Dataset for: A Multi-Agent AI Framework for Explainable Battery System Maintenance

DataForPub Dataset Description

1. Dataset Overview

This dataset contains field operation data collected from an in-service containerized Battery Energy Storage System (BESS). The released data cover eight months of real-world operation and include continuous charge and discharge processes from one battery cluster.

The studied cluster consists of 9 battery racks. Each rack contains 396 lithium-ion cells and 216 temperature measurement points. For each operation, the dataset records rack-level operating variables and cell-level measurements, including cell voltage, temperature, SOC, current, total rack voltage, and power.

The dataset is designed to support research on battery inconsistency evaluation, operation and maintenance (O&M), degradation analysis, abnormal behavior identification, and data-driven decision support for large-scale energy storage systems.

2. System Description

The studied BESS cluster follows a hierarchical structure:

BESS Cluster├── Rack 1│   ├── 396 cell voltage measurements│   ├── 216 temperature measurements│   ├── SOC│   ├── Current│   ├── Total voltage│   └── Power├── Rack 2│   ├── 396 cell voltage measurements│   ├── 216 temperature measurements│   ├── SOC│   ├── Current│   ├── Total voltage│   └── Power├── ...└── Rack 9   ├── 396 cell voltage measurements   ├── 216 temperature measurements   ├── SOC   ├── Current   ├── Total voltage   └── Power

Each monthly file stores all valid charge and discharge operations recorded during that month. An operation represents one continuous charging or discharging process of the studied battery cluster.

3. Dataset File Format

The dataset is provided in MATLAB .mat format. Each .mat file corresponds to one month of operation data.

Example filename:

OperationCondition_202410_All.mat

The file contains one top-level MATLAB structure:

OperationCondition_202410_All

This top-level structure contains multiple operation branches:

OperationCondition_202410_All├── Operation1├── Operation2├── Operation3├── ...└── OperationN

where N is the number of valid operations recorded in the corresponding month.

4. Monthly File Organization

The complete public dataset contains eight monthly .mat files:

DataForPub/├── OperationCondition_202410_All.mat├── OperationCondition_202411_All.mat├── OperationCondition_202412_All.mat├── OperationCondition_202501_All.mat├── OperationCondition_202502_All.mat├── OperationCondition_202503_All.mat├── OperationCondition_202504_All.mat└── OperationCondition_202505_All.mat

Each file follows the same internal structure:

OperationCondition_YYYYMM_All.mat└── OperationCondition_YYYYMM_All   ├── Operation1   ├── Operation2   ├── Operation3   ├── ...   └── OperationN

5. MATLAB Data Structure

Each OperationK branch corresponds to one continuous operation and contains operation-level metadata, rack-level structures, and rack-level time-series variables.

OperationCondition_YYYYMM_All├── Operation1│   ├── Type│   ├── StartTime│   ├── EndTime│   ├── CurrentMean│   ├── Rack1│   ├── Rack2│   ├── Rack3│   ├── Rack4│   ├── Rack5│   ├── Rack6│   ├── Rack7│   ├── Rack8│   ├── Rack9│   ├── Rack1SOCTime│   ├── Rack1SOCValue│   ├── Rack2SOCTime│   ├── Rack2SOCValue│   ├── ...│   ├── Rack9SOCTime│   ├── Rack9SOCValue│   ├── Rack1CurrentTime│   ├── Rack1CurrentValue│   ├── Rack2CurrentTime│   ├── Rack2CurrentValue│   ├── ...│   ├── Rack9CurrentTime│   ├── Rack9CurrentValue│   ├── Rack1VoltageTime│   ├── Rack1VoltageValue│   ├── Rack2VoltageTime│   ├── Rack2VoltageValue│   ├── ...│   ├── Rack9VoltageTime│   ├── Rack9VoltageValue│   ├── Rack1PowerTime│   ├── Rack1PowerValue│   ├── Rack2PowerTime│   ├── Rack2PowerValue│   ├── ...│   └── Rack9PowerValue├── Operation2│   └── Same structure as Operation1├── ...└── OperationN   └── Same structure as Operation1

6. Operation-Level Fields

Each operation contains the following basic fields:

Field name
Data type
Description

Type
char
Operation type. The value is usually Charge or Discharge.

StartTime
datetime
Start time of the operation.

EndTime
datetime
End time of the operation.

CurrentMean
double
Mean current during the operation.

Rack1–Rack9
struct
Cell-level voltage and temperature data for each rack.

7. Rack-Level Time-Series Variables

For each rack, the operation branch contains paired Time and Value arrays for SOC, current, total voltage, and power.

Here, X represents the rack index, ranging from 1 to 9.

