di_{sk}}{dt} + \frac{u_{sk}}{2} + u_k$$
The above equations demonstrate that circulating current control can be achieved by controlling the common-mode voltage. With the grid voltage unchanged, controlling the output current enables control of the equivalent output voltage.
3.2 Battery Cluster Fault Isolation Strategy with Voltage Output Capability Preservation
For faults or alarms in battery clusters or BMS, as shown in Figure 1, the isolation switch configured in the BMS high-voltage box is used to disconnect the battery cluster and BMS from the power unit, achieving fault isolation on the DC side. Since both the positive and negative busbars of the battery cluster are equipped with circuit breakers from the BMS system, the system requires no additional hardware cost. Moreover, after the battery cluster is disconnected, the power unit can still operate in reactive power mode, fully utilizing its voltage support capability on the AC side.
Figure 3 illustrates the implementation logic timing diagram of the proposed faulty battery cluster DC-side exit strategy, which comprises three stages:
Stage 1 (t0–t1): Before executing fault handling, the submodule operates in charging or discharging mode. The battery current and capacitor voltage both contain a DC bias component and a second-harmonic component, which is an inherent characteristic of the H-bridge single-phase structure.
Stage 2 (t1–t2): Upon receiving the alarm information from the battery cluster or BMS at time t1, the modulation voltage of the faulty submodule is modified to be perpendicular to the phase current. At this point, the battery cluster has stopped charging and discharging, the DC bias of the battery current becomes zero, creating zero-crossing conditions for the battery branch current.
Stage 3 (t2–t3): When the battery branch current is detected to be zero at time t2, the submodule controller controls the DC circuit breaker to disconnect the battery cluster from the power converter section, achieving fault exit of the battery cluster. The H-bridge converter section continues operating in reactive power mode to support the grid voltage.
4. Fault-Tolerant Operation Control After Battery Cluster Fault Isolation
Based on the fault handling approach proposed in the previous section, the fault-tolerant operation control of the system after fault occurrence and removal needs further consideration to ensure that the system maintains the charging and discharging power command as much as possible after any number of battery faults occur, without changing the system’s external power factor.
4.1 Basic Principle of System Fault-Tolerant Operation
To simplify the control strategy of system fault-tolerant operation, I adopt the symmetric exit strategy for battery clusters at the same position in the three-phase bridge arms. Define \(v_{nk\_P}\) and \(v_{nk\_Q}\) as the modulation voltage signals of healthy submodules and faulty submodules in the lower bridge arm of phase k, \(v_{pk\_P}\) and \(v_{pk\_Q}\) as the modulation voltage signals of healthy submodules and faulty submodules in the upper bridge arm of phase k, and \(i_{pk}\) and \(i_{nk}\) as the upper and lower bridge arm currents of phase k.
After the battery cluster is isolated and removed, the faulty submodule operates in reactive power mode. Its modulation voltage and bridge arm current vector have a perpendicular relationship. From the perspective of the upper and lower bridge arms individually, the power factor of the bridge arm changes after fault removal. However, by injecting circulating current to achieve mutual compensation of reactive power between the upper and lower bridge arms, the overall power factor of the system remains unaffected.
Considering the general case, define \(N_{nom\_p}\) and \(N_{nom\_n}\) as the number of non-faulty modules in the upper and lower bridge arms, and define the number of faulty modules in the upper and lower bridge arms as \(N_{fal\_p}\) and \(N_{fal\_n}\), respectively.
After a fault occurs, to maintain the set power command as much as possible, the active power command of the system is adjusted to:
$$P_{set} = \min\left\{\frac{N_{nom\_p} + N_{nom\_n}}{2N} , 1\right\} \cdot P_{nom}$$
where \(P_{nom}\) is the rated active power of the system and \(P\) is the active power command before the fault. The active power allocation to the upper and lower bridge arms of each phase can be expressed as:
$$P_{set\_k\_p/n} = \frac{N_{nom\_p/n}}{N_{nom\_p} + N_{nom\_n}} \cdot P_{set}$$
After the battery cluster is isolated and removed, the faulty submodule can continue providing voltage as a reactive power module. The maximum reactive voltage output amplitude of the bridge arm faulty module is:
$$V_{Q\_p/n\_max} = N_{fal\_p/n} \cdot V_{fal\_p/n}$$
The reactive voltage output of the system’s faulty submodules should satisfy:
$$V_{Q\_p/n\_max} > V_{sk} – N_{nom\_p/n} \cdot V_{bat}$$
Considering that the number of faults in the upper and lower bridge arm modules is completely random, it is necessary to inject circulating current to achieve power redistribution between the upper and lower bridge arms within the same phase, ensuring the system maintains normal operation while fully utilizing voltage capability. Without loss of generality, assume that the number of faults in the lower bridge arm is greater than that in the upper bridge arm, i.e., \(N_{fal\_n} > N_{fal\_p}\). According to the fault severity, the system operating states are classified into three fault-tolerant operating conditions.
