In recent years, the rapid expansion of the battery energy storage system market has driven the need for more robust and scalable management architectures. I have observed that centralised energy storage solutions, which connect multiple battery clusters in parallel to a single power conversion system, face significant challenges such as cell imbalance, circulating currents, and complex thermal management. To address these issues, the three-level BMS architecture has emerged as a practical and cost-effective approach. In this paper, I provide a detailed analysis of the three-level BMS architecture applied in a battery energy storage system, covering its hardware composition, functional hierarchy, protection mechanisms, and operational workflow. I also include extensive tables and mathematical formulations to illustrate key parameters and control strategies.
1. Introduction
The battery energy storage system is a critical component in modern power grids, enabling peak shaving, frequency regulation, and renewable energy integration. Two mainstream topologies prevail: centralised and string-based. The centralised topology, widely adopted in early large-scale projects, connects multiple battery clusters (each formed by series-connected battery packs) in parallel to a single bi-directional inverter. However, this configuration suffers from the “barrel effect” — the weakest cluster limits the overall performance — and introduces circulating currents between clusters. The string-based topology, where each cluster has its own inverter, reduces these issues but increases system cost and complexity. For commercial and industrial applications with capacities ranging from several hundred kilowatt-hours to a few megawatt-hours, the centralised architecture remains prevalent due to its lower cost and simpler component selection. In such systems, the three-level BMS architecture provides a hierarchical control framework that ensures safe and efficient operation of the battery energy storage system.
I define the three levels as follows: Battery Management Unit (BMU) at the pack level, Battery Cluster Unit (BCU) at the cluster level, and Battery Array Unit (BAU) at the system level. This layered design enables independent monitoring and control from individual cells to the entire battery energy storage system.
2. System Composition of a Centralised Battery Energy Storage System
A typical centralised battery energy storage system consists of two main enclosures: the energy storage enclosure and the power conversion enclosure. The energy storage enclosure houses the battery packs, high-voltage boxes, a DC busbar cabinet, auxiliary equipment (HVAC, fire suppression, access control, water-leak sensors), and the three-level BMS. The power conversion enclosure contains the centralised PCS (power conversion system) and a step-up transformer, along with an energy management system (EMS) that implements scheduling strategies like peak shaving and green energy consumption.
The three-level BMS manages the entire DC side of the battery energy storage system, ensuring safe operation, thermal management, monitoring, and protection. The hierarchical structure is illustrated conceptually in the following figure, which shows the physical layout of the system.
3. Three-Level BMS Architecture Topology
The three-level BMS architecture partitions the battery energy storage system into three tiers: battery packs (or modules), battery clusters, and the battery array. Each pack consists of cells, a cell monitoring board (BMU), and wiring harnesses. Multiple packs are connected in series to form a cluster, with a high-voltage box (containing the BCU) at the cluster output. Several clusters are then paralleled via a DC busbar cabinet (containing the BAU) to form the complete battery array.
3.1 Battery Pack (Level 1 – BMU)
The BMU is responsible for acquiring each cell’s voltage and temperature, executing passive cell balancing, and detecting wiring faults (e.g., broken voltage or temperature sensor wires). Each BMU provides one fan control output for local thermal management within the pack. Communication with the BCU is via a CAN interface, which also supports firmware upgrades. The measurement ranges are summarised in the table below.
| Parameter | Range / Value |
|---|---|
| Cell voltage measurement | 0 – 5 V |
| Temperature sensor range | -40°C to +125°C (typical NTC) |
| External supply voltage | 12 / 24 VDC |
| Number of cell channels | Typically 4–16 per BMU |
| Passive balancing current | 50–100 mA (resistor-based) |
| Communication interface | CAN 2.0B, up to 1 Mbps |
During operation, the BMU periodically sends cell voltage and temperature data to the BCU. The mathematical relationship between the measured analog values and the digital representation is linear. For a 12-bit ADC with reference voltage of 5 V, the cell voltage is given by:
$$ V_{cell} = \frac{ADC_{raw}}{4096} \times 5.0 \, \text{V} $$
3.2 High-Voltage Box (Level 2 – BCU)
The high-voltage box (or cluster control box) contains the BCU, main circuit breaker, main positive and negative relays, pre-charge resistor and relay, Hall-effect current sensor, fuse, and an isolated DC/DC converter that steps down the cluster voltage to 24 VDC for auxiliary loads. The BCU acts as the cluster-level manager, performing the following tasks:
- Collecting BMU data to compute state of charge (SOC), state of health (SOH), maximum/minimum cell voltage and temperature, and insulation resistance.
