Truace
Subscribe

Decision intelligence for electric bus charging scheduling through LLM-driven reward shaping and reinforcement learning

The rapid expansion of electric bus fleets has made charging scheduling a critical decision-intelligence problem in urban fleet-energy logistics. Reinforcement learning (RL) is promising for this problem, but reward function design still relies on manual trial-and-error and transfers poorly across scenarios. This paper develops an artificial intelligence and operations research (AI-OR) decision intelligence framework that integrates large language model (LLM)-driven reward shaping with RL-based sequential decisi…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-2177· cite & verify
Decision intelligence for electric bus charging scheduling through LLM-driven reward shaping and reinforcement learning
"VDL Evolans Electric Buses at Hermes" by Misanthropic One is licensed under CC BY 2.0. To view a copy of this license, visit https://creativecommons.org/licenses/by/2.0/.

In brief

Researchers built a decision-intelligence framework in which a large language model writes and revises reward functions as executable code, and a reinforcement learning agent uses them to schedule electric bus charging. It was tested on one synthetic scenario and two real bus stations in Guangzhou, using three mainstream LLMs under different initialization conditions.

The work matters because growing electric bus fleets make charging scheduling hard, and manual reward design transfers poorly between settings. Automating that step could ease deployment and help shift charging to valley periods. The evidence comes from simulated experiments at a few stations, so real-world operation is not yet shown.

Main points

  1. The framework has four modules: reward initializer, performance analysis, reward exploration, and rollback.
  2. An event-triggered two-stage charging reallocation mechanism enforces operational constraints through action masking.
  3. Costs fell by up to 32.6 % in nominal within-day charging cost versus traditional methods.
  4. After an approximate adjustment for energy left on board, savings were up to 22.5 %.
  5. Ablation and sensitivity analyses supported each module and compatibility with multiple RL algorithms.

The gain

An LLM-driven reward iteration framework for reinforcement learning scheduled electric bus charging at lower cost than traditional methods across synthetic and real-world Guangzhou bus station scenarios.

The rundown

The paper frames electric bus charging scheduling as a decision-intelligence problem in urban fleet-energy logistics. Reinforcement learning is promising, but reward design currently depends on manual trial-and-error.

The approach represents reward functions as executable code that the LLM iteratively optimizes. Battery safety bounds are managed through readiness conditions, reward shaping, and the acceptance criterion.

Experiments compared three mainstream LLMs on a synthetic case and two Guangzhou bus stations. Analysis of how the rewards evolved identified design principles for charging efficiency, valley-period energy shifting, and fleet-energy logistics resilience.

Sources

  1. Peer-reviewedTransportation Research Part C: Emerging Technologies2026-10-21

The debate