4.2 Operating Mode I: Small Number of Faulty Submodules
When the overall number of faulty submodules is relatively small, the reactive submodules of the lower bridge arm (which has more faults) output as much reactive voltage as possible to prevent over-modulation of its non-faulty submodules. In Mode I, the effective value of the lower bridge arm reactive voltage is expressed as:
$$V_{nk\_Q} = \frac{(N_{nom\_n} – N_{fal\_n}) \cdot V_{bat}}{2}$$
In the analysis of the upper and lower bridge arm operating points, ignoring the influence of bridge arm inductance and grid inductance, the vector diagram of the upper and lower bridge arm voltages and currents and the derivation sequence of voltage and current references in Mode I are shown in Figure 5.
From the voltage and current vector diagram in Figure 5, it can be observed that the active voltage of the lower bridge arm maintains the same phase as the current, and the reactive voltage is perpendicular to the current. According to the Pythagorean theorem, the lower bridge arm’s active voltage effective value and phase angle can be determined:
$$V_{nk\_P} = \sqrt{V_{sk}^2 – V_{nk\_Q}^2}$$
$$\varphi_{nk} = \arcsin\left(\frac{V_{nk\_P}}{V_{sk}}\right)$$
Based on the active power of the lower bridge arm, the current reference of the n bridge arm can be derived, with the current phase angle being the same as the corresponding bridge arm’s active voltage. The effective value of the lower bridge arm current is derived as:
$$I_{nk} = \frac{P_{set\_k\_n}}{V_{nk\_P}}$$
After the amplitude and phase of the lower bridge arm current are determined, according to Kirchhoff’s current law, the upper bridge arm current vector is the vector difference between the grid current and the lower bridge arm current, as shown in Figure 5. The amplitude relationship among the three satisfies the law of cosines for triangles, and the phase angle satisfies the law of sines. From this, the effective value and phase angle of the upper bridge arm current are:
$$I_{pk} = \sqrt{I_{nk}^2 + I_{sk}^2 – 2I_{nk} \cdot I_{sk} \cdot \cos(\varphi_{nk})}$$
$$\varphi_{pk} = \arcsin\left(\frac{I_{nk} \cdot \sin(\varphi_{nk})}{I_{pk}}\right)$$
Since the reactive voltage of the upper bridge arm is perpendicular to the bridge arm current, the active and reactive voltage effective values of the p bridge arm can be calculated by solving the right triangle:
$$V_{pk\_Q} = V_{sk} \cdot \sin(\varphi_{pk})$$
$$V_{pk\_P} = V_{sk} \cdot \cos(\varphi_{pk})$$
Mode I starts the derivation from the lower bridge arm reactive voltage. To control the generation of circulating current, the amplitude and phase reference values of the circulating current are:
$$I_{diff} = \sqrt{I_{nk}^2 + I_{pk}^2 – 2I_{nk} \cdot I_{pk} \cdot \cos(\varphi_{nk} + \varphi_{pk})}$$
$$\varphi_{diff} = \arcsin\left(\frac{I_{pk} \cdot \sin(\varphi_{pk})}{I_{diff}}\right) + \varphi_{nk}$$
By setting the operating points of the upper and lower bridge arms according to the above analysis, when both \(N_{nom\_kp}\) and \(N_{nom\_kn}\) are large and their difference is not significant, the reactive power output from the faulty modules can be ensured to flow between the upper and lower bridge arms within the phase.
