Skip to content
Sai Eswar Gujjala
  • Professional work
  • Deployed internally

EV Battery Degradation Prediction

Forecasts EV battery State of Health (SoH) up to 12 months ahead from high-frequency telemetry, with 200+ engineered ageing features and explainable gradient-boosting models.

My role
End-to-end owner.
Key result
Saved ₹1.4 crore from projected replacement budgets.
Technologies
  • Python
  • pandas
  • XGBoost
  • LightGBM
  • SHAP
Skills shown here

What I owned

  1. Problem framing
  2. Data preparation
  3. Modeling
  4. Evaluation
  5. Deployment
  6. Monitoring

Built at Zetwerk for internal use. Described in simplified form; no customer or supplier names, data, code or internal links are shown.

In [1]:Problem and constraints

Battery health declines with charging and discharging patterns, temperature, current load and cycles. Lower SoH reduces range and performance, and makes maintenance and remaining-life planning harder.

In [2]:Data and preparation

Battery telemetry at 10 Hz: current, voltage, temperature, State of Charge and cycle information.

In [3]:Baseline and evaluation design

  • Cross-validated regression, measured with MSE and R² on the project's evaluation dataset.
  • SHAP explains which features drive each prediction.
  • Cross-validated MSE of 0.0635, with SoH in percent (0–100).

In [4]:Architecture and modeling decisions

Gradient boosting (XGBoost and LightGBM, in cross-validated ensembles) on engineered features. It suits tabular, non-linear data and can be explained with SHAP.

In [5]:Implementation and integration

  • 200+ features: electrochemical ageing indicators, temperature stress, discharge-curve characteristics and C-rate stress.
  • Automated retraining and monitoring as new telemetry arrives.

In [6]:Deployment and operation

Status: Deployed internally

Internal, used by the EV division.

Out[6]:Results and evidence

  • Saved ₹1.4 crore from projected replacement budgets.
  • The EV division can estimate warranty risk 12 months earlier, from SoH forecasts up to 12 months ahead.
  • SHAP showed which telemetry factors drive degradation.
State of Health forecast error (MSE)
0.0635
Cross-validated regression on the project's evaluation dataset, with SoH in percent (0–100).
Offline evaluation

In [7]:Limitations, failures, and lessons

  • Forecasts depend on telemetry quality and on usage staying similar to the past, and the saving is an estimate based on projected budgets.