<p>The increasing integration of connected vehicles in intelligent transportation systems has led to the emergence of the internet of vehicles (IoV), which brings both new opportunities and cybersecurity risks. Traditional intrusion detection systems (IDS) that rely on centralized architectures face challenges in IoV environments due to scalability, latency, and privacy concerns. This work presents FEDQES-IDS, a federated and lightweight IDS specifically designed for IoV scenarios. The proposed framework combines quadratic unconstrained binary optimization (QUBO)-based feature selection with an efficient echo state network (ESN) classifier to enable federated and privacy-preserving model training across distributed vehicular nodes. By parallelizing local optimization and model learning at the edge, the framework reduces end-to-end latency and supports high-rate vehicular traffic processing. To ensure secure communication of model updates, the framework integrates ASCON-80pq, a lightweight and quantum-resistant encryption algorithm. FEDQES-IDS includes an end-to-end federated learning pipeline that performs local feature optimization and collaborative model aggregation while keeping communication and computational costs low. Evaluations on the CICIoV2024 and CAN-FD datasets demonstrate an accuracy increase of up to 4.04 percentage points over non-federated baselines. Furthermore, the framework reduces average inference latency by 40% and decreases per-round communication overhead by over 99% compared to recent federated baselines, providing empirical evidence of its suitability for scalable, real-time deployment.</p>