4.3 Operating Mode II: Larger Numbers of Faulty Submodules with Relatively Balanced Distribution
Mode II takes the bridge arm current equal to the rated value as the starting condition. First, set the P bridge arm current to \(I_{nom\_p}\), and then derive the active and reactive voltage effective values of the P bridge arm:
$$V_{pk\_P} = \frac{P_{set\_k\_p}}{I_{pk}}$$
$$V_{pk\_Q} = \sqrt{V_{sk}^2 – V_{pk\_P}^2}$$
In this mode, the vector relationship that the upper and lower bridge arm voltages and currents must satisfy is shown in Figure 6. Since the active voltage and reactive voltage vectors are perpendicular, the power factor angle of the P bridge arm can be obtained from the active and reactive voltage effective values:
$$\varphi_{pk} = \arcsin\left(\frac{V_{pk\_Q}}{V_{sk}}\right)$$
After the P bridge arm power factor angle and current magnitude are determined, according to Kirchhoff’s current law, the N bridge arm current vector is the vector sum of the grid current and the P bridge arm current. As shown in Figure 6, the amplitude relationship among the three satisfies the law of cosines for triangles, and the phase angle satisfies the law of sines. From this, the effective value and power factor angle of the N bridge arm current are:
$$I_{nk} = \sqrt{I_{pk}^2 + I_{sk}^2 – 2I_{pk} \cdot I_{sk} \cdot \cos(\varphi_{pk})}$$
$$\varphi_{nk} = \arcsin\left(\frac{I_{pk} \cdot \sin(\varphi_{pk})}{I_{nk}}\right)$$
After determining the power factor angle of the N bridge arm, the active and reactive voltages of the N bridge arm are calculated by combining the rated voltage and power angle:
$$V_{nk\_Q} = V_{sk} \cdot \sin(\varphi_{nk})$$
$$V_{nk\_P} = V_{sk} \cdot \cos(\varphi_{nk})$$
Based on the above analysis, by reasonably setting the operating points of the upper and lower bridge arms, when the difference between \(N_{nom\_kp}\) and \(N_{nom\_kn}\) becomes more significant, two key objectives can be achieved: strictly limiting the bridge arm current to not exceed the rated value to avoid overcurrent risk, and ensuring that the reactive power generated by the faulty modules can effectively circulate between the upper and lower bridge arms within the phase.
4.4 Operating Mode III: Large Numbers of Faulty Submodules Concentrated in One Bridge Arm
This situation typically occurs when the system demand reaches the maximum power command and \(N_{nom\_kp}\) and \(N_{nom\_kn}\) exhibit significant disparity. The occurrence of this mode is closely related to the rated current and voltage modulation ratio of the system. In this case, power reduction operation must be adopted to ensure the system maintains grid-connected operation. The vector diagram of the upper and lower bridge arm voltages and currents is shown in Figure 7. Mode III starts deriving other voltage and current command values from the upper bridge arm current and lower bridge arm active voltage.
In this scenario, priority should be given to setting the active voltage command of the more severely faulty N bridge arm, providing as much voltage support as possible based on the limited number of normal operating modules, while adjusting the overall system power level to adapt to the operational constraints under fault conditions, ensuring safe and stable continuous grid-connected operation. Although this power-reduction strategy temporarily prevents the system from reaching rated power output, it can effectively avoid over-modulation and overcurrent phenomena, ensuring continuous availability under severe fault conditions.
At this time, the healthy modules of the N bridge arm with severe faults are all devoted to active voltage output. The active voltage command is:
$$V_{nk\_P} = \frac{N_{nom\_n} \cdot E_{bat}}{2}$$
where \(E_{bat}\) is the battery cluster voltage of the healthy running modules. After determining the active voltage of a single bridge arm, the reactive voltage and power factor angle of the N bridge arm can be obtained:
$$V_{nk\_Q} = \sqrt{\frac{(V_{sk} – V_{nk\_P})^2}{2}}$$
$$\varphi_{nk} = \arcsin\left(\frac{V_{nk\_Q}}{V_{sk}}\right)$$
The bridge arm current command remains consistent with Mode II, where the P bridge arm current of the less severely faulty arm is limited to:
$$I_{pk} = I_{nom\_p}$$
To ensure that the reactive power of the energy storage system only circulates internally without affecting the external grid, according to the reactive power balance condition, the relationship between the N bridge arm current and the P bridge arm power factor angle can be calculated:
$$\varphi_{pk} = \frac{V_{sk} \cdot I_{nom\_p} + E_{bat}^2 \cdot N_{nom\_n}^2 – E_{bat} \cdot N_{nom\_n} \cdot V_{sk}}{I_{pk} \cdot V_{sk}}$$
At this point, the P bridge arm active and reactive voltage calculations follow the equations presented for Mode I. Under Mode III, the system maintains the voltage level unchanged as much as possible while outputting the maximum possible power under the dual constraints of bridge arm current boundary and active voltage boundary. The actual output power of the upper and lower bridge arms is:
$$P_{set\_k\_n} = I_{nk} \cdot V_{nk\_P}$$
$$P_{set\_k\_p} = I_{pk} \cdot V_{pk\_P}$$
When \(N_{nom\_kp} < N_{nom\_kn}\), i.e., the number of normally operating modules in the upper bridge arm is less than that in the lower bridge arm, the current-voltage-power reconstruction should start from the upper bridge arm, with the method being the same as the above process, exchanging the roles of the upper and lower bridge arms.
