KushT/batteryswap2026

BatterySwapAI 2026 Submission — Survival RUL Model + Urgency-Scored Planner

Solution for the BatterySwapAI 2026 challenge: predicts each battery's Remaining Useful Life (RUL) as a calibrated distribution using a right-censored survival model, then schedules a technician's swap route around those predictions.

This repo started from the official example repository and replaces its dummy median-only RUL model and one-swap-per-day planner with the implementation described below.

Approach

Part 1 — RUL prediction (batteryswap_example/features.py, rul_model.py)

• Voltage/temperature features: rolling mean/min/variance, slope and curvature (rate and acceleration of decline), cumulative time spent below key voltage thresholds, an Arrhenius-adjusted thermal-stress integral, and a lightweight knee-point (changepoint) detector.
• Building-pooled feature aggregates as an ID-free proxy for shared environmental risk between co-located batteries — chosen so it still works on buildings not seen during training, since public/private splits use different buildings than the training split.
• SurvivalRULModel: a right-censored Cox proportional-hazards model (via lifelines), with a Weibull AFT fallback, trained on RUL durations synthesized from multiple scenario cuts of the raw timeseries. Returns calibrated p10/p50/p90 RUL quantiles instead of a single point estimate.

Part 2 — Scheduling (batteryswap_example/planner.py)

• UrgencyPlanner: a greedy earliest-deadline-first router for the single traveling technician the evaluator simulates.
• Swap deadlines are derived from the model's p10 RUL quantile — chosen because the evaluator penalizes a late swap 20x more heavily than an early one by default, so the cost-optimal target sits deep in the early tail of the failure-time distribution.
• Visits to the same building are batched together to amortize the fixed building-change/travel cost, and are deferred until shortly before their deadline rather than scheduled as early as possible, to avoid needless early-replacement penalties.
• Batteries not due within the current planning window are given a placeholder day past the window (the evaluator ignores plan rows past the horizon), so the route only spends travel time on batteries that actually need attention this cycle.

Known limitations

• train.py trains on a small, fixed number of scenario cuts by default (see DEFAULT_TRAIN_SCENARIOS and the comment above it). Stacking many more scenarios was tried and made the fitted hazard model noticeably worse in local testing, likely because scenario cuts of the same ~460 physical devices are correlated pseudo-replicates rather than independent training examples. A more careful fix (sample weighting, cross-validated regularization, or a proper train/validation split across scenarios) is a natural next step.
• Evaluated locally only against the train split; not yet validated against the public/private splits or run inside the official Docker image under the full 32GB/CPU-only/30-minute constraints.

Setup

Prerequisites

You need to have the following installed

• git
• Python 3.10+
• Docker

Setting up virtual environment

On Linux / Mac OS / Windows Subsystem for Linux (WSL)

python -m venv venv
source venv/bin/activate

Install dependencies

pip install -r requirements.txt -r requirements.dev.txt

Developing

Develop in virtual environment

Run the training

python batteryswap_example/train.py

Making submissions

Submitting trained models

Trained models or other output used by submission processing (script.py), must be committed in the git repository. An example is batteryswap_example/planners/best.pickle.

Test submissions in Docker - recommended

Using Docker allows to have exactly the same software versions as the submissions system.

This helps to ensure there are no errors when running in the submission environment.

NOTE: this requires around 20 GB+ of disk space.

Build Docker image

docker build -t batteryswapai-2026-example .

Make submissions and run evaluation

docker run --name batteryswapai -v ./dataset:/tmp/data batteryswapai-2026-example bash -c "/app/env/bin/python3 script.py && /app/env/bin/python3 -m batteryswap_public.metric"

Copy submission.csv out of container

docker cp batteryswapai:/app/submission.csv ./submission.csv

Create new submission

NOTE: remember to commit and push your changes to the HuggingFace model repository.

Use New submission in the competition application to submit your current code for evaluation.