- Implementing a three-level alarm protection scheme (mild, moderate, severe) with configurable actions: warning, current derating (50%, 80%, or 100%), or high-voltage shutdown.
- Controlling the pre-charge sequence to limit inrush current when connecting to the DC bus.
- Measuring total cluster voltage, positive bus voltage (P+), and insulation resistance to ground.
- Monitoring relay auxiliary contacts for weld detection.
The BCU communicates with BMUs via CAN1, with the BAU via CAN2, and with external monitoring/EMS via an RS485 interface. It also has multiple programmable digital outputs for relay control and status indication.
| Alarm Category | Examples | Default Action |
|---|---|---|
| Severe (Level 1) | Cell over-voltage, short circuit, insulation failure | Immediate high-voltage disconnect |
| Moderate (Level 2) | High temperature, SOC low, over-current (charge) | Current derating (80%) |
| Mild (Level 3) | Cell voltage difference, communication loss | Warning only |
The full list of alarms supported by a typical BCU includes: total voltage too high/low, cell voltage too high/low, large cell voltage difference, charge/discharge temperature out of range, large temperature difference, rapid temperature rise, SOC low, charge/discharge over-current, low insulation resistance, cell voltage sampling anomaly, temperature sampling anomaly, current sampling anomaly, high-voltage acquisition anomaly, main positive/negative relay weld, BMU communication alarm, maximum charge/discharge current limit exceeded, high-voltage box over-temperature, pre-charge failure, discharge temperature out of range, emergency stop, and more.
The BCU operational state machine follows a deterministic sequence. I describe it mathematically as follows. Let the bus voltage after the main negative relay closes be \(V_{bus}\). Before pre-charge, \(V_{bus} \approx 0\). The pre-charge circuit consists of a resistor \(R_{pre}\) and the system capacitance \(C_{bus}\). The voltage rise is governed by:
$$ V_{bus}(t) = V_{bat} \left(1 – e^{-t / (R_{pre} C_{bus})}\right) $$
The pre-charge period ends when either \(V_{bus}(t) \ge 0.95 \cdot V_{bat}\) or a timeout expires (typically 0.5–2 seconds). Then the main positive relay engages, and after a 500 ms delay, the pre-charge relay is opened. The BCU transitions from STANDBY → SELF_TEST → PRE_CHARGE → HV_ON → NORMAL. If any severe alarm occurs during any phase, the system enters HV_OFF and disconnects.
3.3 DC Busbar Cabinet (Level 3 – BAU)
The DC busbar cabinet is the top-level aggregation point. It houses a circuit breaker for each cluster input, a main output circuit breaker (controlled by the BAU), status indicators, and the BAU module itself. The BAU serves as the brain of the entire battery energy storage system, managing cluster parallelisation, overall system protection, and communication with external devices such as the PCS, EMS, HVAC, and fire suppression system.
Key functions of the BAU include:
- Collecting data from all BCUs via CAN (SOC, voltage, current, min/max cell voltages and temperatures, insulation resistance, alarm status).
- Orchestrating the parallel connection of clusters: start with the cluster having the lowest voltage, then sequentially pre-charge and connect higher-voltage clusters.
- Closing the main output breaker after all clusters are successfully paralleled.
- Reporting aggregated battery array data to the PCS (max charge/discharge current, SOC, voltage, min/max cell voltage and temperature) for power regulation and protection.
- Communicating with the EMS via a LAN interface, and with auxiliary devices (HVAC, fire panel, water-leak sensors, door switches) via RS485.