Field name
Data type
Description

RackXSOCTime
datetime array
Time stamps of SOC measurements for Rack X.

RackXSOCValue
double array
SOC values of Rack X.

RackXCurrentTime
datetime array
Time stamps of current measurements for Rack X.

RackXCurrentValue
double array
Current values of Rack X.

RackXVoltageTime
datetime array
Time stamps of total rack voltage measurements for Rack X.

RackXVoltageValue
double array
Total rack voltage values of Rack X.

RackXPowerTime
datetime array
Time stamps of rack power measurements for Rack X.

RackXPowerValue
double array
Rack power values of Rack X.

8. Rack-Level Structure

Each operation contains nine rack-level structures:

OperationK├── Rack1├── Rack2├── ...└── Rack9

Each RackX structure contains cell-level voltage measurements and temperature measurements for the corresponding rack.

RackX├── Cell voltage data│   ├── Voltage measurement of Cell 1│   ├── Voltage measurement of Cell 2│   ├── ...│   └── Voltage measurement of Cell 396└── Temperature data   ├── Temperature measurement point 1   ├── Temperature measurement point 2   ├── ...   └── Temperature measurement point 216

The voltage data describe the time-series voltage behavior of 396 cells in each rack. The temperature data describe the time-series thermal behavior of 216 temperature measurement points in each rack.

9. Measurement Summary

For each rack, the dataset provides the following measurements:

Measurement
Description

Cell voltage
Time-series voltage measurements of 396 cells in each rack.

Temperature
Time-series temperature measurements of 216 temperature points in each rack.

SOC
State of charge of the rack.

Current
Rack current during the operation.

Total voltage
Total rack voltage during the operation.

Power
Rack power during the operation.

10. Example: One Operation Branch

An example branch of Operation1 is shown below:

Operation1├── Type: 'Charge'├── StartTime: 2024-10-05 10:30:00├── EndTime: 2024-10-05 12:46:47├── CurrentMean: -121.4283├── Rack1│   ├── Cell voltage measurements of 396 cells│   └── Temperature measurements of 216 temperature points├── Rack2│   ├── Cell voltage measurements of 396 cells│   └── Temperature measurements of 216 temperature points├── ...├── Rack9│   ├── Cell voltage measurements of 396 cells│   └── Temperature measurements of 216 temperature points├── Rack1SOCTime├── Rack1SOCValue├── Rack2SOCTime├── Rack2SOCValue├── ...├── Rack9SOCTime├── Rack9SOCValue├── Rack1CurrentTime├── Rack1CurrentValue├── ...├── Rack9CurrentTime├── Rack9CurrentValue├── Rack1VoltageTime├── Rack1VoltageValue├── ...├── Rack9VoltageTime├── Rack9VoltageValue├── Rack1PowerTime├── Rack1PowerValue├── ...└── Rack9PowerValue

11. Data Content Summary

The dataset contains:

Eight months of field operation data from one in-service BESS cluster.

More than 400 charge and discharge operations.

Nine racks in each operation.

396 cell voltage measurements for each rack.

216 temperature measurement points for each rack.

Rack-level SOC, current, total voltage, and power measurements.

Time stamps corresponding to each measured variable.

12. Recommended Citation

If you use this dataset in academic work, please cite the Zenodo dataset record and the following related publications.

Dataset record:

DataForPub Dataset. Zenodo. DOI: 10.5281/zenodo.20094787.

Dataset-related article:

Qu J, Wang Y, Fu Y, et al. A multi-agent AI framework for explainable battery system maintenance[J]. Cell Reports Physical Science, 2026, 7(6).

13. Related Work

This dataset supports and extends the authors’ previous research on diagnosing inconsistencies in battery energy storage systems from electrical, thermal, and aging perspectives [1]. The released data are further associated with the explainable operation and maintenance framework presented in [2], which investigates the transition from inconsistency identification to decision support for BESS operation and maintenance.

14. Data Availability and Access Notes

The dataset supporting this study has been deposited in Zenodo at DOI: 10.5281/zenodo.20094787.

The record metadata are publicly available. The data files are under restricted or embargoed access until the formal publication of the associated article.

Access can be provided to editors and reviewers upon reasonable request.

The dataset record will be updated with the final article DOI after publication, if applicable.

15. Contact

For questions about the dataset, please contact the dataset authors or submit an issue through the public repository.

References

[1] Qu J, Shen J, Li W, et al. Diagnosing inconsistencies in battery energy storage systems: A framework integrating electrical, thermal, and aging perspectives[J]. Applied Energy, 2026, 405: 127203.

[2] Qu J, Wang Y, Fu Y, et al. A multi-agent AI framework for explainable battery system maintenance[J]. Cell Reports Physical Science, 2026, 7(6).