Skip to main content
BatterySOCBenchmark

McMaster University · Electrical & Computer Engineering

The standardized benchmark for battery state-of-charge estimation.

Train on open Tesla Model 3 2170 cell data. Submit your estimator. Get scored on 144 blinded drive cycles from −20 °C to 40 °C plus robustness tests — wrong initial SOC, current-sensor offsets, charging — the same test for every algorithm, from coulomb counting to transformers.

Submit a model View leaderboard

Free for academic and industry use · CC-BY 4.0 dataset · blinded evaluation, results in minutes

Top of the leaderboard

All models
  1. 1
    LSTM-Romulo

    LSTM · McMaster University

    2.40%
  2. 2
    LSTM–Coulomb Counting (Previous Contest) - KOR

    Hybrid · McMaster University

    2.43%
  3. 3
    LSTM Reference Example

    LSTM · McMaster University

    2.73%
  4. 4
    Wisconsin - Old

    Other · McMaster University

    2.88%
  5. 5
    UKF-Hybrid

    Unscented Kalman Filter · McMaster University

    3.20%

Weighted error (% SOC), lower is better.

11

Evaluated models

1

Institutions

0.71%

Best all-cells RMSE

144 + 51

Blinded drive cycles + robustness runs

Supported by

McMaster UniversityNatural Sciences and Engineering Research Council of Canada (NSERC / CRSNG)

New to battery SOC?

The problem in four lines

Read the get-started guide
  • State of chargeHow full the battery is. It can't be measured — only estimated from current, voltage and temperature.
  • Why it's hardFlat voltage curves, 10× resistance at −20 °C, unknown starting charge, drifting sensors.
  • A submissionOne function, MATLAB or Python: it gets one sample per second and returns SOC.
  • The scoreWeighted error in % SOC over hidden cycles at six temperatures. Lower is better; the best are near 3 %.

How it works

Three steps from data to a standardized score

  1. Step 1

    Download the open data

    Characterization tests (HPPC, C/20, C/3, C/2, 1C) and reordered drive cycles for three cells at six temperatures. Everything you need to parameterize a filter or train a network.

    Get the dataset
  2. Step 2

    Build your estimator

    Any method: coulomb counting, Kalman filters, physics-based models, neural networks. Package it as Model.m, Model.p or Model.py, then test it on the site before submitting.

    Submission format
  3. Step 3

    Submit for blinded evaluation

    Your model runs against cycles and a cell you have never seen, plus robustness cases with initial-SOC and current-sensor errors. Results land on the leaderboard with full time-domain plots.

    Submit a model

Open contest

Deadline Oct 9, 2026

2026 Battery SOC Estimation Challenge

Build the most accurate and robust state-of-charge estimator for the Tesla 2170 cell across −20 °C to 40 °C. Cash prizes for the top three weighted-error scores.

CA$5,000 first prize · CA$2,000 second · CA$1,000 third

Contest details & registration

The dataset

  • 4 × Tesla/Panasonic 2170 NCA cells
  • −20, −10, 0, 10, 25, 40 °C
  • 384 drive cycles, 1 Hz
  • Open / blind split by design

Payloads of 80, 448 and 1000 kg with HVAC on/off, modelled on a Tesla Model 3 Standard Range. Licensed CC-BY 4.0 on Borealis.

Explore the dataset

Cite the benchmark

P. J. Kollmeyer, M. Naguib, F. Khanum and A. Emadi, “A Blind Modeling Tool for Standardized Evaluation of Battery State of Charge Estimation Algorithms,” 2022 IEEE Transportation Electrification Conference & Expo (ITEC), pp. 243–248, 2022. doi:10.1109/ITEC53557.2022.9813996

Dataset: P. J. Kollmeyer, F. Khanum, M. Naguib, A. Emadi, “Tesla Model 3 2170 Li-ion Cell Dataset and Battery SOC Estimation Blind Modeling Tool,” Borealis, doi:10.5683/SP3/ZVTR4B.