- Logging system data to internal eMMC Flash, external SD card, or USB drive, with storage capacity meeting the national standard of at least 120 days.
The BAU also implements thermal management by adjusting the HVAC operating state based on the average cell temperature and the system power level. The control logic can be expressed as:
$$ T_{avg} = \frac{1}{N} \sum_{i=1}^{N} T_{i}, \quad P_{sys} = \sum_{j=1}^{M} V_{cluster_j} \cdot I_{cluster_j} $$
where \(N\) is the total number of monitored temperature points and \(M\) is the number of clusters. The HVAC mode is selected from a lookup table, e.g.,
| Condition | HVAC Mode |
|---|---|
| \(T_{avg} < 15^\circ\)C | Heating |
| \(15^\circ\)C \(\le T_{avg} \le 25^\circ\)C | Ventilation |
| \(T_{avg} > 25^\circ\)C | Cooling |
4. Cluster Parallelisation Sequence
One of the most critical tasks performed by the BAU is the sequential parallelisation of battery clusters to minimise circulating currents. Upon power-up, the BAU establishes communication with all BCUs. After verifying that no critical alarms exist, it identifies the cluster with the lowest voltage (denoted as cluster \(k\)). The BAU commands BCU \(k\) to close its main negative relay, then proceed to pre-charge and connect the main positive relay. Once cluster \(k\) is stable on the bus, the BAU selects the next cluster with the next-lowest voltage and commands its pre-charge sequence. The process repeats until all clusters are connected. Finally, the BAU closes the main output breaker, signalling to the EMS and PCS that the battery energy storage system is ready for operation.
If an alarm occurs during parallelisation, the BAU aborts the sequence and may disconnect all clusters. The mathematical condition for a safe connection between a new cluster (voltage \(V_c\)) and the already stabilised bus (voltage \(V_{bus}\)) is that the voltage difference must be below a threshold \(\Delta V_{max}\) (typically 5% of nominal voltage), otherwise the pre-charge circuit will limit the transient current:
$$ I_{inrush} = \frac{|V_c – V_{bus}|}{R_{pre}} \le I_{limit} $$
where \(I_{limit}\) is set by the relay and fuse ratings.
5. Comparison of Centralised and String-Based Architectures
To contextualise the three-level BMS, I provide a comparison between centralised and string-based topologies for a typical commercial battery energy storage system of 1 MWh.
| Feature | Centralised (with 3-level BMS) | String-based |
|---|---|---|
| Number of PCS units | 1 large PCS | Multiple small PCS (one per cluster) |
| DC bus configuration | Multiple clusters in parallel | Each cluster isolated; AC side paralleled |
| BMS complexity | Three-level (BMU/BCU/BAU) | Two-level typically (BMU + cluster controller) |
| Efficiency | Slightly lower due to cluster imbalance | Higher (no circulating current, independent MPPT) |
| Cost | Lower (shared PCS and transformer) | Higher (multiple PCS, more BMS controllers) |
| Scalability | Moderate; limited by parallel cluster count (typically 2–5) | Excellent; plug-and-play clusters |
| Maintenance | Centralised monitoring; one point of failure | Modular; single cluster failure does not stop system |
Despite the advantages of the string-based topology, the centralised architecture with a three-level BMS remains widely used in commercial battery energy storage system installations due to its lower initial investment and simpler integration with existing power conversion equipment. The three-level BMS effectively mitigates the issues of cluster imbalance through intelligent sequential connection and active/passive balancing.
6. Protection and Safety Considerations
Safety is paramount in any battery energy storage system. The three-level BMS provides layered protection. At the BMU level, individual cell over-voltage, under-voltage, and over-temperature are detected and reported. At the BCU level, cluster-level protection includes total voltage limits, current limits, insulation monitoring, relay weld detection, and communication integrity checks. At the BAU level, system-level protection coordinates all clusters and interfaces with external safety devices such as fire suppression panels and emergency stop buttons.