4.5 System Fault-Tolerant Control Method for Battery Cluster Exit
When the overall number of faulty submodules is relatively small, Mode I design allows the reactive submodules of the lower bridge arm (with more faults) to output as much reactive voltage as possible to prevent over-modulation of non-faulty submodules. However, when the current amplitude of any bridge arm exceeds the rated value, the designed operating point cannot operate safely, necessitating redesign of the system’s operating mode.
When the number of normally operating modules in the upper and lower bridge arms differs significantly, a single bridge arm may exceed its rated operating current. When the power command reaches the maximum value allowed by the system, according to the bridge arm current calculation equations of Mode I, the upper and lower bridge arm currents can be expressed as:
$$I_{pk} = \frac{2 \cdot N_{nom\_n} \cdot \sqrt{2} \cdot E_{bat}^2 \cdot N_{nom\_n}^2 – (E_{bat} \cdot N_{nom\_n} – V_{sk})^2}{E_{bat} \cdot N_{nom\_n} \cdot \sqrt{4 \cdot E_{bat}^2 \cdot N_{nom\_n}^2 – (E_{bat} \cdot N_{nom\_n} – V_{sk})^2}}$$
$$I_{nk} = \frac{2 \cdot I_{nom} \cdot N_{nom\_n} \cdot \sqrt{2} \cdot E_{bat} \cdot N_{nom\_n}}{\sqrt{4 \cdot E_{bat}^2 \cdot N_{nom\_n}^2 – (E_{bat} \cdot N_{nom\_n} – V_{sk})^2}}$$
Since the sum of reactive power output to the external grid from the upper and lower bridge arms is zero, and their active power is allocated according to the number of healthy operating modules, the upper and lower bridge arm currents always satisfy the following relationship:
$$I_{nk} \cdot \sin(\varphi_{nk}) = I_{pk} \cdot \sin(\varphi_{pk})$$
$$I_{nk} \cdot \cos(\varphi_{nk}) + I_{pk} \cdot \cos(\varphi_{pk}) = \frac{N_{nom\_n} + N_{nom\_p}}{N_{nom\_n}} \cdot I_{nk} \cdot \cos(\varphi_{nk})$$
Through geometric relationship verification, under the unbalanced module number condition of the upper and lower bridge arms, \(I_{pk}\) always exceeds the bridge arm current rating before \(I_{nk}\). As \(N_{nom\_kp}\) and \(N_{nom\_kn}\) decrease, \(I_{pk}\) exceeds the rated bridge arm current:
$$I_{pk} > I_{nom\_p}$$
At this point, the P bridge arm current exceeds the rated value, which does not satisfy the long-term steady-state operation requirement. The system is subsequently classified as Mode II, where voltage commands are redesigned from the current perspective to effectively avoid bridge arm overcurrent. Under Mode II design, the active voltage analytical expressions of the N bridge arm and P bridge arm are:
$$V_{nk\_P} = \frac{V_{sk} \cdot \sqrt{N_{nom\_n}^2}}{\sqrt{N_{nom\_n}^2 + N_{nom\_p}^2 – N_{nom\_n} \cdot N_{nom\_p} \cdot 1}}$$
$$V_{pk\_P} = \frac{V_{sk} \cdot \sqrt{N_{nom\_p}^2}}{\sqrt{N_{nom\_n}^2 + N_{nom\_p}^2 – N_{nom\_n} \cdot N_{nom\_p} \cdot 1}}$$
However, in extreme unbalanced cases of \(N_{nom\_kp}\) and \(N_{nom\_kn}\), even after the aforementioned power reconstruction of the upper and lower bridge arms, the system may still fail to meet the power output requirements because the number of normally operating modules in the N bridge arm is too small:
$$V_{pk\_P} > V_{pk\_P\_max} \quad \text{or} \quad V_{nk\_P} > V_{nk\_P\_max}$$
When the active voltage provided by the bridge arm battery cannot meet the system requirements, the system automatically reduces the power operation level and subsequently transitions into Mode III operation.
