Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Energy Estimation of Last Mile Electric Vehicle Routes

André Snoeck, Aniruddha Bhargava, Daniel Merchan, Josiah Davis, Julian Pachon

TL;DR

The paper reframes last-mile electrification by treating energy consumption as the key planning metric and develops a spectrum of deep learning models to predict energy use along routes. It compares a segment-level FFNN, a route-level GRU-based RNN, and a compute-optimized decoder-only Transformer named Route Energy Transformer (RET), with RET achieving the largest accuracy gains under a Chinchilla-informed scaling regime. Empirical results show substantial improvements in mean absolute percentage error (MAPE) over distance- and physics-based baselines, particularly on hot and cold routes, and provide a detailed analysis of inference speed across CPU and GPU platforms. The study highlights the practical viability of energy-aware route planning and outlines pathways to combine physics, uncertainty quantification, and real-time adjustments to further enhance last-mile EV operations.

Abstract

Last-mile carriers increasingly incorporate electric vehicles (EVs) into their delivery fleet to achieve sustainability goals. This goal presents many challenges across multiple planning spaces including but not limited to how to plan EV routes. In this paper, we address the problem of predicting energy consumption of EVs for Last-Mile delivery routes using deep learning. We demonstrate the need to move away from thinking about range and we propose using energy as the basic unit of analysis. We share a range of deep learning solutions, beginning with a Feed Forward Neural Network (NN) and Recurrent Neural Network (RNN) and demonstrate significant accuracy improvements relative to pure physics-based and distance-based approaches. Finally, we present Route Energy Transformer (RET) a decoder-only Transformer model sized according to Chinchilla scaling laws. RET yields a +217 Basis Points (bps) improvement in Mean Absolute Percentage Error (MAPE) relative to the Feed Forward NN and a +105 bps improvement relative to the RNN.

Energy Estimation of Last Mile Electric Vehicle Routes

TL;DR

The paper reframes last-mile electrification by treating energy consumption as the key planning metric and develops a spectrum of deep learning models to predict energy use along routes. It compares a segment-level FFNN, a route-level GRU-based RNN, and a compute-optimized decoder-only Transformer named Route Energy Transformer (RET), with RET achieving the largest accuracy gains under a Chinchilla-informed scaling regime. Empirical results show substantial improvements in mean absolute percentage error (MAPE) over distance- and physics-based baselines, particularly on hot and cold routes, and provide a detailed analysis of inference speed across CPU and GPU platforms. The study highlights the practical viability of energy-aware route planning and outlines pathways to combine physics, uncertainty quantification, and real-time adjustments to further enhance last-mile EV operations.

Abstract

Last-mile carriers increasingly incorporate electric vehicles (EVs) into their delivery fleet to achieve sustainability goals. This goal presents many challenges across multiple planning spaces including but not limited to how to plan EV routes. In this paper, we address the problem of predicting energy consumption of EVs for Last-Mile delivery routes using deep learning. We demonstrate the need to move away from thinking about range and we propose using energy as the basic unit of analysis. We share a range of deep learning solutions, beginning with a Feed Forward Neural Network (NN) and Recurrent Neural Network (RNN) and demonstrate significant accuracy improvements relative to pure physics-based and distance-based approaches. Finally, we present Route Energy Transformer (RET) a decoder-only Transformer model sized according to Chinchilla scaling laws. RET yields a +217 Basis Points (bps) improvement in Mean Absolute Percentage Error (MAPE) relative to the Feed Forward NN and a +105 bps improvement relative to the RNN.
Paper Structure (15 sections, 1 equation, 2 figures, 4 tables)

This paper contains 15 sections, 1 equation, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Problem Statement: Electrifying a Last Mile fleet is all about energy
  3. Related Work
  4. Modeling Energy Consumption with Deep Learning
  5. Features.
  6. Predicting individual segments with a Fully Connected Model.
  7. Predicting entire routes with a Recurrent Neural Network.
  8. Training a Compute Optimal Transformer
  9. Modeling Results
  10. Accuracy
  11. Inference Speed
  12. Final Remarks
  13. Beyond route planning.
  14. Future research.
  15. Glossary

Figures (2)

  • Figure 1: Energy consumption (remaining Battery Capacity %) vs. total route length (km). Each dot is one route. Color represents ambient temperature (C$^{\circ}$).
  • Figure 2: Data size vs. Parameter count for Chinchilla Optimal Models. We predict that the Chinchilla optimal model size is orders of magnitude larger than the Feed Forward NN and the RNN.