I formalise the insulation monitoring principle. The BCU measures the positive bus-to-ground resistance \(R_{P+}\) and negative bus-to-ground resistance \(R_{N-}\) using a balanced bridge method. The equivalent insulation resistance \(R_{iso}\) is given by:
$$ \frac{1}{R_{iso}} = \frac{1}{R_{P+}} + \frac{1}{R_{N-}} $$
If \(R_{iso}\) falls below a threshold (e.g., 100 \(\Omega\)/V per national standard), the BCU triggers an insulation fault alarm. The BAU may then command a controlled shutdown of the entire battery energy storage system.
7. Communication Architecture
The three-level BMS relies on a hybrid communication network. CAN buses are used for intra-cluster communication between BMUs and BCU, and for inter-cluster communication between BCUs and BAU. RS485 is employed for connecting to auxiliary devices (HVAC, fire panel). The BAU also provides a LAN (Ethernet) port for high-level data exchange with the EMS. The table below summarises the communication interfaces.
| Interface | From | To | Protocol | Purpose |
|---|---|---|---|---|
| CAN1 | BMU | BCU | CAN 2.0B, 250 kbps | Cell data upload, balancing commands |
| CAN2 | BCU | BAU | CAN 2.0B, 250 kbps | Cluster status, BAU commands |
| RS485 | BCU/BAU | External monitoring | Modbus RTU | SCADA, local HMI |
| LAN | BAU | EMS | Modbus TCP / IEC 61850 | System management, scheduling |
| Digital I/O | BCU | Relays, indicators | Dry contact | Relay control, status indication |
8. Thermal Management and Energy Efficiency
Temperature significantly affects the lifetime and safety of lithium iron phosphate (LFP) cells. The three-level BMS integrates thermal management at all levels. At the pack level, the BMU controls a fan to maintain local airflow. At the cluster level, the BCU can request additional cooling from the BAU if internal temperature sensors indicate overheating. At the system level, the BAU modulates the HVAC operation based on the aggregated temperature readings and system power. The energy consumption of the thermal system can be modelled as:
$$ E_{thermal} = \int P_{fan}(t) + P_{hvac}(t) \, dt $$
where \(P_{fan}\) is the total fan power and \(P_{hvac}\) is the HVAC compressor/ heater power. Optimising the trade-off between cooling energy and cell degradation is a key design goal of a battery energy storage system. The three-level BMS can implement a predictive algorithm that pre-cools the battery before a high-power demand event.
9. Data Logging and Diagnostic Features
To comply with grid codes and facilitate maintenance, the BAU records all critical data, including cell voltages, temperatures, currents, SOC, alarms, and state transitions. The storage medium is typically an internal eMMC (8 GB or larger) plus an external SD card slot. The data retention requirement is a minimum of 120 days. I have calculated the required storage capacity \(C_{log}\) as follows:
$$ C_{log} = N_{clusters} \times N_{cells} \times N_{params} \times F_{log} \times T_{retention} $$
For example, in a system with 4 clusters, each containing 200 cells, logging 10 parameters per cell at 1 Hz for 120 days:
$$ C_{log} = 4 \times 200 \times 10 \times 1 \times (120 \times 86400) \approx 8.3 \times 10^{10} \, \text{bytes} \approx 80 \, \text{GB} $$
Thus, a combination of 8 GB internal memory and a 128 GB SD card is sufficient. The data can be exported via USB for offline analysis.
10. Conclusion
In this paper, I have presented a comprehensive analysis of the three-level BMS architecture applied in centralised battery energy storage systems. The hierarchical design, spanning BMU, BCU, and BAU, provides scalable and robust management for commercial and industrial installations. Through detailed tables, mathematical formulations, and operational descriptions, I have demonstrated how this architecture addresses key challenges such as cell imbalance, circulating currents, thermal management, and system-level protection. While string-based topologies are gaining popularity, the three-level BMS remains a cost-effective and proven solution for many battery energy storage system projects. The integration of intelligent communication, data logging, and alarm management ensures safe and efficient operation over the system’s lifetime. Future work may explore the integration of machine learning algorithms for predictive maintenance and SOC/SOH estimation within this framework.