In practical engineering applications, to ensure the control system automatically responds quickly to various faults, a unified power reconstruction logic needs to be established. The mode classification logic diagram demonstrates the decision-making process of the system when faced with different faults. The logic framework first determines whether overcurrent occurs after system fault isolation: Mode I for minor faults does not cause overcurrent while ensuring full-power operation; if overcurrent occurs but the bridge arm voltage can support the active voltage requirement, it is classified as Mode II; if the bridge arm active voltage cannot support full-power operation, it is classified as Mode III.
4.6 System Control Block Diagram
Figure 10 presents the fault-tolerant operation control block diagram of the system after arbitrary faulty battery cluster exit. The outer loop includes active/reactive power control modules and circulating current control modules, while the inner loop includes upper and lower bridge arm fault-tolerant control modules. The active/reactive power control is consistent with traditional decoupled control strategies. The circulating current control achieves proportional distribution of active power between the upper and lower bridge arms according to the number of healthy units contained, while also controlling the reactive currents of the upper and lower bridge arms to cancel each other, thereby not affecting the overall external power factor.
To achieve fast and accurate control of three-phase circulating currents, in the rotating coordinate system with the grid phase as reference, decoupled control is used to generate the common-mode voltage \(v_{comk}^{*}\) of the three-phase bridge arms, which is superimposed with the modulation voltage generated by the active and reactive module closed-loop control to obtain the total modulation voltage signals of the upper and lower bridge arms. The dq-axis current commands for circulating current control under the three operating modes are:
Mode I:
$$i_{diff\_k\_d}^{*} = I_{nk} \cdot \cos(\varphi_{nk}) + I_{pk} \cdot \cos(\varphi_{pk})$$
$$i_{diff\_k\_q}^{*} = I_{nk} \cdot \sin(\varphi_{nk}) + I_{pk} \cdot \sin(\varphi_{pk})$$
Mode II:
$$i_{diff\_k\_d}^{*} = I_{nom} \cdot \frac{N_{nom\_n} – N_{nom\_p}}{N_{nom\_n} + N_{nom\_p}}$$
$$i_{diff\_k\_q}^{*} = I_{nom} \cdot \frac{2 \cdot \sqrt{N_{nom\_n} \cdot N_{nom\_p}}}{N_{nom\_n} + N_{nom\_p}}$$
Mode III:
$$i_{diff\_k\_d}^{*} = \frac{I_{nom} \cdot V_{sk} \cdot E_{bat} \cdot N_{nom\_n} – E_{bat}^2 \cdot N_{nom\_n}^2}{E_{bat} \cdot N_{nom\_n} \cdot \sqrt{E_{bat}^2 \cdot N_{nom\_n}^2 + E_{bat}^2 \cdot N_{nom\_p}^2 – E_{bat}^2 \cdot N_{nom\_n} \cdot N_{nom\_p}}}$$
$$i_{diff\_k\_q}^{*} = \frac{I_{nom} \cdot E_{bat} \cdot N_{nom\_p} \cdot \sqrt{E_{bat}^2 \cdot N_{nom\_n}^2 + E_{bat}^2 \cdot N_{nom\_p}^2 – E_{bat}^2 \cdot N_{nom\_n} \cdot N_{nom\_p}}}{E_{bat}^2 \cdot N_{nom\_n}^2 + E_{bat}^2 \cdot N_{nom\_p}^2 – E_{bat}^2 \cdot N_{nom\_n} \cdot N_{nom\_p}}$$
Within the bridge arm, the fault-tolerant control needs to generate modulation voltages for both faulty submodules and healthy submodules. The total modulation voltage of faulty submodules is obtained through dq/abc coordinate transformation based on the previously analyzed operating point, while the modulation voltage of healthy submodules is obtained by subtracting the faulty submodule modulation voltage from the total bridge arm modulation voltage.
Since the battery cluster of the faulty submodule has been removed, the submodule capacitor voltage is no longer clamped by the battery. The capacitor voltage may become excessively high, leading to device overvoltage, or may become too low, resulting in insufficient submodule modulation ratio. Therefore, it is necessary to control and regulate the average value of all faulty submodule capacitor voltages \(V_{Ck\_avg}\) to equal the given value \(V_{C\_ref}\). The k-phase reactive modulation voltage is:
$$v_{kQ}^{*} = \frac{2 \cdot K \cdot (N – N_{nom}) \cdot V_{C\_ref} \cdot N_{nom}}{(N – N_{nom}) \cdot V_C} \cdot \cos(\omega t + \varphi_k)$$
where \(\Delta V_{Ck} = V_{Ck\_ref} – V_{C\_avg}\) is the error between the k-phase capacitor voltage and the given value, and K is the phase capacitor voltage control coefficient.
