Optimal Scheduling of Electric Heavy-Truck Battery Swapping Stations via Distributionally Robust Deep Reinforcement Learning
Year: 2026
Type: article
Abstract: Electric heavy-truck battery swapping stations (BSSs) face coupled challenges in practical operation, including uncertain swapping demand, time-of-use (TOU) electricity-price fluctuations, and difficult on-site battery dispatch. This paper proposes a robust optimal scheduling method based on the Distributionally Robust Soft Actor–Critic (DR-SAC) algorithm. First, a battery-cluster model based on state-of-charge (SOC) classification compresses the decision space into level-wise charging-count scheduling, reducing dimensionality. Second, the scheduling problem is formulated as a Robust Markov Decision Process (RMDP), optimizing the worst-case entropy-regularized cumulative reward within a Kullback–Leibler (KL) divergence-constrained ambiguity set. To cope with the unknown transition kernel in offline learning, a Variational Auto-Encoder (VAE) estimates the transition distribution from historical data, while a strong-duality transformation and functional approximation efficiently solve the robust Bellman operator. Results show that under joint upward perturbations of demand and price, the relative variation in the daily profit of DR-SAC is 3.7%, and the peak-shaving rate reaches 78.9% at a KL radius δ of 0.067 calibrated from historical forecasting errors. Across varying demand intensities, DR-SAC ranks first in daily profit with the smallest run-to-run variation among the compared algorithms on the studied station. DR-SAC provides a practical scheduling solution for electric heavy-truck BSSs under multiple uncertainties.
Source: Applied Sciences
DOI: 10.3390/app162211346
Institutions: Ministry of Transport, Research Institute of Highway
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,896
OpenAlex ID: W7233985844
