Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Mixer-Informer-Based Two-Stage Transfer Learning for Long-Sequence Load Forecasting in Newly Constructed Electric Vehicle Charging Stations

Zhenhua Zhou, Bozhen Jiang, Qin Wang

TL;DR

This work tackles long-sequence EV charging load forecasting in newly constructed stations where data are scarce. It introduces MIK-TST, a two-stage transfer learning framework that combines Mixer for feature fusion, Informer with ProbSparse attention for long-range temporal modeling, and KAN for learnable activation-based nonlinear mapping. The model pre-trains on a large, multi-station dataset and fine-tunes on limited target data, achieving 4% MAE and 8% MSE improvements over strong baselines on 26 Boulder charging stations, with ablation and hyperparameter analyses supporting the design choices. The approach offers a scalable, accurate solution for smart-grid planning and dynamic EV infrastructure expansion, while outlining avenues to incorporate external factors and uncertainty estimation for even greater reliability.

Abstract

The rapid rise in electric vehicle (EV) adoption demands precise charging station load forecasting, challenged by long-sequence temporal dependencies and limited data in new facilities. This study proposes MIK-TST, a novel two-stage transfer learning framework integrating Mixer, Informer, and Kolmogorov-Arnold Networks (KAN). The Mixer fuses multi-source features, Informer captures long-range dependencies via ProbSparse attention, and KAN enhances nonlinear modeling with learnable activation functions. Pre-trained on extensive data and fine-tuned on limited target data, MIK-TST achieves 4% and 8% reductions in MAE and MSE, respectively, outperforming baselines on a dataset of 26 charging stations in Boulder, USA. This scalable solution enhances smart grid efficiency and supports sustainable EV infrastructure expansion.

Mixer-Informer-Based Two-Stage Transfer Learning for Long-Sequence Load Forecasting in Newly Constructed Electric Vehicle Charging Stations

TL;DR

This work tackles long-sequence EV charging load forecasting in newly constructed stations where data are scarce. It introduces MIK-TST, a two-stage transfer learning framework that combines Mixer for feature fusion, Informer with ProbSparse attention for long-range temporal modeling, and KAN for learnable activation-based nonlinear mapping. The model pre-trains on a large, multi-station dataset and fine-tunes on limited target data, achieving 4% MAE and 8% MSE improvements over strong baselines on 26 Boulder charging stations, with ablation and hyperparameter analyses supporting the design choices. The approach offers a scalable, accurate solution for smart-grid planning and dynamic EV infrastructure expansion, while outlining avenues to incorporate external factors and uncertainty estimation for even greater reliability.

Abstract

The rapid rise in electric vehicle (EV) adoption demands precise charging station load forecasting, challenged by long-sequence temporal dependencies and limited data in new facilities. This study proposes MIK-TST, a novel two-stage transfer learning framework integrating Mixer, Informer, and Kolmogorov-Arnold Networks (KAN). The Mixer fuses multi-source features, Informer captures long-range dependencies via ProbSparse attention, and KAN enhances nonlinear modeling with learnable activation functions. Pre-trained on extensive data and fine-tuned on limited target data, MIK-TST achieves 4% and 8% reductions in MAE and MSE, respectively, outperforming baselines on a dataset of 26 charging stations in Boulder, USA. This scalable solution enhances smart grid efficiency and supports sustainable EV infrastructure expansion.
Paper Structure (25 sections, 15 equations, 7 figures, 2 tables)

This paper contains 25 sections, 15 equations, 7 figures, 2 tables.

Table of Contents

  1. Introduction
  2. PRELIMINARIES
  3. Problem Formulation
  4. Related Studies on EV Charging Load Forecasting:
  5. METHODOLOGY
  6. Mixer Module
  7. Network Structure
  8. Advantages of the Mixing Mechanism
  9. Informer Module
  10. Network Structure
  11. Advantages of the Informer
  12. KAN Module
  13. Network Structure
  14. Advantages of the KAN
  15. Two-Stage Transfer Learning Strategy
  16. ...and 10 more sections

Figures (7)

  • Figure 1: Overall architecture of the MIK-TST
  • Figure 2: Overall architecture of the Mixer
  • Figure 3: Overall architecture of the Informer
  • Figure 4: Overall architecture of the KAN
  • Figure 5: Parameter sensitivity analysis of MIK-TST with respect to the hidden state dimension $d$.
  • ...and 2 more figures