Real-time balancing of capacitor voltages among faulty submodules within the same bridge arm is achieved by varying the AC voltage amplitude of the reactive submodules. Specifically, the imbalance degree of the capacitor voltage average is used as feedback, and a fundamental voltage regulation component, in-phase or anti-phase with the phase current, is superimposed on the modulation voltage of the unbalanced reactive power modules. This controls the active power absorbed by each power module, achieving capacitor voltage balancing of the reactive modules.
Defining the average capacitor voltage of the i-th module among the \(N – N_{nom}\) faulty submodules in phase k as \(V_{Cki}\), its imbalance degree can be expressed as:
$$\Delta V_{Cki} = V_{Cki} – \frac{1}{N – N_{nom}} \sum_{j=1}^{N-N_{nom}} V_{Ckj}$$
Therefore, the modulation voltage of the i-th faulty reactive module is:
$$v_{kiQ} = K_1 \cdot \Delta V_{Cki} + \frac{v_{kQ}}{N – N_{nom}} \cdot \frac{V_{bat}}{E_{bat}} \cdot \cos(\omega t + \varphi_k)$$
where \(v_{kQ}\) is the k-phase reactive modulation voltage and \(K_1\) is the intra-phase capacitor voltage balancing coefficient.
5. Simulation Verification
To verify the proposed fault-tolerant control strategy for battery clusters in modular multilevel high-voltage transformerless energy storage systems, I constructed a three-phase 10 kV energy storage system simulation model using MATLAB/Simulink. The simulation model structure is consistent with the topological description, and Table 2 lists the main system parameters.
| Parameter | Value |
|---|---|
| AC grid line voltage \(V_{ab}\) / kV | 10 |
| System rated power \(P_{nom}\) / MW | 10 |
| System rated capacity \(W_{nom}\) / MWh | 10 |
| Battery cluster rated voltage \(V_{bat}\) / V | 1200 |
| Number of submodules per bridge arm \(n\) | 14 |
| Grid-side filter inductance \(L_f\) / mH | 3 |
| Control frequency \(f_{control}\) / kHz | 10 |
| DC-side filter inductance \(L_{dc}\) / mH | 2.5 |
| DC-side filter capacitance \(C_{dc}\) / mF | 7.2 |
5.1 Simulation Case 1: Operating Mode I
In the Mode I simulation, at 0.2 s, two faulty battery clusters are removed from the upper bridge arm and four faulty battery clusters from the lower bridge arm. Figure 11 illustrates the system currents, upper and lower bridge arm currents, circulating current, and active and reactive power. After the fault occurs at 0.2 s, the three-phase currents maintain stable sinusoidal waveforms. The circulating current immediately increases from nearly zero and stabilizes into a three-phase AC waveform, reflecting that the control strategy activates circulating current control to balance the power distribution between bridge arms.
Simultaneously, the active power of the upper bridge arm decreases smoothly from the initial 5 MW to approximately 4.29 MW, and the lower bridge arm to approximately 3.57 MW, indicating that the system operates with expected derating. The reactive power of the upper and lower bridge arms rapidly adjusts from nearly zero to approximately ±2.5 MVar, ensuring the system’s external reactive power output is zero, maintaining unity power factor operation of the energy storage system.
Figures 12 and 13 show the detailed voltage and current variations of the A-phase upper and lower bridge arms, respectively. The upper bridge arm reactive voltage leads the bridge arm current, while the bridge arm current and active voltage are in phase. In Mode I, the faulty submodule capacitor voltage rapidly rises from near zero and stabilizes at approximately 1200 V, while the healthy submodule voltage remains essentially unchanged. This confirms that faulty submodules have been effectively isolated and switched to reactive power output mode. For the lower bridge arm, the reactive voltage lags the bridge arm current, and the bridge arm current is in phase with the active voltage. Compared to the upper bridge arm, the lower bridge arm has larger reactive voltage, and the difference between active and reactive voltage amplitudes is smaller for the lower bridge arm. The lower bridge arm current is smaller than the upper bridge arm current. These simulation results align well with the vector diagram and analytical calculations.
5.2 Simulation Case 2: Operating Mode II
For Mode II simulation, at 0.2 s, two faulty battery clusters are removed from the upper bridge arm and six faulty battery clusters from the lower bridge arm. Figure 14 shows the system currents, upper and lower bridge arm currents, circulating current, and active and reactive power under Mode II fault-tolerant operation.
