Kenneth ChengMathematical modeling

Editorial project overview

Interpretable EV Dispatch and Replenishment

Can an auditable model balance fleet energy, profit, fairness, and disruption response?

Reproducible studyUpdated 30 Apr 2026
Technical figure from Interpretable EV Dispatch and Replenishment
A reviewed research figure synchronized from ScienceProject. The original aspect ratio and labels are preserved. Source and generation notes.

Overview

This project links a physics-based energy estimate to an interpretable residual model, then uses the estimates inside a multi-objective dispatch search. EWMA and CUSUM statistics trigger receding-horizon replanning when residual behavior changes.

Evidence and interpretation

The reported run has a high coefficient of determination but a 42.512% MAPE. Both numbers matter: the model captures broad variation while retaining substantial relative error. The optimization results are therefore a computational study, not a production dispatch guarantee.

Reproduction

Run the light pipeline with python SJMMA2026/ProblemE/run.py --mode all --scope light. The repository contains bilingual reports and stable generated figures.

Key findings

  • Route and environment features dominate the configured residual correction in the reported run.
  • The schedule experiment exposes a non-degenerate profit-energy-fairness frontier.

Limitations

  • The reported energy MAPE is 42.512%, despite a high R².
  • Scheduling uses a simplified assignment structure rather than a full exact MILP.
  • Disruptions are generated with a stylized simulation.

Technical record

Detailed source, calculations, generated figures, and reproduction instructions remain in ScienceProject. Open the technical project.

Version history

2026-04-30 — Curated overview reviewed against repository evidence.