Facilitating Battery Swapping Services for Freight Trucks with Spatial-Temporal Demand Prediction

Linyu Liu, Zhen Dai, Shiji Song, Xiaocheng Li, Guanting Chen

Abstract: Electrifying heavy-duty trucks offers a substantial opportunity to curtail carbon emissions, advancing toward a carbon-neutral future. However, the inherent challenges of limited battery energy and the sheer weight of heavy-duty trucks lead to reduced mileage and prolonged charging durations. Consequently, battery-swapping services emerge as an attractive solution for these trucks. This paper employs a two-fold approach to investigate the potential and enhance the efficacy of such services. Firstly, spatial-temporal demand prediction models are adopted to predict the traffic patterns for the upcoming hours. Subsequently, the prediction guides an optimization module for efficient battery allocation and deployment. Analyzing the heavy-duty truck data on a highway network spanning over 2,500 miles, our model and analysis underscore the value of prediction/machine learning in facilitating future decision-makings. In particular, we find that the initial phase of implementing battery-swapping services favors mobile battery-swapping stations, but as the system matures, fixed-location stations are preferred.

Introduction

Heavy-duty trucks (HDTs) represent a significant climate challenge, contributing over 40% of road transport CO2 emissions despite making up less than 10% of vehicles. Electrification is difficult because massive batteries reduce cargo capacity and require long charging times. Battery Swapping Stations (BSS) offer a diesel-like refueling experience, but infrastructure deployment faces a "chicken-and-egg" problem: operators need precise demand forecasts to justify capital investment. This paper addresses the operational challenge of managing battery inventory across a vast highway network using a mix of fixed stations and mobile swapping units.

Methodology

The researchers employ a two-stage "predict-then-optimize" pipeline to manage the BSS infrastructure.

The predict-then-optimize pipeline combining ML demand forecasting with an integer programming planning model.
  1. Demand Prediction: Two spatial-temporal neural networks forecast truck traffic volume for a 6-hour horizon. The Temporal Graph Convolutional Network (T-GCN) combines graph convolutions with Gated Recurrent Units to capture network topology and time-series patterns. The Attention Graph Convolutional Network (A3T-GCN) adds an attention mechanism to capture long-term correlations.
  2. Optimization: Predictions are fed into an Integer Program that determines battery inventory levels at fixed stations and dispatches mobile BSS units to service areas. The objective is to minimize "lost demand"—instances where trucks require a swap but no batteries are available.

The study utilizes a high-fidelity dataset from the Chongqing Expressway Group Company (CEGC), covering 2,500 miles of highway monitored by 1,800 camera gantries. Computer vision was used to isolate heavy-duty truck traffic over a three-month period.

Key Results and Insights

The Mobility Shift

A primary finding is that mobile BSS units are superior during the early stages of electrification. When EV adoption is low, traffic patterns are irregular and unpredictable. Mobile units can dynamically "chase" these peaks. As the system matures and traffic stabilizes, fixed-location stations become the more cost-effective backbone.

Lost demand across different mobile ratios and traffic shifts, comparing T-GCN, A3T-GCN, and hindsight models.

Model Performance and Utility

While the attention-based A3T-GCN model provided more accurate long-term forecasts (3–5 hours), the simpler T-GCN model actually yielded better results within the optimization pipeline. This suggests that for infrastructure scheduling, accuracy in the immediate 1–2 hour window is more critical than long-term precision. Overall, the ML-based scheduling achieved results within 109% of an "oracle" policy (perfect hindsight), demonstrating the robustness of the framework.

Conclusion and Future Work

This research proves that a "predict-then-optimize" approach can effectively manage battery-swapping infrastructure for freight logistics. Future work will address the complexities of grid load, electricity costs, and the potential for reinforcement learning to handle more dynamic scheduling variables such as weather and road accidents.

Summary updated

Abstract

Electrifying heavy-duty trucks offers a substantial opportunity to curtail carbon emissions, advancing toward a carbon-neutral future. However, the inherent challenges of limited battery energy and the sheer weight of heavy-duty trucks lead to reduced mileage and prolonged charging durations. Consequently, battery-swapping services emerge as an attractive solution for these trucks. This paper employs a two-fold approach to investigate the potential and enhance the efficacy of such services. Firstly, spatial-temporal demand prediction models are adopted to predict the traffic patterns for the upcoming hours. Subsequently, the prediction guides an optimization module for efficient battery allocation and deployment. Analyzing the heavy-duty truck data on a highway network spanning over 2,500 miles, our model and analysis underscore the value of prediction/machine learning in facilitating future decision-makings. In particular, we find that the initial phase of implementing battery-swapping services favors mobile battery-swapping stations, but as the system matures, fixed-location stations are preferred.
BibTeX
@misc{liu2023facilitating,
  author = {Linyu Liu and Zhen Dai and Shiji Song and Xiaocheng Li and Guanting Chen},
  title = {{Facilitating Battery Swapping Services for Freight Trucks with Spatial-Temporal Demand Prediction}},
  year = {2023},
  eprint = {2310.04440},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2310.04440},
}
Figure