When the system experiences more severe faults—using the example of two battery clusters removed from the upper bridge arm and six from the lower bridge arm—the system maintains stable operation with unity power factor. The three-phase currents remain stable after the fault occurs at 0.2 s, and the upper and lower bridge arm currents exhibit notable differences, directly related to the increased asymmetry in fault numbers between the upper and lower bridge arms. The circulating current after the fault is larger than that in Mode I, indicating that a greater fault asymmetry requires a larger circulating current to maintain reactive power balancing.
Figures 15 and 16 respectively display the voltage and current variations of the A-phase upper and lower bridge arms under Mode II. The faulty submodules begin outputting reactive voltage components at 0.2 s, confirming effective isolation and reactive mode operation. The healthy submodule voltages remain essentially unchanged. For the lower bridge arm, the reactive voltage lags the bridge arm current, and the bridge arm current is in phase with the active voltage. Compared to Mode I, the lower bridge arm reactive voltage further increases, and the gap between active and reactive voltages narrows. The lower bridge arm fault submodule capacitor voltage does not rise significantly. The upper bridge arm current remains at the rated arm current value. These simulation results satisfy the Mode II analytical calculations, and the simulation confirms that the energy storage system maintains unity power factor operation under fault conditions.
5.3 Simulation Case 3: Operating Mode III
For Mode III simulation, at 0.2 s, one faulty battery cluster is removed from the upper bridge arm and twelve faulty battery clusters from the lower bridge arm. Figure 17 shows the system currents, upper and lower bridge arm currents, circulating current, and active and reactive power under Mode III operation.
After the fault occurs at 0.2 s, the three-phase currents remain essentially stable, and the system output reactive power remains zero. The system current is reduced compared to the previous two modes, indicating a decrease in system output power. The upper bridge arm current amplitude decreases only slightly, while the lower bridge arm current amplitude decreases after the fault. The circulating current is established rapidly after the fault, with a larger amplitude than in the previous two modes. This indicates that when the difference in the number of operating modules between the upper and lower bridge arms is significant, a larger circulating current is needed to maintain reactive power flow between the upper and lower bridge arms.
Figures 18 and 19 respectively show the current and voltage variations of the A-phase upper and lower bridge arms under Mode III. The upper bridge arm current waveform remains stable and maintains its rated value after the fault, meeting the system command design requirements. Compared to the previous mode, the number of faulty modules in the upper bridge arm decreases, but the reactive voltage component remains essentially unchanged because the reactive voltage magnitude in Mode III is determined by the more severely faulty bridge arm. The faulty submodule capacitor voltage rapidly rises to approximately 1200 V and remains stable, while the healthy submodule voltage changes little during the fault.
The lower bridge arm current, active voltage, and reactive voltage show that compared to the other two modes, the current amplitude further decreases after the fault, and the lower bridge arm reactive voltage exceeds the active voltage. This is consistent with the vector relationship illustrated in the Mode III diagram. This mode arises mainly from the extremely large difference in the number of faulty modules between the upper and lower bridge arms, so the current difference between the upper and lower bridge arms further increases. In Mode III, the lower bridge arm is required to provide a larger reactive voltage than in Mode II, but the number of faulty modules in the lower bridge arm also increases compared to Mode II, so the faulty module voltage and healthy module voltage do not rise significantly.
5.4 Simulation Case 4: Continuous Transitions Between Operating Modes
In the simulation model, I simulate the random automatic variation of module fault numbers. Figure 20 demonstrates the effectiveness of the system control strategy under various fault conditions.
During the simulation period from 0.2 s to 0.4 s, the system has 2 faulty modules in the upper bridge arm and 3 faulty modules in the lower bridge arm. The control algorithm classifies this as Mode I. At this time, the upper bridge arm active power is 4.29 MW and reactive power is 1.86 MVar; the lower bridge arm active power is 3.93 MW and reactive power is −1.86 MVar.
During the period from 0.4 s to 0.6 s, the system has 2 faulty modules in the upper bridge arm and 6 faulty modules in the lower bridge arm. The control algorithm classifies this as Mode II. At this time, the upper bridge arm active power is 4.29 MW and reactive power is 2.58 MVar; the lower bridge arm active power is 2.86 MW and reactive power is −2.58 MVar. The difference in the number of normally operating modules between the upper and lower bridge arms becomes larger. Under the premise of ensuring that the bridge arm current does not exceed the limit, the control strategy increases the circulating current to achieve reactive power balance between the upper and lower bridge arms, with the system’s external reactive power output always remaining zero.
