Federated Learning-based Distributed Control for Heterogeneous Platoons of Connected Autonomous Electric Vehicles

To promote the EU's sustainable development of environment and economy, EU planned to reduce emissions from conventional vehicles in European Green Deal.

Heterogeneous platoons of connected autonomous electric vehicles (HPCAEVs) are thought to be the most efficient, viable, and promising solution to reduce both energy consumption and pollutant emissions produced by conventional vehicles and improve road efficiency.

However, since HPCAEVs are still in the evolving stage, the secure and energy-saving operation of HPCAEVs in mixed traffic flow is suffering significant challenges.

The FEASIBLE project is proposed in the context of addressing these challenges by providing some scientific foundations that enable the secure and energy-saving operation of HPCAEVs in mixed traffic flow, with particular focus on networked modeling, stability analysis, and control design.

First, network-level federated learning (FL) models will be set up by incorporating intrinsic characteristics of HPCAEVs, macroscopic traffic flow, surrounding vehicle flow, and their interactions.

Second, novel stability analysis criteria for HPCAEVs will be derived with expected dynamic performance by designing a FL-based Lyapunov function with sufficiently small and bounded modeling error.

Third, FL-based distributed robust model predictive control methods for HPCAEVs with intervehicle distance constraints will be developed under surrounding road conditions.

Finally, advanced simulations and tests of HPCAEVs will be conducted to demonstrate the effectiveness and practicality of the proposed modeling and control methods, and the control performance will be compared with cutting-edge methods.