Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Federated Graph Learning for EV Charging Demand Forecasting with Personalization Against Cyberattacks

Yi Li, Renyou Xie, Chaojie Li, Yi Wang, Zhaoyang Dong

TL;DR

The paper tackles privacy and cybersecurity challenges in EV charging demand forecasting by proposing a personalized federated graph learning framework that fuses graph neural networks with attention-based, robust aggregation. The PFGL system integrates a spatial-temporal graph convolutional backbone with a credit-based attention mechanism to learn per-station models that share information while resisting Byzantine attacks. A convergence theorem and robustness analysis accompany the method, and experiments on the Palo Alto dataset demonstrate improved probabilistic forecasts and resilience under attack. The work advances secure, scalable, and individualized forecasting for EV charging infrastructure planning and grid stability.

Abstract

Mitigating cybersecurity risk in electric vehicle (EV) charging demand forecasting plays a crucial role in the safe operation of collective EV chargings, the stability of the power grid, and the cost-effective infrastructure expansion. However, existing methods either suffer from the data privacy issue and the susceptibility to cyberattacks or fail to consider the spatial correlation among different stations. To address these challenges, a federated graph learning approach involving multiple charging stations is proposed to collaboratively train a more generalized deep learning model for demand forecasting while capturing spatial correlations among various stations and enhancing robustness against potential attacks. Firstly, for better model performance, a Graph Neural Network (GNN) model is leveraged to characterize the geographic correlation among different charging stations in a federated manner. Secondly, to ensure robustness and deal with the data heterogeneity in a federated setting, a message passing that utilizes a global attention mechanism to aggregate personalized models for each client is proposed. Thirdly, by concerning cyberattacks, a special credit-based function is designed to mitigate potential threats from malicious clients or unwanted attacks. Extensive experiments on a public EV charging dataset are conducted using various deep learning techniques and federated learning methods to demonstrate the prediction accuracy and robustness of the proposed approach.

Federated Graph Learning for EV Charging Demand Forecasting with Personalization Against Cyberattacks

TL;DR

The paper tackles privacy and cybersecurity challenges in EV charging demand forecasting by proposing a personalized federated graph learning framework that fuses graph neural networks with attention-based, robust aggregation. The PFGL system integrates a spatial-temporal graph convolutional backbone with a credit-based attention mechanism to learn per-station models that share information while resisting Byzantine attacks. A convergence theorem and robustness analysis accompany the method, and experiments on the Palo Alto dataset demonstrate improved probabilistic forecasts and resilience under attack. The work advances secure, scalable, and individualized forecasting for EV charging infrastructure planning and grid stability.

Abstract

Mitigating cybersecurity risk in electric vehicle (EV) charging demand forecasting plays a crucial role in the safe operation of collective EV chargings, the stability of the power grid, and the cost-effective infrastructure expansion. However, existing methods either suffer from the data privacy issue and the susceptibility to cyberattacks or fail to consider the spatial correlation among different stations. To address these challenges, a federated graph learning approach involving multiple charging stations is proposed to collaboratively train a more generalized deep learning model for demand forecasting while capturing spatial correlations among various stations and enhancing robustness against potential attacks. Firstly, for better model performance, a Graph Neural Network (GNN) model is leveraged to characterize the geographic correlation among different charging stations in a federated manner. Secondly, to ensure robustness and deal with the data heterogeneity in a federated setting, a message passing that utilizes a global attention mechanism to aggregate personalized models for each client is proposed. Thirdly, by concerning cyberattacks, a special credit-based function is designed to mitigate potential threats from malicious clients or unwanted attacks. Extensive experiments on a public EV charging dataset are conducted using various deep learning techniques and federated learning methods to demonstrate the prediction accuracy and robustness of the proposed approach.
Paper Structure (21 sections, 1 theorem, 25 equations, 4 figures, 7 tables, 1 algorithm)

This paper contains 21 sections, 1 theorem, 25 equations, 4 figures, 7 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Problem Formulation
  3. EV Charging Demand Forecasting
  4. New FL for Charging Demand Forecasting
  5. Proposed Personalized Federated Graph Learning Method
  6. Spatial-Temporal Graph Convolution Networks
  7. Personalized Federated Learning
  8. Convergence Analysis
  9. Robust Analysis Against Cyberattacks
  10. Case Studies
  11. Baseline Approaches
  12. Experimental Setups
  13. Implementation Details
  14. Evaluation Metrics
  15. Performance without Malicious Attacks
  16. ...and 6 more sections

Key Result

Theorem 1

Suppose that Assumptions asp:L-smooth, asp:bounded_grad, and asp:bounded_diff hold, if $\eta_t \le \min\{\frac{1}{LK},\frac{1}{\beta_i \sqrt{3(K-1)(K+1)}}\}$ and if $\eta_t = \frac{1}{\sqrt{T}}$, then for $T \ge \max\{L^2K^2, 3\beta_i^2 (K-1)(K+1)\}$, Algorithm alg:server ensure : where $f_i^*$ is the optimal value of $f_i()$, $\Phi(\beta_i,K)=27(L^2+\beta_i^2)(K+1)M^2$, and $\Psi(\beta_i,K,\sigm

Figures (4)

  • Figure 1: The proposed personalized federated graph learning framework of EV charging demand forecasting.
  • Figure 2: Prediction result of PFGL model (6 step prediction) for stations (a) WEBS, (b) BRYANT, (c) TED, (d) RINCO, (e) HAMLT, (f) HIGH, (g) CAMB, (h) MPL.
  • Figure 3: Heatmap of global attention
  • Figure 4: Weight $\lambda_{i,j}$ attributed to 'HIGH' station (a) $\sigma=0.6$. (b) $\sigma=0.8$, (c) $\sigma=0.9$, (d) $\sigma=0.95$.

Theorems & Definitions (2)

  • Theorem 1
  • proof