During the period from 0.6 s to 0.8 s, the system has 1 faulty module in the upper bridge arm and 12 faulty modules in the lower bridge arm. The control algorithm classifies this as Mode III. At this time, the upper bridge arm active power is 4.44 MW and reactive power is 2.29 MVar; the lower bridge arm active power is 0.68 MW and reactive power is −2.29 MVar.
The simulation demonstrates that the system can maintain unity power factor to the grid under all operating modes, achieve mutual compensation of reactive power between the upper and lower bridge arms through circulating current control, and ensure zero external reactive power. Simultaneously, active power is reasonably allocated according to the ratio of normally operating module numbers in each bridge arm, maximizing system output power under fault conditions. This validates the proposed fault-tolerant control strategy for battery clusters in modular multilevel high-voltage transformerless energy storage systems.
A summary of the three operating modes and their key characteristics is presented in Table 3.
| Operating Mode | Fault Distribution | Key Control Principle | Active Power | Reactive Power | Bridge Arm Current |
|---|---|---|---|---|---|
| Mode I | Small number of faulty submodules | Reactive voltage output maximized for the more faulty bridge arm | Proportional to healthy module count | Internal circulation, zero external | Below rated value |
| Mode II | More faulty submodules, relatively balanced distribution | Bridge arm current limited to rated value | Reduced from rated | Internal circulation, zero external | One arm at rated, other below |
| Mode III | Many faults concentrated in one bridge arm | Maximum voltage support with power reduction | Significantly reduced | Internal circulation, zero external | Less severe arm at rated |
The simulation results presented in Table 4 quantitatively compare the key performance metrics of the energy storage system under different operating modes during the fault period.
| Quantity | Mode I | Mode II | Mode III |
|---|---|---|---|
| Faulty submodules in upper arm \(N_{fal\_p}\) | 2 | 2 | 1 |
| Faulty submodules in lower arm \(N_{fal\_n}\) | 4 | 6 | 12 |
| Upper arm active power / MW | 4.29 | 4.29 | 4.44 |
| Lower arm active power / MW | 3.93 | 2.86 | 0.68 |
| Upper arm reactive power / MVar | 1.86 | 2.58 | 2.29 |
| Lower arm reactive power / MVar | −1.86 | −2.58 | −2.29 |
| External reactive power / MVar | 0 | 0 | 0 |
| System power factor | 1 | 1 | 1 |
| Faulty submodule capacitor voltage / V | ≈1200 | ≈1200 | ≈1200 |
The comprehensive simulation studies across the three operating modes and the continuous transition case confirm the effectiveness of the proposed fault-tolerant control strategy. The energy storage system maintains stable three-phase currents, unity power factor, and proper reactive power circulation under all fault scenarios, while ensuring that healthy battery clusters continue normal charging and discharging operations. The faulty submodule capacitor voltages remain well-regulated at the reference value, preventing overvoltage and undervoltage conditions regardless of the fault severity.
6. Conclusion
This article addresses the operational reliability challenges in single-unit large-capacity high-voltage transformerless energy storage systems caused by battery cluster and BMS faults. The main conclusions are summarized as follows:
1) A new battery cluster fault isolation method is proposed. By utilizing the isolation switches configured by the battery management system on the positive and negative terminals of the battery cluster, the faulty battery cluster is disconnected from the power unit. After the faulty battery cluster exits, the H-bridge section can output reactive voltage to support the grid-side voltage, avoiding the over-modulation risk that traditional AC bypass strategies often trigger.
2) After faulty battery cluster isolation and removal, the proposed system-level fault-tolerant control based on mutual reactive power compensation between the upper and lower bridge arms preserves the voltage output capability of the faulty submodule’s power unit and coordinates operation with healthy submodules, without changing the system’s external power factor. The active power is appropriately reduced and allocated proportionally to the number of healthy operating modules in each bridge arm, while the circulating current injection strategy compensates for the power factor deviations.
3) Simulation studies under three different fault distribution scenarios and continuous mode transitions verify that the proposed control strategy enables arbitrary faulty battery clusters to exit without affecting the normal charging and discharging operations of other energy storage submodules or the system power factor. The system maintains unity power factor operation under all fault conditions, significantly enhancing the operational reliability and online operation time of large-capacity energy storage systems.
The proposed fault-tolerant control strategy provides a practical engineering solution for improving the reliability and availability of high-voltage transformerless energy storage systems, which are increasingly important for supporting new power systems with high penetration of renewable energy sources.